wandb 0.18.2__py3-none-musllinux_1_2_x86_64.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package_readme.md +89 -0
- wandb/__init__.py +245 -0
- wandb/__init__.pyi +1139 -0
- wandb/__main__.py +3 -0
- wandb/_globals.py +19 -0
- wandb/agents/__init__.py +0 -0
- wandb/agents/pyagent.py +363 -0
- wandb/analytics/__init__.py +3 -0
- wandb/analytics/sentry.py +266 -0
- wandb/apis/__init__.py +48 -0
- wandb/apis/attrs.py +40 -0
- wandb/apis/importers/__init__.py +1 -0
- wandb/apis/importers/internals/internal.py +385 -0
- wandb/apis/importers/internals/protocols.py +99 -0
- wandb/apis/importers/internals/util.py +78 -0
- wandb/apis/importers/mlflow.py +254 -0
- wandb/apis/importers/validation.py +108 -0
- wandb/apis/importers/wandb.py +1603 -0
- wandb/apis/internal.py +232 -0
- wandb/apis/normalize.py +89 -0
- wandb/apis/paginator.py +81 -0
- wandb/apis/public/__init__.py +34 -0
- wandb/apis/public/api.py +1305 -0
- wandb/apis/public/artifacts.py +1090 -0
- wandb/apis/public/const.py +4 -0
- wandb/apis/public/files.py +195 -0
- wandb/apis/public/history.py +149 -0
- wandb/apis/public/jobs.py +659 -0
- wandb/apis/public/projects.py +154 -0
- wandb/apis/public/query_generator.py +166 -0
- wandb/apis/public/reports.py +469 -0
- wandb/apis/public/runs.py +914 -0
- wandb/apis/public/sweeps.py +240 -0
- wandb/apis/public/teams.py +198 -0
- wandb/apis/public/users.py +136 -0
- wandb/apis/reports/__init__.py +1 -0
- wandb/apis/reports/v1/__init__.py +8 -0
- wandb/apis/reports/v2/__init__.py +8 -0
- wandb/apis/workspaces/__init__.py +8 -0
- wandb/beta/workflows.py +288 -0
- wandb/bin/nvidia_gpu_stats +0 -0
- wandb/bin/wandb-core +0 -0
- wandb/cli/__init__.py +0 -0
- wandb/cli/cli.py +3004 -0
- wandb/data_types.py +63 -0
- wandb/docker/__init__.py +342 -0
- wandb/docker/auth.py +436 -0
- wandb/docker/wandb-entrypoint.sh +33 -0
- wandb/docker/www_authenticate.py +94 -0
- wandb/env.py +514 -0
- wandb/errors/__init__.py +17 -0
- wandb/errors/errors.py +37 -0
- wandb/errors/term.py +103 -0
- wandb/errors/util.py +57 -0
- wandb/errors/warnings.py +2 -0
- wandb/filesync/__init__.py +0 -0
- wandb/filesync/dir_watcher.py +403 -0
- wandb/filesync/stats.py +100 -0
- wandb/filesync/step_checksum.py +142 -0
- wandb/filesync/step_prepare.py +179 -0
- wandb/filesync/step_upload.py +290 -0
- wandb/filesync/upload_job.py +142 -0
- wandb/integration/__init__.py +0 -0
- wandb/integration/catboost/__init__.py +5 -0
- wandb/integration/catboost/catboost.py +178 -0
- wandb/integration/cohere/__init__.py +3 -0
- wandb/integration/cohere/cohere.py +21 -0
- wandb/integration/cohere/resolver.py +347 -0
- wandb/integration/diffusers/__init__.py +3 -0
- wandb/integration/diffusers/autologger.py +76 -0
- wandb/integration/diffusers/pipeline_resolver.py +50 -0
- wandb/integration/diffusers/resolvers/__init__.py +9 -0
- wandb/integration/diffusers/resolvers/multimodal.py +882 -0
- wandb/integration/diffusers/resolvers/utils.py +102 -0
- wandb/integration/fastai/__init__.py +249 -0
- wandb/integration/gym/__init__.py +105 -0
- wandb/integration/huggingface/__init__.py +3 -0
- wandb/integration/huggingface/huggingface.py +18 -0
- wandb/integration/huggingface/resolver.py +213 -0
- wandb/integration/keras/__init__.py +11 -0
- wandb/integration/keras/callbacks/__init__.py +5 -0
- wandb/integration/keras/callbacks/metrics_logger.py +136 -0
- wandb/integration/keras/callbacks/model_checkpoint.py +195 -0
- wandb/integration/keras/callbacks/tables_builder.py +226 -0
- wandb/integration/keras/keras.py +1091 -0
- wandb/integration/kfp/__init__.py +6 -0
- wandb/integration/kfp/helpers.py +28 -0
- wandb/integration/kfp/kfp_patch.py +324 -0
- wandb/integration/kfp/wandb_logging.py +182 -0
- wandb/integration/langchain/__init__.py +3 -0
- wandb/integration/langchain/wandb_tracer.py +48 -0
- wandb/integration/lightgbm/__init__.py +239 -0
- wandb/integration/lightning/__init__.py +0 -0
- wandb/integration/lightning/fabric/__init__.py +3 -0
- wandb/integration/lightning/fabric/logger.py +762 -0
- wandb/integration/magic.py +556 -0
- wandb/integration/metaflow/__init__.py +3 -0
- wandb/integration/metaflow/metaflow.py +383 -0
- wandb/integration/openai/__init__.py +3 -0
- wandb/integration/openai/fine_tuning.py +480 -0
- wandb/integration/openai/openai.py +22 -0
- wandb/integration/openai/resolver.py +240 -0
- wandb/integration/prodigy/__init__.py +3 -0
- wandb/integration/prodigy/prodigy.py +299 -0
- wandb/integration/sacred/__init__.py +117 -0
- wandb/integration/sagemaker/__init__.py +12 -0
- wandb/integration/sagemaker/auth.py +28 -0
- wandb/integration/sagemaker/config.py +49 -0
- wandb/integration/sagemaker/files.py +3 -0
- wandb/integration/sagemaker/resources.py +34 -0
- wandb/integration/sb3/__init__.py +3 -0
- wandb/integration/sb3/sb3.py +153 -0
- wandb/integration/sklearn/__init__.py +37 -0
- wandb/integration/sklearn/calculate/__init__.py +32 -0
- wandb/integration/sklearn/calculate/calibration_curves.py +125 -0
- wandb/integration/sklearn/calculate/class_proportions.py +68 -0
- wandb/integration/sklearn/calculate/confusion_matrix.py +93 -0
- wandb/integration/sklearn/calculate/decision_boundaries.py +40 -0
- wandb/integration/sklearn/calculate/elbow_curve.py +55 -0
- wandb/integration/sklearn/calculate/feature_importances.py +67 -0
- wandb/integration/sklearn/calculate/learning_curve.py +64 -0
- wandb/integration/sklearn/calculate/outlier_candidates.py +69 -0
- wandb/integration/sklearn/calculate/residuals.py +86 -0
- wandb/integration/sklearn/calculate/silhouette.py +118 -0
- wandb/integration/sklearn/calculate/summary_metrics.py +62 -0
- wandb/integration/sklearn/plot/__init__.py +35 -0
- wandb/integration/sklearn/plot/classifier.py +329 -0
- wandb/integration/sklearn/plot/clusterer.py +146 -0
- wandb/integration/sklearn/plot/regressor.py +121 -0
- wandb/integration/sklearn/plot/shared.py +91 -0
- wandb/integration/sklearn/utils.py +183 -0
- wandb/integration/tensorboard/__init__.py +10 -0
- wandb/integration/tensorboard/log.py +355 -0
- wandb/integration/tensorboard/monkeypatch.py +185 -0
- wandb/integration/tensorflow/__init__.py +5 -0
- wandb/integration/tensorflow/estimator_hook.py +54 -0
- wandb/integration/torch/__init__.py +0 -0
- wandb/integration/torch/wandb_torch.py +554 -0
- wandb/integration/ultralytics/__init__.py +11 -0
- wandb/integration/ultralytics/bbox_utils.py +208 -0
- wandb/integration/ultralytics/callback.py +524 -0
- wandb/integration/ultralytics/classification_utils.py +83 -0
- wandb/integration/ultralytics/mask_utils.py +202 -0
- wandb/integration/ultralytics/pose_utils.py +103 -0
- wandb/integration/xgboost/__init__.py +11 -0
- wandb/integration/xgboost/xgboost.py +189 -0
- wandb/integration/yolov8/__init__.py +0 -0
- wandb/integration/yolov8/yolov8.py +284 -0
- wandb/jupyter.py +515 -0
- wandb/magic.py +3 -0
- wandb/mpmain/__init__.py +0 -0
- wandb/mpmain/__main__.py +1 -0
- wandb/old/__init__.py +0 -0
- wandb/old/core.py +53 -0
- wandb/old/settings.py +173 -0
- wandb/old/summary.py +440 -0
- wandb/plot/__init__.py +19 -0
- wandb/plot/bar.py +45 -0
- wandb/plot/confusion_matrix.py +100 -0
- wandb/plot/histogram.py +39 -0
- wandb/plot/line.py +43 -0
- wandb/plot/line_series.py +88 -0
- wandb/plot/pr_curve.py +136 -0
- wandb/plot/roc_curve.py +118 -0
- wandb/plot/scatter.py +32 -0
- wandb/plot/utils.py +183 -0
- wandb/plot/viz.py +123 -0
- wandb/proto/__init__.py +0 -0
- wandb/proto/v3/__init__.py +0 -0
- wandb/proto/v3/wandb_base_pb2.py +55 -0
- wandb/proto/v3/wandb_internal_pb2.py +1608 -0
- wandb/proto/v3/wandb_server_pb2.py +208 -0
- wandb/proto/v3/wandb_settings_pb2.py +112 -0
- wandb/proto/v3/wandb_telemetry_pb2.py +106 -0
- wandb/proto/v4/__init__.py +0 -0
- wandb/proto/v4/wandb_base_pb2.py +30 -0
- wandb/proto/v4/wandb_internal_pb2.py +360 -0
- wandb/proto/v4/wandb_server_pb2.py +63 -0
- wandb/proto/v4/wandb_settings_pb2.py +45 -0
- wandb/proto/v4/wandb_telemetry_pb2.py +41 -0
- wandb/proto/v5/wandb_base_pb2.py +31 -0
- wandb/proto/v5/wandb_internal_pb2.py +361 -0
- wandb/proto/v5/wandb_server_pb2.py +64 -0
- wandb/proto/v5/wandb_settings_pb2.py +46 -0
- wandb/proto/v5/wandb_telemetry_pb2.py +42 -0
- wandb/proto/wandb_base_pb2.py +10 -0
- wandb/proto/wandb_deprecated.py +53 -0
- wandb/proto/wandb_generate_deprecated.py +34 -0
- wandb/proto/wandb_generate_proto.py +49 -0
- wandb/proto/wandb_internal_pb2.py +16 -0
- wandb/proto/wandb_server_pb2.py +10 -0
- wandb/proto/wandb_settings_pb2.py +10 -0
- wandb/proto/wandb_telemetry_pb2.py +10 -0
- wandb/py.typed +0 -0
- wandb/sdk/__init__.py +37 -0
- wandb/sdk/artifacts/__init__.py +0 -0
- wandb/sdk/artifacts/_validators.py +90 -0
- wandb/sdk/artifacts/artifact.py +2389 -0
- wandb/sdk/artifacts/artifact_download_logger.py +43 -0
- wandb/sdk/artifacts/artifact_file_cache.py +253 -0
- wandb/sdk/artifacts/artifact_instance_cache.py +17 -0
- wandb/sdk/artifacts/artifact_manifest.py +74 -0
- wandb/sdk/artifacts/artifact_manifest_entry.py +249 -0
- wandb/sdk/artifacts/artifact_manifests/__init__.py +0 -0
- wandb/sdk/artifacts/artifact_manifests/artifact_manifest_v1.py +92 -0
- wandb/sdk/artifacts/artifact_saver.py +269 -0
- wandb/sdk/artifacts/artifact_state.py +11 -0
- wandb/sdk/artifacts/artifact_ttl.py +7 -0
- wandb/sdk/artifacts/exceptions.py +57 -0
- wandb/sdk/artifacts/staging.py +25 -0
- wandb/sdk/artifacts/storage_handler.py +62 -0
- wandb/sdk/artifacts/storage_handlers/__init__.py +0 -0
- wandb/sdk/artifacts/storage_handlers/azure_handler.py +208 -0
- wandb/sdk/artifacts/storage_handlers/gcs_handler.py +228 -0
- wandb/sdk/artifacts/storage_handlers/http_handler.py +114 -0
- wandb/sdk/artifacts/storage_handlers/local_file_handler.py +141 -0
- wandb/sdk/artifacts/storage_handlers/multi_handler.py +56 -0
- wandb/sdk/artifacts/storage_handlers/s3_handler.py +300 -0
- wandb/sdk/artifacts/storage_handlers/tracking_handler.py +72 -0
- wandb/sdk/artifacts/storage_handlers/wb_artifact_handler.py +135 -0
- wandb/sdk/artifacts/storage_handlers/wb_local_artifact_handler.py +74 -0
- wandb/sdk/artifacts/storage_layout.py +6 -0
- wandb/sdk/artifacts/storage_policies/__init__.py +4 -0
- wandb/sdk/artifacts/storage_policies/register.py +1 -0
- wandb/sdk/artifacts/storage_policies/wandb_storage_policy.py +378 -0
- wandb/sdk/artifacts/storage_policy.py +72 -0
- wandb/sdk/backend/__init__.py +0 -0
- wandb/sdk/backend/backend.py +222 -0
- wandb/sdk/data_types/__init__.py +0 -0
- wandb/sdk/data_types/_dtypes.py +914 -0
- wandb/sdk/data_types/_private.py +10 -0
- wandb/sdk/data_types/audio.py +165 -0
- wandb/sdk/data_types/base_types/__init__.py +0 -0
- wandb/sdk/data_types/base_types/json_metadata.py +55 -0
- wandb/sdk/data_types/base_types/media.py +315 -0
- wandb/sdk/data_types/base_types/wb_value.py +272 -0
- wandb/sdk/data_types/bokeh.py +70 -0
- wandb/sdk/data_types/graph.py +405 -0
- wandb/sdk/data_types/helper_types/__init__.py +0 -0
- wandb/sdk/data_types/helper_types/bounding_boxes_2d.py +295 -0
- wandb/sdk/data_types/helper_types/classes.py +159 -0
- wandb/sdk/data_types/helper_types/image_mask.py +235 -0
- wandb/sdk/data_types/histogram.py +96 -0
- wandb/sdk/data_types/html.py +115 -0
- wandb/sdk/data_types/image.py +845 -0
- wandb/sdk/data_types/molecule.py +241 -0
- wandb/sdk/data_types/object_3d.py +474 -0
- wandb/sdk/data_types/plotly.py +82 -0
- wandb/sdk/data_types/saved_model.py +446 -0
- wandb/sdk/data_types/table.py +1204 -0
- wandb/sdk/data_types/trace_tree.py +438 -0
- wandb/sdk/data_types/utils.py +229 -0
- wandb/sdk/data_types/video.py +247 -0
- wandb/sdk/integration_utils/__init__.py +0 -0
- wandb/sdk/integration_utils/auto_logging.py +239 -0
- wandb/sdk/integration_utils/data_logging.py +475 -0
- wandb/sdk/interface/__init__.py +0 -0
- wandb/sdk/interface/constants.py +4 -0
- wandb/sdk/interface/interface.py +972 -0
- wandb/sdk/interface/interface_queue.py +59 -0
- wandb/sdk/interface/interface_relay.py +53 -0
- wandb/sdk/interface/interface_shared.py +537 -0
- wandb/sdk/interface/interface_sock.py +61 -0
- wandb/sdk/interface/message_future.py +27 -0
- wandb/sdk/interface/message_future_poll.py +50 -0
- wandb/sdk/interface/router.py +118 -0
- wandb/sdk/interface/router_queue.py +44 -0
- wandb/sdk/interface/router_relay.py +39 -0
- wandb/sdk/interface/router_sock.py +36 -0
- wandb/sdk/interface/summary_record.py +67 -0
- wandb/sdk/internal/__init__.py +0 -0
- wandb/sdk/internal/context.py +89 -0
- wandb/sdk/internal/datastore.py +297 -0
- wandb/sdk/internal/file_pusher.py +181 -0
- wandb/sdk/internal/file_stream.py +695 -0
- wandb/sdk/internal/flow_control.py +263 -0
- wandb/sdk/internal/handler.py +901 -0
- wandb/sdk/internal/internal.py +417 -0
- wandb/sdk/internal/internal_api.py +4358 -0
- wandb/sdk/internal/internal_util.py +100 -0
- wandb/sdk/internal/job_builder.py +629 -0
- wandb/sdk/internal/profiler.py +78 -0
- wandb/sdk/internal/progress.py +83 -0
- wandb/sdk/internal/run.py +25 -0
- wandb/sdk/internal/sample.py +70 -0
- wandb/sdk/internal/sender.py +1686 -0
- wandb/sdk/internal/sender_config.py +197 -0
- wandb/sdk/internal/settings_static.py +90 -0
- wandb/sdk/internal/system/__init__.py +0 -0
- wandb/sdk/internal/system/assets/__init__.py +27 -0
- wandb/sdk/internal/system/assets/aggregators.py +37 -0
- wandb/sdk/internal/system/assets/asset_registry.py +20 -0
- wandb/sdk/internal/system/assets/cpu.py +163 -0
- wandb/sdk/internal/system/assets/disk.py +210 -0
- wandb/sdk/internal/system/assets/gpu.py +416 -0
- wandb/sdk/internal/system/assets/gpu_amd.py +239 -0
- wandb/sdk/internal/system/assets/gpu_apple.py +177 -0
- wandb/sdk/internal/system/assets/interfaces.py +207 -0
- wandb/sdk/internal/system/assets/ipu.py +177 -0
- wandb/sdk/internal/system/assets/memory.py +166 -0
- wandb/sdk/internal/system/assets/network.py +125 -0
- wandb/sdk/internal/system/assets/open_metrics.py +299 -0
- wandb/sdk/internal/system/assets/tpu.py +154 -0
- wandb/sdk/internal/system/assets/trainium.py +399 -0
- wandb/sdk/internal/system/env_probe_helpers.py +13 -0
- wandb/sdk/internal/system/system_info.py +249 -0
- wandb/sdk/internal/system/system_monitor.py +229 -0
- wandb/sdk/internal/tb_watcher.py +518 -0
- wandb/sdk/internal/thread_local_settings.py +18 -0
- wandb/sdk/internal/writer.py +206 -0
- wandb/sdk/launch/__init__.py +14 -0
- wandb/sdk/launch/_launch.py +330 -0
- wandb/sdk/launch/_launch_add.py +255 -0
- wandb/sdk/launch/_project_spec.py +566 -0
- wandb/sdk/launch/agent/__init__.py +5 -0
- wandb/sdk/launch/agent/agent.py +924 -0
- wandb/sdk/launch/agent/config.py +296 -0
- wandb/sdk/launch/agent/job_status_tracker.py +53 -0
- wandb/sdk/launch/agent/run_queue_item_file_saver.py +45 -0
- wandb/sdk/launch/builder/__init__.py +0 -0
- wandb/sdk/launch/builder/abstract.py +156 -0
- wandb/sdk/launch/builder/build.py +297 -0
- wandb/sdk/launch/builder/context_manager.py +235 -0
- wandb/sdk/launch/builder/docker_builder.py +177 -0
- wandb/sdk/launch/builder/kaniko_builder.py +595 -0
- wandb/sdk/launch/builder/noop.py +58 -0
- wandb/sdk/launch/builder/templates/_wandb_bootstrap.py +188 -0
- wandb/sdk/launch/builder/templates/dockerfile.py +92 -0
- wandb/sdk/launch/create_job.py +528 -0
- wandb/sdk/launch/environment/abstract.py +29 -0
- wandb/sdk/launch/environment/aws_environment.py +322 -0
- wandb/sdk/launch/environment/azure_environment.py +105 -0
- wandb/sdk/launch/environment/gcp_environment.py +335 -0
- wandb/sdk/launch/environment/local_environment.py +66 -0
- wandb/sdk/launch/errors.py +19 -0
- wandb/sdk/launch/git_reference.py +109 -0
- wandb/sdk/launch/inputs/files.py +148 -0
- wandb/sdk/launch/inputs/internal.py +315 -0
- wandb/sdk/launch/inputs/manage.py +113 -0
- wandb/sdk/launch/inputs/schema.py +39 -0
- wandb/sdk/launch/loader.py +249 -0
- wandb/sdk/launch/registry/abstract.py +48 -0
- wandb/sdk/launch/registry/anon.py +29 -0
- wandb/sdk/launch/registry/azure_container_registry.py +124 -0
- wandb/sdk/launch/registry/elastic_container_registry.py +192 -0
- wandb/sdk/launch/registry/google_artifact_registry.py +219 -0
- wandb/sdk/launch/registry/local_registry.py +67 -0
- wandb/sdk/launch/runner/__init__.py +0 -0
- wandb/sdk/launch/runner/abstract.py +195 -0
- wandb/sdk/launch/runner/kubernetes_monitor.py +474 -0
- wandb/sdk/launch/runner/kubernetes_runner.py +963 -0
- wandb/sdk/launch/runner/local_container.py +301 -0
- wandb/sdk/launch/runner/local_process.py +78 -0
- wandb/sdk/launch/runner/sagemaker_runner.py +426 -0
- wandb/sdk/launch/runner/vertex_runner.py +230 -0
- wandb/sdk/launch/sweeps/__init__.py +39 -0
- wandb/sdk/launch/sweeps/scheduler.py +742 -0
- wandb/sdk/launch/sweeps/scheduler_sweep.py +91 -0
- wandb/sdk/launch/sweeps/utils.py +316 -0
- wandb/sdk/launch/utils.py +746 -0
- wandb/sdk/launch/wandb_reference.py +138 -0
- wandb/sdk/lib/__init__.py +5 -0
- wandb/sdk/lib/_settings_toposort_generate.py +159 -0
- wandb/sdk/lib/_settings_toposort_generated.py +250 -0
- wandb/sdk/lib/_wburls_generate.py +25 -0
- wandb/sdk/lib/_wburls_generated.py +22 -0
- wandb/sdk/lib/apikey.py +273 -0
- wandb/sdk/lib/capped_dict.py +26 -0
- wandb/sdk/lib/config_util.py +101 -0
- wandb/sdk/lib/credentials.py +141 -0
- wandb/sdk/lib/deprecate.py +42 -0
- wandb/sdk/lib/disabled.py +29 -0
- wandb/sdk/lib/exit_hooks.py +54 -0
- wandb/sdk/lib/file_stream_utils.py +118 -0
- wandb/sdk/lib/filenames.py +64 -0
- wandb/sdk/lib/filesystem.py +372 -0
- wandb/sdk/lib/fsm.py +174 -0
- wandb/sdk/lib/gitlib.py +239 -0
- wandb/sdk/lib/gql_request.py +65 -0
- wandb/sdk/lib/handler_util.py +21 -0
- wandb/sdk/lib/hashutil.py +84 -0
- wandb/sdk/lib/import_hooks.py +275 -0
- wandb/sdk/lib/ipython.py +146 -0
- wandb/sdk/lib/json_util.py +80 -0
- wandb/sdk/lib/lazyloader.py +63 -0
- wandb/sdk/lib/mailbox.py +460 -0
- wandb/sdk/lib/module.py +69 -0
- wandb/sdk/lib/paths.py +106 -0
- wandb/sdk/lib/preinit.py +42 -0
- wandb/sdk/lib/printer.py +313 -0
- wandb/sdk/lib/proto_util.py +90 -0
- wandb/sdk/lib/redirect.py +845 -0
- wandb/sdk/lib/reporting.py +99 -0
- wandb/sdk/lib/retry.py +289 -0
- wandb/sdk/lib/run_moment.py +78 -0
- wandb/sdk/lib/runid.py +12 -0
- wandb/sdk/lib/server.py +52 -0
- wandb/sdk/lib/service_connection.py +216 -0
- wandb/sdk/lib/service_token.py +94 -0
- wandb/sdk/lib/sock_client.py +295 -0
- wandb/sdk/lib/sparkline.py +45 -0
- wandb/sdk/lib/telemetry.py +100 -0
- wandb/sdk/lib/timed_input.py +133 -0
- wandb/sdk/lib/timer.py +19 -0
- wandb/sdk/lib/tracelog.py +255 -0
- wandb/sdk/lib/wburls.py +46 -0
- wandb/sdk/service/__init__.py +0 -0
- wandb/sdk/service/_startup_debug.py +22 -0
- wandb/sdk/service/port_file.py +53 -0
- wandb/sdk/service/server.py +116 -0
- wandb/sdk/service/server_sock.py +276 -0
- wandb/sdk/service/service.py +242 -0
- wandb/sdk/service/streams.py +417 -0
- wandb/sdk/verify/__init__.py +0 -0
- wandb/sdk/verify/verify.py +501 -0
- wandb/sdk/wandb_alerts.py +12 -0
- wandb/sdk/wandb_config.py +322 -0
- wandb/sdk/wandb_helper.py +54 -0
- wandb/sdk/wandb_init.py +1266 -0
- wandb/sdk/wandb_login.py +349 -0
- wandb/sdk/wandb_metric.py +110 -0
- wandb/sdk/wandb_require.py +97 -0
- wandb/sdk/wandb_require_helpers.py +44 -0
- wandb/sdk/wandb_run.py +4236 -0
- wandb/sdk/wandb_settings.py +2001 -0
- wandb/sdk/wandb_setup.py +409 -0
- wandb/sdk/wandb_summary.py +150 -0
- wandb/sdk/wandb_sweep.py +119 -0
- wandb/sdk/wandb_sync.py +81 -0
- wandb/sdk/wandb_watch.py +144 -0
- wandb/sklearn.py +35 -0
- wandb/sync/__init__.py +3 -0
- wandb/sync/sync.py +443 -0
- wandb/trigger.py +29 -0
- wandb/util.py +1956 -0
- wandb/vendor/__init__.py +0 -0
- wandb/vendor/gql-0.2.0/setup.py +40 -0
- wandb/vendor/gql-0.2.0/tests/__init__.py +0 -0
- wandb/vendor/gql-0.2.0/tests/starwars/__init__.py +0 -0
- wandb/vendor/gql-0.2.0/tests/starwars/fixtures.py +96 -0
- wandb/vendor/gql-0.2.0/tests/starwars/schema.py +146 -0
- wandb/vendor/gql-0.2.0/tests/starwars/test_dsl.py +293 -0
- wandb/vendor/gql-0.2.0/tests/starwars/test_query.py +355 -0
- wandb/vendor/gql-0.2.0/tests/starwars/test_validation.py +171 -0
- wandb/vendor/gql-0.2.0/tests/test_client.py +31 -0
- wandb/vendor/gql-0.2.0/tests/test_transport.py +89 -0
- wandb/vendor/gql-0.2.0/wandb_gql/__init__.py +4 -0
- wandb/vendor/gql-0.2.0/wandb_gql/client.py +75 -0
- wandb/vendor/gql-0.2.0/wandb_gql/dsl.py +152 -0
- wandb/vendor/gql-0.2.0/wandb_gql/gql.py +10 -0
- wandb/vendor/gql-0.2.0/wandb_gql/transport/__init__.py +0 -0
- wandb/vendor/gql-0.2.0/wandb_gql/transport/http.py +6 -0
- wandb/vendor/gql-0.2.0/wandb_gql/transport/local_schema.py +15 -0
- wandb/vendor/gql-0.2.0/wandb_gql/transport/requests.py +46 -0
- wandb/vendor/gql-0.2.0/wandb_gql/utils.py +21 -0
- wandb/vendor/graphql-core-1.1/setup.py +86 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/__init__.py +287 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/error/__init__.py +6 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/error/base.py +42 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/error/format_error.py +11 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/error/located_error.py +29 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/error/syntax_error.py +36 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/__init__.py +26 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/base.py +311 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executor.py +398 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/__init__.py +0 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/asyncio.py +53 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/gevent.py +22 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/process.py +32 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/sync.py +7 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/thread.py +35 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/utils.py +6 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/__init__.py +0 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/executor.py +66 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/fragment.py +252 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/resolver.py +151 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/utils.py +7 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/middleware.py +57 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/execution/values.py +145 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/graphql.py +60 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/__init__.py +0 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/ast.py +1349 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/base.py +19 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/lexer.py +435 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/location.py +30 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/parser.py +779 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/printer.py +193 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/source.py +18 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/visitor.py +222 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/language/visitor_meta.py +82 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/__init__.py +0 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/cached_property.py +17 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/contain_subset.py +28 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/default_ordered_dict.py +40 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/ordereddict.py +8 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/pair_set.py +43 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/version.py +78 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/__init__.py +67 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/definition.py +619 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/directives.py +132 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/introspection.py +440 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/scalars.py +131 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/schema.py +100 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/type/typemap.py +145 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/__init__.py +0 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/assert_valid_name.py +9 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_from_value.py +65 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_to_code.py +49 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_to_dict.py +24 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/base.py +75 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/build_ast_schema.py +291 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/build_client_schema.py +250 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/concat_ast.py +9 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/extend_schema.py +357 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/get_field_def.py +27 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/get_operation_ast.py +21 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/introspection_query.py +90 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/is_valid_literal_value.py +67 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/is_valid_value.py +66 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/quoted_or_list.py +21 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/schema_printer.py +168 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/suggestion_list.py +56 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_comparators.py +69 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_from_ast.py +21 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_info.py +149 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/utils/value_from_ast.py +69 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/__init__.py +4 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/__init__.py +79 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/arguments_of_correct_type.py +24 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/base.py +8 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/default_values_of_correct_type.py +44 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/fields_on_correct_type.py +113 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/fragments_on_composite_types.py +33 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_argument_names.py +70 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_directives.py +97 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_fragment_names.py +19 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_type_names.py +43 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/lone_anonymous_operation.py +23 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_fragment_cycles.py +59 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_undefined_variables.py +36 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_unused_fragments.py +38 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_unused_variables.py +37 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/overlapping_fields_can_be_merged.py +529 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/possible_fragment_spreads.py +44 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/provided_non_null_arguments.py +46 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/scalar_leafs.py +33 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_argument_names.py +32 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_fragment_names.py +28 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_input_field_names.py +33 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_operation_names.py +31 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_variable_names.py +27 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/variables_are_input_types.py +21 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/variables_in_allowed_position.py +53 -0
- wandb/vendor/graphql-core-1.1/wandb_graphql/validation/validation.py +158 -0
- wandb/vendor/promise-2.3.0/conftest.py +30 -0
- wandb/vendor/promise-2.3.0/setup.py +64 -0
- wandb/vendor/promise-2.3.0/tests/__init__.py +0 -0
- wandb/vendor/promise-2.3.0/tests/conftest.py +8 -0
- wandb/vendor/promise-2.3.0/tests/test_awaitable.py +32 -0
- wandb/vendor/promise-2.3.0/tests/test_awaitable_35.py +47 -0
- wandb/vendor/promise-2.3.0/tests/test_benchmark.py +116 -0
- wandb/vendor/promise-2.3.0/tests/test_complex_threads.py +23 -0
- wandb/vendor/promise-2.3.0/tests/test_dataloader.py +452 -0
- wandb/vendor/promise-2.3.0/tests/test_dataloader_awaitable_35.py +99 -0
- wandb/vendor/promise-2.3.0/tests/test_dataloader_extra.py +65 -0
- wandb/vendor/promise-2.3.0/tests/test_extra.py +670 -0
- wandb/vendor/promise-2.3.0/tests/test_issues.py +132 -0
- wandb/vendor/promise-2.3.0/tests/test_promise_list.py +70 -0
- wandb/vendor/promise-2.3.0/tests/test_spec.py +584 -0
- wandb/vendor/promise-2.3.0/tests/test_thread_safety.py +115 -0
- wandb/vendor/promise-2.3.0/tests/utils.py +3 -0
- wandb/vendor/promise-2.3.0/wandb_promise/__init__.py +38 -0
- wandb/vendor/promise-2.3.0/wandb_promise/async_.py +135 -0
- wandb/vendor/promise-2.3.0/wandb_promise/compat.py +32 -0
- wandb/vendor/promise-2.3.0/wandb_promise/dataloader.py +326 -0
- wandb/vendor/promise-2.3.0/wandb_promise/iterate_promise.py +12 -0
- wandb/vendor/promise-2.3.0/wandb_promise/promise.py +848 -0
- wandb/vendor/promise-2.3.0/wandb_promise/promise_list.py +151 -0
- wandb/vendor/promise-2.3.0/wandb_promise/pyutils/__init__.py +0 -0
- wandb/vendor/promise-2.3.0/wandb_promise/pyutils/version.py +83 -0
- wandb/vendor/promise-2.3.0/wandb_promise/schedulers/__init__.py +0 -0
- wandb/vendor/promise-2.3.0/wandb_promise/schedulers/asyncio.py +22 -0
- wandb/vendor/promise-2.3.0/wandb_promise/schedulers/gevent.py +21 -0
- wandb/vendor/promise-2.3.0/wandb_promise/schedulers/immediate.py +27 -0
- wandb/vendor/promise-2.3.0/wandb_promise/schedulers/thread.py +18 -0
- wandb/vendor/promise-2.3.0/wandb_promise/utils.py +56 -0
- wandb/vendor/pygments/__init__.py +90 -0
- wandb/vendor/pygments/cmdline.py +568 -0
- wandb/vendor/pygments/console.py +74 -0
- wandb/vendor/pygments/filter.py +74 -0
- wandb/vendor/pygments/filters/__init__.py +350 -0
- wandb/vendor/pygments/formatter.py +95 -0
- wandb/vendor/pygments/formatters/__init__.py +153 -0
- wandb/vendor/pygments/formatters/_mapping.py +85 -0
- wandb/vendor/pygments/formatters/bbcode.py +109 -0
- wandb/vendor/pygments/formatters/html.py +851 -0
- wandb/vendor/pygments/formatters/img.py +600 -0
- wandb/vendor/pygments/formatters/irc.py +182 -0
- wandb/vendor/pygments/formatters/latex.py +482 -0
- wandb/vendor/pygments/formatters/other.py +160 -0
- wandb/vendor/pygments/formatters/rtf.py +147 -0
- wandb/vendor/pygments/formatters/svg.py +153 -0
- wandb/vendor/pygments/formatters/terminal.py +136 -0
- wandb/vendor/pygments/formatters/terminal256.py +309 -0
- wandb/vendor/pygments/lexer.py +871 -0
- wandb/vendor/pygments/lexers/__init__.py +329 -0
- wandb/vendor/pygments/lexers/_asy_builtins.py +1645 -0
- wandb/vendor/pygments/lexers/_cl_builtins.py +232 -0
- wandb/vendor/pygments/lexers/_cocoa_builtins.py +72 -0
- wandb/vendor/pygments/lexers/_csound_builtins.py +1346 -0
- wandb/vendor/pygments/lexers/_lasso_builtins.py +5327 -0
- wandb/vendor/pygments/lexers/_lua_builtins.py +295 -0
- wandb/vendor/pygments/lexers/_mapping.py +500 -0
- wandb/vendor/pygments/lexers/_mql_builtins.py +1172 -0
- wandb/vendor/pygments/lexers/_openedge_builtins.py +2547 -0
- wandb/vendor/pygments/lexers/_php_builtins.py +4756 -0
- wandb/vendor/pygments/lexers/_postgres_builtins.py +621 -0
- wandb/vendor/pygments/lexers/_scilab_builtins.py +3094 -0
- wandb/vendor/pygments/lexers/_sourcemod_builtins.py +1163 -0
- wandb/vendor/pygments/lexers/_stan_builtins.py +532 -0
- wandb/vendor/pygments/lexers/_stata_builtins.py +419 -0
- wandb/vendor/pygments/lexers/_tsql_builtins.py +1004 -0
- wandb/vendor/pygments/lexers/_vim_builtins.py +1939 -0
- wandb/vendor/pygments/lexers/actionscript.py +240 -0
- wandb/vendor/pygments/lexers/agile.py +24 -0
- wandb/vendor/pygments/lexers/algebra.py +221 -0
- wandb/vendor/pygments/lexers/ambient.py +76 -0
- wandb/vendor/pygments/lexers/ampl.py +87 -0
- wandb/vendor/pygments/lexers/apl.py +101 -0
- wandb/vendor/pygments/lexers/archetype.py +318 -0
- wandb/vendor/pygments/lexers/asm.py +641 -0
- wandb/vendor/pygments/lexers/automation.py +374 -0
- wandb/vendor/pygments/lexers/basic.py +500 -0
- wandb/vendor/pygments/lexers/bibtex.py +160 -0
- wandb/vendor/pygments/lexers/business.py +612 -0
- wandb/vendor/pygments/lexers/c_cpp.py +252 -0
- wandb/vendor/pygments/lexers/c_like.py +541 -0
- wandb/vendor/pygments/lexers/capnproto.py +78 -0
- wandb/vendor/pygments/lexers/chapel.py +102 -0
- wandb/vendor/pygments/lexers/clean.py +288 -0
- wandb/vendor/pygments/lexers/compiled.py +34 -0
- wandb/vendor/pygments/lexers/configs.py +833 -0
- wandb/vendor/pygments/lexers/console.py +114 -0
- wandb/vendor/pygments/lexers/crystal.py +393 -0
- wandb/vendor/pygments/lexers/csound.py +366 -0
- wandb/vendor/pygments/lexers/css.py +689 -0
- wandb/vendor/pygments/lexers/d.py +251 -0
- wandb/vendor/pygments/lexers/dalvik.py +125 -0
- wandb/vendor/pygments/lexers/data.py +555 -0
- wandb/vendor/pygments/lexers/diff.py +165 -0
- wandb/vendor/pygments/lexers/dotnet.py +691 -0
- wandb/vendor/pygments/lexers/dsls.py +878 -0
- wandb/vendor/pygments/lexers/dylan.py +289 -0
- wandb/vendor/pygments/lexers/ecl.py +125 -0
- wandb/vendor/pygments/lexers/eiffel.py +65 -0
- wandb/vendor/pygments/lexers/elm.py +121 -0
- wandb/vendor/pygments/lexers/erlang.py +533 -0
- wandb/vendor/pygments/lexers/esoteric.py +277 -0
- wandb/vendor/pygments/lexers/ezhil.py +69 -0
- wandb/vendor/pygments/lexers/factor.py +344 -0
- wandb/vendor/pygments/lexers/fantom.py +250 -0
- wandb/vendor/pygments/lexers/felix.py +273 -0
- wandb/vendor/pygments/lexers/forth.py +177 -0
- wandb/vendor/pygments/lexers/fortran.py +205 -0
- wandb/vendor/pygments/lexers/foxpro.py +428 -0
- wandb/vendor/pygments/lexers/functional.py +21 -0
- wandb/vendor/pygments/lexers/go.py +101 -0
- wandb/vendor/pygments/lexers/grammar_notation.py +213 -0
- wandb/vendor/pygments/lexers/graph.py +80 -0
- wandb/vendor/pygments/lexers/graphics.py +553 -0
- wandb/vendor/pygments/lexers/haskell.py +843 -0
- wandb/vendor/pygments/lexers/haxe.py +936 -0
- wandb/vendor/pygments/lexers/hdl.py +382 -0
- wandb/vendor/pygments/lexers/hexdump.py +103 -0
- wandb/vendor/pygments/lexers/html.py +602 -0
- wandb/vendor/pygments/lexers/idl.py +270 -0
- wandb/vendor/pygments/lexers/igor.py +288 -0
- wandb/vendor/pygments/lexers/inferno.py +96 -0
- wandb/vendor/pygments/lexers/installers.py +322 -0
- wandb/vendor/pygments/lexers/int_fiction.py +1343 -0
- wandb/vendor/pygments/lexers/iolang.py +63 -0
- wandb/vendor/pygments/lexers/j.py +146 -0
- wandb/vendor/pygments/lexers/javascript.py +1525 -0
- wandb/vendor/pygments/lexers/julia.py +333 -0
- wandb/vendor/pygments/lexers/jvm.py +1573 -0
- wandb/vendor/pygments/lexers/lisp.py +2621 -0
- wandb/vendor/pygments/lexers/make.py +202 -0
- wandb/vendor/pygments/lexers/markup.py +595 -0
- wandb/vendor/pygments/lexers/math.py +21 -0
- wandb/vendor/pygments/lexers/matlab.py +663 -0
- wandb/vendor/pygments/lexers/ml.py +769 -0
- wandb/vendor/pygments/lexers/modeling.py +358 -0
- wandb/vendor/pygments/lexers/modula2.py +1561 -0
- wandb/vendor/pygments/lexers/monte.py +204 -0
- wandb/vendor/pygments/lexers/ncl.py +894 -0
- wandb/vendor/pygments/lexers/nimrod.py +159 -0
- wandb/vendor/pygments/lexers/nit.py +64 -0
- wandb/vendor/pygments/lexers/nix.py +136 -0
- wandb/vendor/pygments/lexers/oberon.py +105 -0
- wandb/vendor/pygments/lexers/objective.py +504 -0
- wandb/vendor/pygments/lexers/ooc.py +85 -0
- wandb/vendor/pygments/lexers/other.py +41 -0
- wandb/vendor/pygments/lexers/parasail.py +79 -0
- wandb/vendor/pygments/lexers/parsers.py +835 -0
- wandb/vendor/pygments/lexers/pascal.py +644 -0
- wandb/vendor/pygments/lexers/pawn.py +199 -0
- wandb/vendor/pygments/lexers/perl.py +620 -0
- wandb/vendor/pygments/lexers/php.py +267 -0
- wandb/vendor/pygments/lexers/praat.py +294 -0
- wandb/vendor/pygments/lexers/prolog.py +306 -0
- wandb/vendor/pygments/lexers/python.py +939 -0
- wandb/vendor/pygments/lexers/qvt.py +152 -0
- wandb/vendor/pygments/lexers/r.py +453 -0
- wandb/vendor/pygments/lexers/rdf.py +270 -0
- wandb/vendor/pygments/lexers/rebol.py +431 -0
- wandb/vendor/pygments/lexers/resource.py +85 -0
- wandb/vendor/pygments/lexers/rnc.py +67 -0
- wandb/vendor/pygments/lexers/roboconf.py +82 -0
- wandb/vendor/pygments/lexers/robotframework.py +560 -0
- wandb/vendor/pygments/lexers/ruby.py +519 -0
- wandb/vendor/pygments/lexers/rust.py +220 -0
- wandb/vendor/pygments/lexers/sas.py +228 -0
- wandb/vendor/pygments/lexers/scripting.py +1222 -0
- wandb/vendor/pygments/lexers/shell.py +794 -0
- wandb/vendor/pygments/lexers/smalltalk.py +195 -0
- wandb/vendor/pygments/lexers/smv.py +79 -0
- wandb/vendor/pygments/lexers/snobol.py +83 -0
- wandb/vendor/pygments/lexers/special.py +103 -0
- wandb/vendor/pygments/lexers/sql.py +681 -0
- wandb/vendor/pygments/lexers/stata.py +108 -0
- wandb/vendor/pygments/lexers/supercollider.py +90 -0
- wandb/vendor/pygments/lexers/tcl.py +145 -0
- wandb/vendor/pygments/lexers/templates.py +2283 -0
- wandb/vendor/pygments/lexers/testing.py +207 -0
- wandb/vendor/pygments/lexers/text.py +25 -0
- wandb/vendor/pygments/lexers/textedit.py +169 -0
- wandb/vendor/pygments/lexers/textfmts.py +297 -0
- wandb/vendor/pygments/lexers/theorem.py +458 -0
- wandb/vendor/pygments/lexers/trafficscript.py +54 -0
- wandb/vendor/pygments/lexers/typoscript.py +226 -0
- wandb/vendor/pygments/lexers/urbi.py +133 -0
- wandb/vendor/pygments/lexers/varnish.py +190 -0
- wandb/vendor/pygments/lexers/verification.py +111 -0
- wandb/vendor/pygments/lexers/web.py +24 -0
- wandb/vendor/pygments/lexers/webmisc.py +988 -0
- wandb/vendor/pygments/lexers/whiley.py +116 -0
- wandb/vendor/pygments/lexers/x10.py +69 -0
- wandb/vendor/pygments/modeline.py +44 -0
- wandb/vendor/pygments/plugin.py +68 -0
- wandb/vendor/pygments/regexopt.py +92 -0
- wandb/vendor/pygments/scanner.py +105 -0
- wandb/vendor/pygments/sphinxext.py +158 -0
- wandb/vendor/pygments/style.py +155 -0
- wandb/vendor/pygments/styles/__init__.py +80 -0
- wandb/vendor/pygments/styles/abap.py +29 -0
- wandb/vendor/pygments/styles/algol.py +63 -0
- wandb/vendor/pygments/styles/algol_nu.py +63 -0
- wandb/vendor/pygments/styles/arduino.py +98 -0
- wandb/vendor/pygments/styles/autumn.py +65 -0
- wandb/vendor/pygments/styles/borland.py +51 -0
- wandb/vendor/pygments/styles/bw.py +49 -0
- wandb/vendor/pygments/styles/colorful.py +81 -0
- wandb/vendor/pygments/styles/default.py +73 -0
- wandb/vendor/pygments/styles/emacs.py +72 -0
- wandb/vendor/pygments/styles/friendly.py +72 -0
- wandb/vendor/pygments/styles/fruity.py +42 -0
- wandb/vendor/pygments/styles/igor.py +29 -0
- wandb/vendor/pygments/styles/lovelace.py +97 -0
- wandb/vendor/pygments/styles/manni.py +75 -0
- wandb/vendor/pygments/styles/monokai.py +106 -0
- wandb/vendor/pygments/styles/murphy.py +80 -0
- wandb/vendor/pygments/styles/native.py +65 -0
- wandb/vendor/pygments/styles/paraiso_dark.py +125 -0
- wandb/vendor/pygments/styles/paraiso_light.py +125 -0
- wandb/vendor/pygments/styles/pastie.py +75 -0
- wandb/vendor/pygments/styles/perldoc.py +69 -0
- wandb/vendor/pygments/styles/rainbow_dash.py +89 -0
- wandb/vendor/pygments/styles/rrt.py +33 -0
- wandb/vendor/pygments/styles/sas.py +44 -0
- wandb/vendor/pygments/styles/stata.py +40 -0
- wandb/vendor/pygments/styles/tango.py +141 -0
- wandb/vendor/pygments/styles/trac.py +63 -0
- wandb/vendor/pygments/styles/vim.py +63 -0
- wandb/vendor/pygments/styles/vs.py +38 -0
- wandb/vendor/pygments/styles/xcode.py +51 -0
- wandb/vendor/pygments/token.py +213 -0
- wandb/vendor/pygments/unistring.py +217 -0
- wandb/vendor/pygments/util.py +388 -0
- wandb/vendor/pynvml/__init__.py +0 -0
- wandb/vendor/pynvml/pynvml.py +4779 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/__init__.py +17 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/events.py +615 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/__init__.py +98 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/api.py +369 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/fsevents.py +172 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/fsevents2.py +239 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify.py +218 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify_buffer.py +81 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify_c.py +575 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/kqueue.py +730 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/polling.py +145 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/read_directory_changes.py +133 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/winapi.py +348 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/patterns.py +265 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/tricks/__init__.py +174 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/__init__.py +151 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/bricks.py +249 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/compat.py +29 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/decorators.py +198 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/delayed_queue.py +88 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/dirsnapshot.py +293 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/echo.py +157 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/event_backport.py +41 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/importlib2.py +40 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/platform.py +57 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/unicode_paths.py +64 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/win32stat.py +123 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/version.py +28 -0
- wandb/vendor/watchdog_0_9_0/wandb_watchdog/watchmedo.py +577 -0
- wandb/wandb_agent.py +588 -0
- wandb/wandb_controller.py +721 -0
- wandb/wandb_run.py +9 -0
- wandb-0.18.2.dist-info/METADATA +213 -0
- wandb-0.18.2.dist-info/RECORD +827 -0
- wandb-0.18.2.dist-info/WHEEL +5 -0
- wandb-0.18.2.dist-info/entry_points.txt +3 -0
- wandb-0.18.2.dist-info/licenses/LICENSE +21 -0
wandb/sdk/wandb_run.py
ADDED
@@ -0,0 +1,4236 @@
|
|
1
|
+
import _thread as thread
|
2
|
+
import atexit
|
3
|
+
import functools
|
4
|
+
import glob
|
5
|
+
import json
|
6
|
+
import logging
|
7
|
+
import numbers
|
8
|
+
import os
|
9
|
+
import pathlib
|
10
|
+
import re
|
11
|
+
import sys
|
12
|
+
import threading
|
13
|
+
import time
|
14
|
+
import traceback
|
15
|
+
import warnings
|
16
|
+
from collections.abc import Mapping
|
17
|
+
from dataclasses import dataclass, field
|
18
|
+
from datetime import datetime, timedelta, timezone
|
19
|
+
from enum import IntEnum
|
20
|
+
from types import TracebackType
|
21
|
+
from typing import (
|
22
|
+
TYPE_CHECKING,
|
23
|
+
Any,
|
24
|
+
Callable,
|
25
|
+
Dict,
|
26
|
+
List,
|
27
|
+
NamedTuple,
|
28
|
+
Optional,
|
29
|
+
Sequence,
|
30
|
+
TextIO,
|
31
|
+
Tuple,
|
32
|
+
Type,
|
33
|
+
Union,
|
34
|
+
)
|
35
|
+
|
36
|
+
import requests
|
37
|
+
|
38
|
+
import wandb
|
39
|
+
import wandb.env
|
40
|
+
from wandb import errors, trigger
|
41
|
+
from wandb._globals import _datatypes_set_callback
|
42
|
+
from wandb.apis import internal, public
|
43
|
+
from wandb.apis.internal import Api
|
44
|
+
from wandb.apis.public import Api as PublicApi
|
45
|
+
from wandb.errors import CommError
|
46
|
+
from wandb.integration.torch import wandb_torch
|
47
|
+
from wandb.plot.viz import CustomChart, Visualize, custom_chart
|
48
|
+
from wandb.proto.wandb_internal_pb2 import (
|
49
|
+
MetricRecord,
|
50
|
+
PollExitResponse,
|
51
|
+
Result,
|
52
|
+
RunRecord,
|
53
|
+
)
|
54
|
+
from wandb.sdk.artifacts.artifact import Artifact
|
55
|
+
from wandb.sdk.internal import job_builder
|
56
|
+
from wandb.sdk.lib.import_hooks import (
|
57
|
+
register_post_import_hook,
|
58
|
+
unregister_post_import_hook,
|
59
|
+
)
|
60
|
+
from wandb.sdk.lib.paths import FilePathStr, LogicalPath, StrPath
|
61
|
+
from wandb.util import (
|
62
|
+
_is_artifact_object,
|
63
|
+
_is_artifact_string,
|
64
|
+
_is_artifact_version_weave_dict,
|
65
|
+
_is_py_requirements_or_dockerfile,
|
66
|
+
_resolve_aliases,
|
67
|
+
add_import_hook,
|
68
|
+
parse_artifact_string,
|
69
|
+
)
|
70
|
+
|
71
|
+
from . import wandb_config, wandb_metric, wandb_summary
|
72
|
+
from .artifacts._validators import validate_aliases, validate_tags
|
73
|
+
from .data_types._dtypes import TypeRegistry
|
74
|
+
from .interface.interface import FilesDict, GlobStr, InterfaceBase, PolicyName
|
75
|
+
from .interface.summary_record import SummaryRecord
|
76
|
+
from .lib import (
|
77
|
+
config_util,
|
78
|
+
deprecate,
|
79
|
+
filenames,
|
80
|
+
filesystem,
|
81
|
+
ipython,
|
82
|
+
module,
|
83
|
+
proto_util,
|
84
|
+
redirect,
|
85
|
+
telemetry,
|
86
|
+
)
|
87
|
+
from .lib.exit_hooks import ExitHooks
|
88
|
+
from .lib.gitlib import GitRepo
|
89
|
+
from .lib.mailbox import MailboxError, MailboxHandle, MailboxProbe, MailboxProgress
|
90
|
+
from .lib.printer import get_printer
|
91
|
+
from .lib.proto_util import message_to_dict
|
92
|
+
from .lib.reporting import Reporter
|
93
|
+
from .lib.wburls import wburls
|
94
|
+
from .wandb_alerts import AlertLevel
|
95
|
+
from .wandb_settings import Settings
|
96
|
+
from .wandb_setup import _WandbSetup
|
97
|
+
|
98
|
+
if TYPE_CHECKING:
|
99
|
+
if sys.version_info >= (3, 8):
|
100
|
+
from typing import TypedDict
|
101
|
+
else:
|
102
|
+
from typing_extensions import TypedDict
|
103
|
+
|
104
|
+
import wandb.apis.public
|
105
|
+
import wandb.sdk.backend.backend
|
106
|
+
import wandb.sdk.interface.interface_queue
|
107
|
+
from wandb.proto.wandb_internal_pb2 import (
|
108
|
+
GetSummaryResponse,
|
109
|
+
InternalMessagesResponse,
|
110
|
+
SampledHistoryResponse,
|
111
|
+
)
|
112
|
+
|
113
|
+
from .lib.printer import PrinterJupyter, PrinterTerm
|
114
|
+
|
115
|
+
class GitSourceDict(TypedDict):
|
116
|
+
remote: str
|
117
|
+
commit: str
|
118
|
+
entrypoint: List[str]
|
119
|
+
args: Sequence[str]
|
120
|
+
|
121
|
+
class ArtifactSourceDict(TypedDict):
|
122
|
+
artifact: str
|
123
|
+
entrypoint: List[str]
|
124
|
+
args: Sequence[str]
|
125
|
+
|
126
|
+
class ImageSourceDict(TypedDict):
|
127
|
+
image: str
|
128
|
+
args: Sequence[str]
|
129
|
+
|
130
|
+
class JobSourceDict(TypedDict, total=False):
|
131
|
+
_version: str
|
132
|
+
source_type: str
|
133
|
+
source: Union[GitSourceDict, ArtifactSourceDict, ImageSourceDict]
|
134
|
+
input_types: Dict[str, Any]
|
135
|
+
output_types: Dict[str, Any]
|
136
|
+
runtime: Optional[str]
|
137
|
+
|
138
|
+
|
139
|
+
logger = logging.getLogger("wandb")
|
140
|
+
EXIT_TIMEOUT = 60
|
141
|
+
RE_LABEL = re.compile(r"[a-zA-Z0-9_-]+$")
|
142
|
+
|
143
|
+
|
144
|
+
class TeardownStage(IntEnum):
|
145
|
+
EARLY = 1
|
146
|
+
LATE = 2
|
147
|
+
|
148
|
+
|
149
|
+
class TeardownHook(NamedTuple):
|
150
|
+
call: Callable[[], None]
|
151
|
+
stage: TeardownStage
|
152
|
+
|
153
|
+
|
154
|
+
class RunStatusChecker:
|
155
|
+
"""Periodically polls the background process for relevant updates.
|
156
|
+
|
157
|
+
- check if the user has requested a stop.
|
158
|
+
- check the network status.
|
159
|
+
- check the run sync status.
|
160
|
+
"""
|
161
|
+
|
162
|
+
_stop_status_lock: threading.Lock
|
163
|
+
_stop_status_handle: Optional[MailboxHandle]
|
164
|
+
_network_status_lock: threading.Lock
|
165
|
+
_network_status_handle: Optional[MailboxHandle]
|
166
|
+
_internal_messages_lock: threading.Lock
|
167
|
+
_internal_messages_handle: Optional[MailboxHandle]
|
168
|
+
|
169
|
+
def __init__(
|
170
|
+
self,
|
171
|
+
interface: InterfaceBase,
|
172
|
+
stop_polling_interval: int = 15,
|
173
|
+
retry_polling_interval: int = 5,
|
174
|
+
internal_messages_polling_interval: int = 10,
|
175
|
+
) -> None:
|
176
|
+
self._interface = interface
|
177
|
+
self._stop_polling_interval = stop_polling_interval
|
178
|
+
self._retry_polling_interval = retry_polling_interval
|
179
|
+
self._internal_messages_polling_interval = internal_messages_polling_interval
|
180
|
+
|
181
|
+
self._join_event = threading.Event()
|
182
|
+
|
183
|
+
self._stop_status_lock = threading.Lock()
|
184
|
+
self._stop_status_handle = None
|
185
|
+
self._stop_thread = threading.Thread(
|
186
|
+
target=self.check_stop_status,
|
187
|
+
name="ChkStopThr",
|
188
|
+
daemon=True,
|
189
|
+
)
|
190
|
+
|
191
|
+
self._network_status_lock = threading.Lock()
|
192
|
+
self._network_status_handle = None
|
193
|
+
self._network_status_thread = threading.Thread(
|
194
|
+
target=self.check_network_status,
|
195
|
+
name="NetStatThr",
|
196
|
+
daemon=True,
|
197
|
+
)
|
198
|
+
|
199
|
+
self._internal_messages_lock = threading.Lock()
|
200
|
+
self._internal_messages_handle = None
|
201
|
+
self._internal_messages_thread = threading.Thread(
|
202
|
+
target=self.check_internal_messages,
|
203
|
+
name="IntMsgThr",
|
204
|
+
daemon=True,
|
205
|
+
)
|
206
|
+
|
207
|
+
def start(self) -> None:
|
208
|
+
self._stop_thread.start()
|
209
|
+
self._network_status_thread.start()
|
210
|
+
self._internal_messages_thread.start()
|
211
|
+
|
212
|
+
@staticmethod
|
213
|
+
def _abandon_status_check(
|
214
|
+
lock: threading.Lock,
|
215
|
+
handle: Optional[MailboxHandle],
|
216
|
+
):
|
217
|
+
with lock:
|
218
|
+
if handle:
|
219
|
+
handle.abandon()
|
220
|
+
|
221
|
+
def _loop_check_status(
|
222
|
+
self,
|
223
|
+
*,
|
224
|
+
lock: threading.Lock,
|
225
|
+
set_handle: Any,
|
226
|
+
timeout: int,
|
227
|
+
request: Any,
|
228
|
+
process: Any,
|
229
|
+
) -> None:
|
230
|
+
local_handle: Optional[MailboxHandle] = None
|
231
|
+
join_requested = False
|
232
|
+
while not join_requested:
|
233
|
+
time_probe = time.monotonic()
|
234
|
+
if not local_handle:
|
235
|
+
local_handle = request()
|
236
|
+
assert local_handle
|
237
|
+
|
238
|
+
with lock:
|
239
|
+
if self._join_event.is_set():
|
240
|
+
break
|
241
|
+
set_handle(local_handle)
|
242
|
+
try:
|
243
|
+
result = local_handle.wait(timeout=timeout, release=False)
|
244
|
+
except MailboxError:
|
245
|
+
# background threads are oportunistically getting results
|
246
|
+
# from the internal process but the internal process could
|
247
|
+
# be shutdown at any time. In this case assume that the
|
248
|
+
# thread should exit silently. This is possible
|
249
|
+
# because we do not have an atexit handler for the user
|
250
|
+
# process which quiesces active threads.
|
251
|
+
break
|
252
|
+
with lock:
|
253
|
+
set_handle(None)
|
254
|
+
|
255
|
+
if result:
|
256
|
+
process(result)
|
257
|
+
# if request finished, clear the handle to send on the next interval
|
258
|
+
local_handle.abandon()
|
259
|
+
local_handle = None
|
260
|
+
|
261
|
+
time_elapsed = time.monotonic() - time_probe
|
262
|
+
wait_time = max(timeout - time_elapsed, 0)
|
263
|
+
join_requested = self._join_event.wait(timeout=wait_time)
|
264
|
+
|
265
|
+
def check_network_status(self) -> None:
|
266
|
+
def _process_network_status(result: Result) -> None:
|
267
|
+
network_status = result.response.network_status_response
|
268
|
+
for hr in network_status.network_responses:
|
269
|
+
if (
|
270
|
+
hr.http_status_code == 200 or hr.http_status_code == 0
|
271
|
+
): # we use 0 for non-http errors (eg wandb errors)
|
272
|
+
wandb.termlog(f"{hr.http_response_text}")
|
273
|
+
else:
|
274
|
+
wandb.termlog(
|
275
|
+
"{} encountered ({}), retrying request".format(
|
276
|
+
hr.http_status_code, hr.http_response_text.rstrip()
|
277
|
+
)
|
278
|
+
)
|
279
|
+
|
280
|
+
try:
|
281
|
+
self._loop_check_status(
|
282
|
+
lock=self._network_status_lock,
|
283
|
+
set_handle=lambda x: setattr(self, "_network_status_handle", x),
|
284
|
+
timeout=self._retry_polling_interval,
|
285
|
+
request=self._interface.deliver_network_status,
|
286
|
+
process=_process_network_status,
|
287
|
+
)
|
288
|
+
except BrokenPipeError:
|
289
|
+
self._abandon_status_check(
|
290
|
+
self._network_status_lock,
|
291
|
+
self._network_status_handle,
|
292
|
+
)
|
293
|
+
|
294
|
+
def check_stop_status(self) -> None:
|
295
|
+
def _process_stop_status(result: Result) -> None:
|
296
|
+
stop_status = result.response.stop_status_response
|
297
|
+
if stop_status.run_should_stop:
|
298
|
+
# TODO(frz): This check is required
|
299
|
+
# until WB-3606 is resolved on server side.
|
300
|
+
if not wandb.agents.pyagent.is_running(): # type: ignore
|
301
|
+
thread.interrupt_main()
|
302
|
+
return
|
303
|
+
|
304
|
+
try:
|
305
|
+
self._loop_check_status(
|
306
|
+
lock=self._stop_status_lock,
|
307
|
+
set_handle=lambda x: setattr(self, "_stop_status_handle", x),
|
308
|
+
timeout=self._stop_polling_interval,
|
309
|
+
request=self._interface.deliver_stop_status,
|
310
|
+
process=_process_stop_status,
|
311
|
+
)
|
312
|
+
except BrokenPipeError:
|
313
|
+
self._abandon_status_check(
|
314
|
+
self._stop_status_lock,
|
315
|
+
self._stop_status_handle,
|
316
|
+
)
|
317
|
+
|
318
|
+
def check_internal_messages(self) -> None:
|
319
|
+
def _process_internal_messages(result: Result) -> None:
|
320
|
+
internal_messages = result.response.internal_messages_response
|
321
|
+
for msg in internal_messages.messages.warning:
|
322
|
+
wandb.termwarn(msg)
|
323
|
+
|
324
|
+
try:
|
325
|
+
self._loop_check_status(
|
326
|
+
lock=self._internal_messages_lock,
|
327
|
+
set_handle=lambda x: setattr(self, "_internal_messages_handle", x),
|
328
|
+
timeout=self._internal_messages_polling_interval,
|
329
|
+
request=self._interface.deliver_internal_messages,
|
330
|
+
process=_process_internal_messages,
|
331
|
+
)
|
332
|
+
except BrokenPipeError:
|
333
|
+
self._abandon_status_check(
|
334
|
+
self._internal_messages_lock,
|
335
|
+
self._internal_messages_handle,
|
336
|
+
)
|
337
|
+
|
338
|
+
def stop(self) -> None:
|
339
|
+
self._join_event.set()
|
340
|
+
self._abandon_status_check(
|
341
|
+
self._stop_status_lock,
|
342
|
+
self._stop_status_handle,
|
343
|
+
)
|
344
|
+
self._abandon_status_check(
|
345
|
+
self._network_status_lock,
|
346
|
+
self._network_status_handle,
|
347
|
+
)
|
348
|
+
self._abandon_status_check(
|
349
|
+
self._internal_messages_lock,
|
350
|
+
self._internal_messages_handle,
|
351
|
+
)
|
352
|
+
|
353
|
+
def join(self) -> None:
|
354
|
+
self.stop()
|
355
|
+
self._stop_thread.join()
|
356
|
+
self._network_status_thread.join()
|
357
|
+
self._internal_messages_thread.join()
|
358
|
+
|
359
|
+
|
360
|
+
class _run_decorator: # noqa: N801
|
361
|
+
_is_attaching: str = ""
|
362
|
+
|
363
|
+
class Dummy: ...
|
364
|
+
|
365
|
+
@classmethod
|
366
|
+
def _attach(cls, func: Callable) -> Callable:
|
367
|
+
@functools.wraps(func)
|
368
|
+
def wrapper(self: Type["Run"], *args: Any, **kwargs: Any) -> Any:
|
369
|
+
# * `_attach_id` is only assigned in service hence for all non-service cases
|
370
|
+
# it will be a passthrough.
|
371
|
+
# * `_attach_pid` is only assigned in _init (using _attach_pid guarantees single attach):
|
372
|
+
# - for non-fork case the object is shared through pickling so will be None.
|
373
|
+
# - for fork case the new process share mem space hence the value would be of parent process.
|
374
|
+
if (
|
375
|
+
getattr(self, "_attach_id", None)
|
376
|
+
and getattr(self, "_attach_pid", None) != os.getpid()
|
377
|
+
):
|
378
|
+
if cls._is_attaching:
|
379
|
+
message = (
|
380
|
+
f"Trying to attach `{func.__name__}` "
|
381
|
+
f"while in the middle of attaching `{cls._is_attaching}`"
|
382
|
+
)
|
383
|
+
raise RuntimeError(message)
|
384
|
+
cls._is_attaching = func.__name__
|
385
|
+
try:
|
386
|
+
wandb._attach(run=self) # type: ignore
|
387
|
+
except Exception as e:
|
388
|
+
# In case the attach fails we will raise the exception that caused the issue.
|
389
|
+
# This exception should be caught and fail the execution of the program.
|
390
|
+
cls._is_attaching = ""
|
391
|
+
raise e
|
392
|
+
cls._is_attaching = ""
|
393
|
+
return func(self, *args, **kwargs)
|
394
|
+
|
395
|
+
return wrapper
|
396
|
+
|
397
|
+
@classmethod
|
398
|
+
def _noop_on_finish(cls, message: str = "", only_warn: bool = False) -> Callable:
|
399
|
+
def decorator_fn(func: Callable) -> Callable:
|
400
|
+
@functools.wraps(func)
|
401
|
+
def wrapper_fn(self: Type["Run"], *args: Any, **kwargs: Any) -> Any:
|
402
|
+
if not getattr(self, "_is_finished", False):
|
403
|
+
return func(self, *args, **kwargs)
|
404
|
+
|
405
|
+
default_message = (
|
406
|
+
f"Run ({self.id}) is finished. The call to `{func.__name__}` will be ignored. "
|
407
|
+
f"Please make sure that you are using an active run."
|
408
|
+
)
|
409
|
+
resolved_message = message or default_message
|
410
|
+
if only_warn:
|
411
|
+
warnings.warn(resolved_message, UserWarning, stacklevel=2)
|
412
|
+
else:
|
413
|
+
raise errors.UsageError(resolved_message)
|
414
|
+
|
415
|
+
return wrapper_fn
|
416
|
+
|
417
|
+
return decorator_fn
|
418
|
+
|
419
|
+
@classmethod
|
420
|
+
def _noop(cls, func: Callable) -> Callable:
|
421
|
+
@functools.wraps(func)
|
422
|
+
def wrapper(self: Type["Run"], *args: Any, **kwargs: Any) -> Any:
|
423
|
+
# `_attach_id` is only assigned in service hence for all service cases
|
424
|
+
# it will be a passthrough. We don't pickle non-service so again a way
|
425
|
+
# to see that we are in non-service case
|
426
|
+
if getattr(self, "_attach_id", None) is None:
|
427
|
+
# `_init_pid` is only assigned in __init__ (this will be constant check for mp):
|
428
|
+
# - for non-fork case the object is shared through pickling,
|
429
|
+
# and we don't pickle non-service so will be None
|
430
|
+
# - for fork case the new process share mem space hence the value would be of parent process.
|
431
|
+
_init_pid = getattr(self, "_init_pid", None)
|
432
|
+
if _init_pid != os.getpid():
|
433
|
+
message = "`{}` ignored (called from pid={}, `init` called from pid={}). See: {}".format(
|
434
|
+
func.__name__,
|
435
|
+
os.getpid(),
|
436
|
+
_init_pid,
|
437
|
+
wburls.get("multiprocess"),
|
438
|
+
)
|
439
|
+
# - if this process was pickled in non-service case,
|
440
|
+
# we ignore the attributes (since pickle is not supported)
|
441
|
+
# - for fork case will use the settings of the parent process
|
442
|
+
# - only point of inconsistent behavior from forked and non-forked cases
|
443
|
+
settings = getattr(self, "_settings", None)
|
444
|
+
if settings and settings["strict"]:
|
445
|
+
wandb.termerror(message, repeat=False)
|
446
|
+
raise errors.UnsupportedError(
|
447
|
+
f"`{func.__name__}` does not support multiprocessing"
|
448
|
+
)
|
449
|
+
wandb.termwarn(message, repeat=False)
|
450
|
+
return cls.Dummy()
|
451
|
+
|
452
|
+
return func(self, *args, **kwargs)
|
453
|
+
|
454
|
+
return wrapper
|
455
|
+
|
456
|
+
|
457
|
+
@dataclass
|
458
|
+
class RunStatus:
|
459
|
+
sync_items_total: int = field(default=0)
|
460
|
+
sync_items_pending: int = field(default=0)
|
461
|
+
sync_time: Optional[datetime] = field(default=None)
|
462
|
+
|
463
|
+
|
464
|
+
class Run:
|
465
|
+
"""A unit of computation logged by wandb. Typically, this is an ML experiment.
|
466
|
+
|
467
|
+
Create a run with `wandb.init()`:
|
468
|
+
<!--yeadoc-test:run-object-basic-->
|
469
|
+
```python
|
470
|
+
import wandb
|
471
|
+
|
472
|
+
run = wandb.init()
|
473
|
+
```
|
474
|
+
|
475
|
+
There is only ever at most one active `wandb.Run` in any process,
|
476
|
+
and it is accessible as `wandb.run`:
|
477
|
+
<!--yeadoc-test:global-run-object-->
|
478
|
+
```python
|
479
|
+
import wandb
|
480
|
+
|
481
|
+
assert wandb.run is None
|
482
|
+
|
483
|
+
wandb.init()
|
484
|
+
|
485
|
+
assert wandb.run is not None
|
486
|
+
```
|
487
|
+
anything you log with `wandb.log` will be sent to that run.
|
488
|
+
|
489
|
+
If you want to start more runs in the same script or notebook, you'll need to
|
490
|
+
finish the run that is in-flight. Runs can be finished with `wandb.finish` or
|
491
|
+
by using them in a `with` block:
|
492
|
+
<!--yeadoc-test:run-context-manager-->
|
493
|
+
```python
|
494
|
+
import wandb
|
495
|
+
|
496
|
+
wandb.init()
|
497
|
+
wandb.finish()
|
498
|
+
|
499
|
+
assert wandb.run is None
|
500
|
+
|
501
|
+
with wandb.init() as run:
|
502
|
+
pass # log data here
|
503
|
+
|
504
|
+
assert wandb.run is None
|
505
|
+
```
|
506
|
+
|
507
|
+
See the documentation for `wandb.init` for more on creating runs, or check out
|
508
|
+
[our guide to `wandb.init`](https://docs.wandb.ai/guides/track/launch).
|
509
|
+
|
510
|
+
In distributed training, you can either create a single run in the rank 0 process
|
511
|
+
and then log information only from that process, or you can create a run in each process,
|
512
|
+
logging from each separately, and group the results together with the `group` argument
|
513
|
+
to `wandb.init`. For more details on distributed training with W&B, check out
|
514
|
+
[our guide](https://docs.wandb.ai/guides/track/log/distributed-training).
|
515
|
+
|
516
|
+
Currently, there is a parallel `Run` object in the `wandb.Api`. Eventually these
|
517
|
+
two objects will be merged.
|
518
|
+
|
519
|
+
Attributes:
|
520
|
+
summary: (Summary) Single values set for each `wandb.log()` key. By
|
521
|
+
default, summary is set to the last value logged. You can manually
|
522
|
+
set summary to the best value, like max accuracy, instead of the
|
523
|
+
final value.
|
524
|
+
"""
|
525
|
+
|
526
|
+
_telemetry_obj: telemetry.TelemetryRecord
|
527
|
+
_telemetry_obj_active: bool
|
528
|
+
_telemetry_obj_dirty: bool
|
529
|
+
_telemetry_obj_flushed: bytes
|
530
|
+
|
531
|
+
_teardown_hooks: List[TeardownHook]
|
532
|
+
_tags: Optional[Tuple[Any, ...]]
|
533
|
+
|
534
|
+
_entity: Optional[str]
|
535
|
+
_project: Optional[str]
|
536
|
+
_group: Optional[str]
|
537
|
+
_job_type: Optional[str]
|
538
|
+
_name: Optional[str]
|
539
|
+
_notes: Optional[str]
|
540
|
+
_sweep_id: Optional[str]
|
541
|
+
|
542
|
+
_run_obj: Optional[RunRecord]
|
543
|
+
# Use string literal annotation because of type reference loop
|
544
|
+
_backend: Optional["wandb.sdk.backend.backend.Backend"]
|
545
|
+
_internal_run_interface: Optional[
|
546
|
+
"wandb.sdk.interface.interface_queue.InterfaceQueue"
|
547
|
+
]
|
548
|
+
_wl: Optional[_WandbSetup]
|
549
|
+
|
550
|
+
_out_redir: Optional[redirect.RedirectBase]
|
551
|
+
_err_redir: Optional[redirect.RedirectBase]
|
552
|
+
_redirect_cb: Optional[Callable[[str, str], None]]
|
553
|
+
_redirect_raw_cb: Optional[Callable[[str, str], None]]
|
554
|
+
_output_writer: Optional["filesystem.CRDedupedFile"]
|
555
|
+
_quiet: Optional[bool]
|
556
|
+
|
557
|
+
_atexit_cleanup_called: bool
|
558
|
+
_hooks: Optional[ExitHooks]
|
559
|
+
_exit_code: Optional[int]
|
560
|
+
|
561
|
+
_run_status_checker: Optional[RunStatusChecker]
|
562
|
+
|
563
|
+
_sampled_history: Optional["SampledHistoryResponse"]
|
564
|
+
_final_summary: Optional["GetSummaryResponse"]
|
565
|
+
_poll_exit_handle: Optional[MailboxHandle]
|
566
|
+
_poll_exit_response: Optional[PollExitResponse]
|
567
|
+
_internal_messages_response: Optional["InternalMessagesResponse"]
|
568
|
+
|
569
|
+
_stdout_slave_fd: Optional[int]
|
570
|
+
_stderr_slave_fd: Optional[int]
|
571
|
+
_artifact_slots: List[str]
|
572
|
+
|
573
|
+
_init_pid: int
|
574
|
+
_attach_pid: int
|
575
|
+
_iface_pid: Optional[int]
|
576
|
+
_iface_port: Optional[int]
|
577
|
+
|
578
|
+
_attach_id: Optional[str]
|
579
|
+
_is_attached: bool
|
580
|
+
_is_finished: bool
|
581
|
+
_settings: Settings
|
582
|
+
|
583
|
+
_launch_artifacts: Optional[Dict[str, Any]]
|
584
|
+
_printer: Union["PrinterTerm", "PrinterJupyter"]
|
585
|
+
|
586
|
+
def __init__(
|
587
|
+
self,
|
588
|
+
settings: Settings,
|
589
|
+
config: Optional[Dict[str, Any]] = None,
|
590
|
+
sweep_config: Optional[Dict[str, Any]] = None,
|
591
|
+
launch_config: Optional[Dict[str, Any]] = None,
|
592
|
+
) -> None:
|
593
|
+
# pid is set, so we know if this run object was initialized by this process
|
594
|
+
self._init_pid = os.getpid()
|
595
|
+
|
596
|
+
if settings._noop:
|
597
|
+
# TODO: properly handle setting for disabled mode
|
598
|
+
self._settings = settings
|
599
|
+
return
|
600
|
+
|
601
|
+
self._init(
|
602
|
+
settings=settings,
|
603
|
+
config=config,
|
604
|
+
sweep_config=sweep_config,
|
605
|
+
launch_config=launch_config,
|
606
|
+
)
|
607
|
+
|
608
|
+
def _init(
|
609
|
+
self,
|
610
|
+
settings: Settings,
|
611
|
+
config: Optional[Dict[str, Any]] = None,
|
612
|
+
sweep_config: Optional[Dict[str, Any]] = None,
|
613
|
+
launch_config: Optional[Dict[str, Any]] = None,
|
614
|
+
) -> None:
|
615
|
+
self._settings = settings
|
616
|
+
|
617
|
+
self._config = wandb_config.Config()
|
618
|
+
self._config._set_callback(self._config_callback)
|
619
|
+
self._config._set_artifact_callback(self._config_artifact_callback)
|
620
|
+
self._config._set_settings(self._settings)
|
621
|
+
self._backend = None
|
622
|
+
self._internal_run_interface = None
|
623
|
+
# todo: perhaps this should be a property that is a noop on a finished run
|
624
|
+
self.summary = wandb_summary.Summary(
|
625
|
+
self._summary_get_current_summary_callback,
|
626
|
+
)
|
627
|
+
self.summary._set_update_callback(self._summary_update_callback)
|
628
|
+
self._step = 0
|
629
|
+
self._torch_history: Optional[wandb_torch.TorchHistory] = None # type: ignore
|
630
|
+
|
631
|
+
# todo: eventually would be nice to make this configurable using self._settings._start_time
|
632
|
+
# need to test (jhr): if you set start time to 2 days ago and run a test for 15 minutes,
|
633
|
+
# does the total time get calculated right (not as 2 days and 15 minutes)?
|
634
|
+
self._start_time = time.time()
|
635
|
+
|
636
|
+
_datatypes_set_callback(self._datatypes_callback)
|
637
|
+
|
638
|
+
self._printer = get_printer(self._settings._jupyter)
|
639
|
+
self._wl = None
|
640
|
+
self._reporter: Optional[Reporter] = None
|
641
|
+
|
642
|
+
self._entity = None
|
643
|
+
self._project = None
|
644
|
+
self._group = None
|
645
|
+
self._job_type = None
|
646
|
+
self._run_id = self._settings.run_id
|
647
|
+
self._starting_step = 0
|
648
|
+
self._name = None
|
649
|
+
self._notes = None
|
650
|
+
self._tags = None
|
651
|
+
self._remote_url = None
|
652
|
+
self._commit = None
|
653
|
+
self._sweep_id = None
|
654
|
+
|
655
|
+
self._hooks = None
|
656
|
+
self._teardown_hooks = []
|
657
|
+
self._out_redir = None
|
658
|
+
self._err_redir = None
|
659
|
+
self._stdout_slave_fd = None
|
660
|
+
self._stderr_slave_fd = None
|
661
|
+
self._exit_code = None
|
662
|
+
self._exit_result = None
|
663
|
+
self._quiet = self._settings.quiet
|
664
|
+
|
665
|
+
self._output_writer = None
|
666
|
+
self._used_artifact_slots: Dict[str, str] = {}
|
667
|
+
|
668
|
+
# Returned from backend request_run(), set from wandb_init?
|
669
|
+
self._run_obj = None
|
670
|
+
|
671
|
+
# Created when the run "starts".
|
672
|
+
self._run_status_checker = None
|
673
|
+
|
674
|
+
self._sampled_history = None
|
675
|
+
self._final_summary = None
|
676
|
+
self._poll_exit_response = None
|
677
|
+
self._internal_messages_response = None
|
678
|
+
self._poll_exit_handle = None
|
679
|
+
|
680
|
+
# Initialize telemetry object
|
681
|
+
self._telemetry_obj = telemetry.TelemetryRecord()
|
682
|
+
self._telemetry_obj_active = False
|
683
|
+
self._telemetry_obj_flushed = b""
|
684
|
+
self._telemetry_obj_dirty = False
|
685
|
+
|
686
|
+
self._atexit_cleanup_called = False
|
687
|
+
|
688
|
+
# Pull info from settings
|
689
|
+
self._init_from_settings(self._settings)
|
690
|
+
|
691
|
+
# Initial scope setup for sentry.
|
692
|
+
# This might get updated when the actual run comes back.
|
693
|
+
wandb._sentry.configure_scope(
|
694
|
+
tags=dict(self._settings),
|
695
|
+
process_context="user",
|
696
|
+
)
|
697
|
+
|
698
|
+
# Populate config
|
699
|
+
config = config or dict()
|
700
|
+
wandb_key = "_wandb"
|
701
|
+
config.setdefault(wandb_key, dict())
|
702
|
+
self._launch_artifact_mapping: Dict[str, Any] = {}
|
703
|
+
self._unique_launch_artifact_sequence_names: Dict[str, Any] = {}
|
704
|
+
if self._settings.save_code and self._settings.program_relpath:
|
705
|
+
config[wandb_key]["code_path"] = LogicalPath(
|
706
|
+
os.path.join("code", self._settings.program_relpath)
|
707
|
+
)
|
708
|
+
|
709
|
+
if self._settings.fork_from is not None:
|
710
|
+
config[wandb_key]["branch_point"] = {
|
711
|
+
"run_id": self._settings.fork_from.run,
|
712
|
+
"step": self._settings.fork_from.value,
|
713
|
+
}
|
714
|
+
|
715
|
+
if self._settings.resume_from is not None:
|
716
|
+
config[wandb_key]["branch_point"] = {
|
717
|
+
"run_id": self._settings.resume_from.run,
|
718
|
+
"step": self._settings.resume_from.value,
|
719
|
+
}
|
720
|
+
|
721
|
+
self._config._update(config, ignore_locked=True)
|
722
|
+
|
723
|
+
if sweep_config:
|
724
|
+
self._config.merge_locked(
|
725
|
+
sweep_config, user="sweep", _allow_val_change=True
|
726
|
+
)
|
727
|
+
|
728
|
+
if launch_config:
|
729
|
+
self._config.merge_locked(
|
730
|
+
launch_config, user="launch", _allow_val_change=True
|
731
|
+
)
|
732
|
+
|
733
|
+
# if run is from a launch queue, add queue id to _wandb config
|
734
|
+
launch_queue_name = wandb.env.get_launch_queue_name()
|
735
|
+
if launch_queue_name:
|
736
|
+
self._config[wandb_key]["launch_queue_name"] = launch_queue_name
|
737
|
+
|
738
|
+
launch_queue_entity = wandb.env.get_launch_queue_entity()
|
739
|
+
if launch_queue_entity:
|
740
|
+
self._config[wandb_key]["launch_queue_entity"] = launch_queue_entity
|
741
|
+
|
742
|
+
launch_trace_id = wandb.env.get_launch_trace_id()
|
743
|
+
if launch_trace_id:
|
744
|
+
self._config[wandb_key]["launch_trace_id"] = launch_trace_id
|
745
|
+
|
746
|
+
# interface pid and port configured when backend is configured (See _hack_set_run)
|
747
|
+
# TODO: using pid isn't the best for windows as pid reuse can happen more often than unix
|
748
|
+
self._iface_pid = None
|
749
|
+
self._iface_port = None
|
750
|
+
self._attach_id = None
|
751
|
+
self._is_attached = False
|
752
|
+
self._is_finished = False
|
753
|
+
|
754
|
+
self._attach_pid = os.getpid()
|
755
|
+
|
756
|
+
# for now, use runid as attach id, this could/should be versioned in the future
|
757
|
+
if not self._settings._disable_service:
|
758
|
+
self._attach_id = self._settings.run_id
|
759
|
+
|
760
|
+
def _set_iface_pid(self, iface_pid: int) -> None:
|
761
|
+
self._iface_pid = iface_pid
|
762
|
+
|
763
|
+
def _set_iface_port(self, iface_port: int) -> None:
|
764
|
+
self._iface_port = iface_port
|
765
|
+
|
766
|
+
def _handle_launch_artifact_overrides(self) -> None:
|
767
|
+
if self._settings.launch and (os.environ.get("WANDB_ARTIFACTS") is not None):
|
768
|
+
try:
|
769
|
+
artifacts: Dict[str, Any] = json.loads(
|
770
|
+
os.environ.get("WANDB_ARTIFACTS", "{}")
|
771
|
+
)
|
772
|
+
except (ValueError, SyntaxError):
|
773
|
+
wandb.termwarn("Malformed WANDB_ARTIFACTS, using original artifacts")
|
774
|
+
else:
|
775
|
+
self._initialize_launch_artifact_maps(artifacts)
|
776
|
+
|
777
|
+
elif (
|
778
|
+
self._settings.launch
|
779
|
+
and self._settings.launch_config_path
|
780
|
+
and os.path.exists(self._settings.launch_config_path)
|
781
|
+
):
|
782
|
+
self.save(self._settings.launch_config_path)
|
783
|
+
with open(self._settings.launch_config_path) as fp:
|
784
|
+
launch_config = json.loads(fp.read())
|
785
|
+
if launch_config.get("overrides", {}).get("artifacts") is not None:
|
786
|
+
artifacts = launch_config.get("overrides").get("artifacts")
|
787
|
+
self._initialize_launch_artifact_maps(artifacts)
|
788
|
+
|
789
|
+
def _initialize_launch_artifact_maps(self, artifacts: Dict[str, Any]) -> None:
|
790
|
+
for key, item in artifacts.items():
|
791
|
+
self._launch_artifact_mapping[key] = item
|
792
|
+
artifact_sequence_tuple_or_slot = key.split(":")
|
793
|
+
|
794
|
+
if len(artifact_sequence_tuple_or_slot) == 2:
|
795
|
+
sequence_name = artifact_sequence_tuple_or_slot[0].split("/")[-1]
|
796
|
+
if self._unique_launch_artifact_sequence_names.get(sequence_name):
|
797
|
+
self._unique_launch_artifact_sequence_names.pop(sequence_name)
|
798
|
+
else:
|
799
|
+
self._unique_launch_artifact_sequence_names[sequence_name] = item
|
800
|
+
|
801
|
+
def _telemetry_callback(self, telem_obj: telemetry.TelemetryRecord) -> None:
|
802
|
+
if not hasattr(self, "_telemetry_obj"):
|
803
|
+
return
|
804
|
+
self._telemetry_obj.MergeFrom(telem_obj)
|
805
|
+
self._telemetry_obj_dirty = True
|
806
|
+
self._telemetry_flush()
|
807
|
+
|
808
|
+
def _telemetry_flush(self) -> None:
|
809
|
+
if not hasattr(self, "_telemetry_obj"):
|
810
|
+
return
|
811
|
+
if not self._telemetry_obj_active:
|
812
|
+
return
|
813
|
+
if not self._telemetry_obj_dirty:
|
814
|
+
return
|
815
|
+
if self._backend and self._backend.interface:
|
816
|
+
serialized = self._telemetry_obj.SerializeToString()
|
817
|
+
if serialized == self._telemetry_obj_flushed:
|
818
|
+
return
|
819
|
+
self._backend.interface._publish_telemetry(self._telemetry_obj)
|
820
|
+
self._telemetry_obj_flushed = serialized
|
821
|
+
self._telemetry_obj_dirty = False
|
822
|
+
|
823
|
+
def _freeze(self) -> None:
|
824
|
+
self._frozen = True
|
825
|
+
|
826
|
+
def __setattr__(self, attr: str, value: object) -> None:
|
827
|
+
if getattr(self, "_frozen", None) and not hasattr(self, attr):
|
828
|
+
raise Exception(f"Attribute {attr} is not supported on Run object.")
|
829
|
+
super().__setattr__(attr, value)
|
830
|
+
|
831
|
+
def _update_settings(self, settings: Settings) -> None:
|
832
|
+
self._settings = settings
|
833
|
+
self._init_from_settings(settings)
|
834
|
+
|
835
|
+
def _init_from_settings(self, settings: Settings) -> None:
|
836
|
+
if settings.entity is not None:
|
837
|
+
self._entity = settings.entity
|
838
|
+
if settings.project is not None:
|
839
|
+
self._project = settings.project
|
840
|
+
if settings.run_group is not None:
|
841
|
+
self._group = settings.run_group
|
842
|
+
if settings.run_job_type is not None:
|
843
|
+
self._job_type = settings.run_job_type
|
844
|
+
if settings.run_name is not None:
|
845
|
+
self._name = settings.run_name
|
846
|
+
if settings.run_notes is not None:
|
847
|
+
self._notes = settings.run_notes
|
848
|
+
if settings.run_tags is not None:
|
849
|
+
self._tags = settings.run_tags
|
850
|
+
if settings.sweep_id is not None:
|
851
|
+
self._sweep_id = settings.sweep_id
|
852
|
+
|
853
|
+
def _make_proto_run(self, run: RunRecord) -> None:
|
854
|
+
"""Populate protocol buffer RunData for interface/interface."""
|
855
|
+
if self._entity is not None:
|
856
|
+
run.entity = self._entity
|
857
|
+
if self._project is not None:
|
858
|
+
run.project = self._project
|
859
|
+
if self._group is not None:
|
860
|
+
run.run_group = self._group
|
861
|
+
if self._job_type is not None:
|
862
|
+
run.job_type = self._job_type
|
863
|
+
if self._run_id is not None:
|
864
|
+
run.run_id = self._run_id
|
865
|
+
if self._name is not None:
|
866
|
+
run.display_name = self._name
|
867
|
+
if self._notes is not None:
|
868
|
+
run.notes = self._notes
|
869
|
+
if self._tags is not None:
|
870
|
+
for tag in self._tags:
|
871
|
+
run.tags.append(tag)
|
872
|
+
if self._start_time is not None:
|
873
|
+
run.start_time.FromMicroseconds(int(self._start_time * 1e6))
|
874
|
+
if self._remote_url is not None:
|
875
|
+
run.git.remote_url = self._remote_url
|
876
|
+
if self._commit is not None:
|
877
|
+
run.git.commit = self._commit
|
878
|
+
if self._sweep_id is not None:
|
879
|
+
run.sweep_id = self._sweep_id
|
880
|
+
# Note: run.config is set in interface/interface:_make_run()
|
881
|
+
|
882
|
+
def _populate_git_info(self) -> None:
|
883
|
+
# Use user provided git info if available otherwise resolve it from the environment
|
884
|
+
try:
|
885
|
+
repo = GitRepo(
|
886
|
+
root=self._settings.git_root,
|
887
|
+
remote=self._settings.git_remote,
|
888
|
+
remote_url=self._settings.git_remote_url,
|
889
|
+
commit=self._settings.git_commit,
|
890
|
+
lazy=False,
|
891
|
+
)
|
892
|
+
self._remote_url, self._commit = repo.remote_url, repo.last_commit
|
893
|
+
except Exception:
|
894
|
+
wandb.termwarn("Cannot find valid git repo associated with this directory.")
|
895
|
+
|
896
|
+
def __deepcopy__(self, memo: Dict[int, Any]) -> "Run":
|
897
|
+
return self
|
898
|
+
|
899
|
+
def __getstate__(self) -> Any:
|
900
|
+
"""Return run state as a custom pickle."""
|
901
|
+
# We only pickle in service mode
|
902
|
+
if not self._settings or self._settings._disable_service:
|
903
|
+
return
|
904
|
+
|
905
|
+
_attach_id = self._attach_id
|
906
|
+
if not _attach_id:
|
907
|
+
return
|
908
|
+
|
909
|
+
return dict(
|
910
|
+
_attach_id=_attach_id,
|
911
|
+
_init_pid=self._init_pid,
|
912
|
+
_is_finished=self._is_finished,
|
913
|
+
)
|
914
|
+
|
915
|
+
def __setstate__(self, state: Any) -> None:
|
916
|
+
"""Set run state from a custom pickle."""
|
917
|
+
if not state:
|
918
|
+
return
|
919
|
+
|
920
|
+
_attach_id = state.get("_attach_id")
|
921
|
+
if not _attach_id:
|
922
|
+
return
|
923
|
+
|
924
|
+
if state["_init_pid"] == os.getpid():
|
925
|
+
raise RuntimeError("attach in the same process is not supported currently")
|
926
|
+
|
927
|
+
self.__dict__.update(state)
|
928
|
+
|
929
|
+
@property
|
930
|
+
def _torch(self) -> "wandb_torch.TorchHistory": # type: ignore
|
931
|
+
if self._torch_history is None:
|
932
|
+
self._torch_history = wandb_torch.TorchHistory() # type: ignore
|
933
|
+
return self._torch_history
|
934
|
+
|
935
|
+
@property
|
936
|
+
@_run_decorator._attach
|
937
|
+
def settings(self) -> Settings:
|
938
|
+
"""A frozen copy of run's Settings object."""
|
939
|
+
cp = self._settings.copy()
|
940
|
+
cp.freeze()
|
941
|
+
return cp
|
942
|
+
|
943
|
+
@property
|
944
|
+
@_run_decorator._attach
|
945
|
+
def dir(self) -> str:
|
946
|
+
"""The directory where files associated with the run are saved."""
|
947
|
+
return self._settings.files_dir
|
948
|
+
|
949
|
+
@property
|
950
|
+
@_run_decorator._attach
|
951
|
+
def config(self) -> wandb_config.Config:
|
952
|
+
"""Config object associated with this run."""
|
953
|
+
return self._config
|
954
|
+
|
955
|
+
@property
|
956
|
+
@_run_decorator._attach
|
957
|
+
def config_static(self) -> wandb_config.ConfigStatic:
|
958
|
+
return wandb_config.ConfigStatic(self._config)
|
959
|
+
|
960
|
+
@property
|
961
|
+
@_run_decorator._attach
|
962
|
+
def name(self) -> Optional[str]:
|
963
|
+
"""Display name of the run.
|
964
|
+
|
965
|
+
Display names are not guaranteed to be unique and may be descriptive.
|
966
|
+
By default, they are randomly generated.
|
967
|
+
"""
|
968
|
+
if self._name:
|
969
|
+
return self._name
|
970
|
+
if not self._run_obj:
|
971
|
+
return None
|
972
|
+
return self._run_obj.display_name
|
973
|
+
|
974
|
+
@name.setter
|
975
|
+
@_run_decorator._noop_on_finish()
|
976
|
+
def name(self, name: str) -> None:
|
977
|
+
with telemetry.context(run=self) as tel:
|
978
|
+
tel.feature.set_run_name = True
|
979
|
+
self._name = name
|
980
|
+
if self._backend and self._backend.interface:
|
981
|
+
self._backend.interface.publish_run(self)
|
982
|
+
|
983
|
+
@property
|
984
|
+
@_run_decorator._attach
|
985
|
+
def notes(self) -> Optional[str]:
|
986
|
+
"""Notes associated with the run, if there are any.
|
987
|
+
|
988
|
+
Notes can be a multiline string and can also use markdown and latex equations
|
989
|
+
inside `$$`, like `$x + 3$`.
|
990
|
+
"""
|
991
|
+
if self._notes:
|
992
|
+
return self._notes
|
993
|
+
if not self._run_obj:
|
994
|
+
return None
|
995
|
+
return self._run_obj.notes
|
996
|
+
|
997
|
+
@notes.setter
|
998
|
+
@_run_decorator._noop_on_finish()
|
999
|
+
def notes(self, notes: str) -> None:
|
1000
|
+
self._notes = notes
|
1001
|
+
if self._backend and self._backend.interface:
|
1002
|
+
self._backend.interface.publish_run(self)
|
1003
|
+
|
1004
|
+
@property
|
1005
|
+
@_run_decorator._attach
|
1006
|
+
def tags(self) -> Optional[Tuple]:
|
1007
|
+
"""Tags associated with the run, if there are any."""
|
1008
|
+
if self._tags:
|
1009
|
+
return self._tags
|
1010
|
+
if self._run_obj:
|
1011
|
+
return tuple(self._run_obj.tags)
|
1012
|
+
return None
|
1013
|
+
|
1014
|
+
@tags.setter
|
1015
|
+
@_run_decorator._noop_on_finish()
|
1016
|
+
def tags(self, tags: Sequence) -> None:
|
1017
|
+
with telemetry.context(run=self) as tel:
|
1018
|
+
tel.feature.set_run_tags = True
|
1019
|
+
self._tags = tuple(tags)
|
1020
|
+
if self._backend and self._backend.interface:
|
1021
|
+
self._backend.interface.publish_run(self)
|
1022
|
+
|
1023
|
+
@property
|
1024
|
+
@_run_decorator._attach
|
1025
|
+
def id(self) -> str:
|
1026
|
+
"""Identifier for this run."""
|
1027
|
+
if TYPE_CHECKING:
|
1028
|
+
assert self._run_id is not None
|
1029
|
+
return self._run_id
|
1030
|
+
|
1031
|
+
@property
|
1032
|
+
@_run_decorator._attach
|
1033
|
+
def sweep_id(self) -> Optional[str]:
|
1034
|
+
"""ID of the sweep associated with the run, if there is one."""
|
1035
|
+
if not self._run_obj:
|
1036
|
+
return None
|
1037
|
+
return self._run_obj.sweep_id or None
|
1038
|
+
|
1039
|
+
def _get_path(self) -> str:
|
1040
|
+
parts = [
|
1041
|
+
e for e in [self._entity, self._project, self._run_id] if e is not None
|
1042
|
+
]
|
1043
|
+
return "/".join(parts)
|
1044
|
+
|
1045
|
+
@property
|
1046
|
+
@_run_decorator._attach
|
1047
|
+
def path(self) -> str:
|
1048
|
+
"""Path to the run.
|
1049
|
+
|
1050
|
+
Run paths include entity, project, and run ID, in the format
|
1051
|
+
`entity/project/run_id`.
|
1052
|
+
"""
|
1053
|
+
return self._get_path()
|
1054
|
+
|
1055
|
+
def _get_start_time(self) -> float:
|
1056
|
+
return (
|
1057
|
+
self._start_time
|
1058
|
+
if not self._run_obj
|
1059
|
+
else (self._run_obj.start_time.ToMicroseconds() / 1e6)
|
1060
|
+
)
|
1061
|
+
|
1062
|
+
@property
|
1063
|
+
@_run_decorator._attach
|
1064
|
+
def start_time(self) -> float:
|
1065
|
+
"""Unix timestamp (in seconds) of when the run started."""
|
1066
|
+
return self._get_start_time()
|
1067
|
+
|
1068
|
+
def _get_starting_step(self) -> int:
|
1069
|
+
return self._starting_step if not self._run_obj else self._run_obj.starting_step
|
1070
|
+
|
1071
|
+
@property
|
1072
|
+
@_run_decorator._attach
|
1073
|
+
def starting_step(self) -> int:
|
1074
|
+
"""The first step of the run."""
|
1075
|
+
return self._get_starting_step()
|
1076
|
+
|
1077
|
+
@property
|
1078
|
+
@_run_decorator._attach
|
1079
|
+
def resumed(self) -> bool:
|
1080
|
+
"""True if the run was resumed, False otherwise."""
|
1081
|
+
return self._run_obj.resumed if self._run_obj else False
|
1082
|
+
|
1083
|
+
@property
|
1084
|
+
@_run_decorator._attach
|
1085
|
+
def step(self) -> int:
|
1086
|
+
"""Current value of the step.
|
1087
|
+
|
1088
|
+
This counter is incremented by `wandb.log`.
|
1089
|
+
"""
|
1090
|
+
return self._step
|
1091
|
+
|
1092
|
+
def project_name(self) -> str:
|
1093
|
+
return self._run_obj.project if self._run_obj else ""
|
1094
|
+
|
1095
|
+
@property
|
1096
|
+
@_run_decorator._attach
|
1097
|
+
def mode(self) -> str:
|
1098
|
+
"""For compatibility with `0.9.x` and earlier, deprecate eventually."""
|
1099
|
+
if hasattr(self, "_telemetry_obj"):
|
1100
|
+
deprecate.deprecate(
|
1101
|
+
field_name=deprecate.Deprecated.run__mode,
|
1102
|
+
warning_message=(
|
1103
|
+
"The mode property of wandb.run is deprecated "
|
1104
|
+
"and will be removed in a future release."
|
1105
|
+
),
|
1106
|
+
)
|
1107
|
+
return "dryrun" if self._settings._offline else "run"
|
1108
|
+
|
1109
|
+
@property
|
1110
|
+
@_run_decorator._attach
|
1111
|
+
def offline(self) -> bool:
|
1112
|
+
return self._settings._offline
|
1113
|
+
|
1114
|
+
@property
|
1115
|
+
@_run_decorator._attach
|
1116
|
+
def disabled(self) -> bool:
|
1117
|
+
return self._settings._noop
|
1118
|
+
|
1119
|
+
def _get_group(self) -> str:
|
1120
|
+
return self._run_obj.run_group if self._run_obj else ""
|
1121
|
+
|
1122
|
+
@property
|
1123
|
+
@_run_decorator._attach
|
1124
|
+
def group(self) -> str:
|
1125
|
+
"""Name of the group associated with the run.
|
1126
|
+
|
1127
|
+
Setting a group helps the W&B UI organize runs in a sensible way.
|
1128
|
+
|
1129
|
+
If you are doing a distributed training you should give all of the
|
1130
|
+
runs in the training the same group.
|
1131
|
+
If you are doing cross-validation you should give all the cross-validation
|
1132
|
+
folds the same group.
|
1133
|
+
"""
|
1134
|
+
return self._get_group()
|
1135
|
+
|
1136
|
+
@property
|
1137
|
+
@_run_decorator._attach
|
1138
|
+
def job_type(self) -> str:
|
1139
|
+
return self._run_obj.job_type if self._run_obj else ""
|
1140
|
+
|
1141
|
+
@property
|
1142
|
+
@_run_decorator._attach
|
1143
|
+
def project(self) -> str:
|
1144
|
+
"""Name of the W&B project associated with the run."""
|
1145
|
+
return self.project_name()
|
1146
|
+
|
1147
|
+
@_run_decorator._noop_on_finish()
|
1148
|
+
@_run_decorator._attach
|
1149
|
+
def log_code(
|
1150
|
+
self,
|
1151
|
+
root: Optional[str] = ".",
|
1152
|
+
name: Optional[str] = None,
|
1153
|
+
include_fn: Union[
|
1154
|
+
Callable[[str, str], bool], Callable[[str], bool]
|
1155
|
+
] = _is_py_requirements_or_dockerfile,
|
1156
|
+
exclude_fn: Union[
|
1157
|
+
Callable[[str, str], bool], Callable[[str], bool]
|
1158
|
+
] = filenames.exclude_wandb_fn,
|
1159
|
+
) -> Optional[Artifact]:
|
1160
|
+
"""Save the current state of your code to a W&B Artifact.
|
1161
|
+
|
1162
|
+
By default, it walks the current directory and logs all files that end with `.py`.
|
1163
|
+
|
1164
|
+
Arguments:
|
1165
|
+
root: The relative (to `os.getcwd()`) or absolute path to recursively find code from.
|
1166
|
+
name: (str, optional) The name of our code artifact. By default, we'll name
|
1167
|
+
the artifact `source-$PROJECT_ID-$ENTRYPOINT_RELPATH`. There may be scenarios where you want
|
1168
|
+
many runs to share the same artifact. Specifying name allows you to achieve that.
|
1169
|
+
include_fn: A callable that accepts a file path and (optionally) root path and
|
1170
|
+
returns True when it should be included and False otherwise. This
|
1171
|
+
defaults to: `lambda path, root: path.endswith(".py")`
|
1172
|
+
exclude_fn: A callable that accepts a file path and (optionally) root path and
|
1173
|
+
returns `True` when it should be excluded and `False` otherwise. This
|
1174
|
+
defaults to a function that excludes all files within `<root>/.wandb/`
|
1175
|
+
and `<root>/wandb/` directories.
|
1176
|
+
|
1177
|
+
Examples:
|
1178
|
+
Basic usage
|
1179
|
+
```python
|
1180
|
+
run.log_code()
|
1181
|
+
```
|
1182
|
+
|
1183
|
+
Advanced usage
|
1184
|
+
```python
|
1185
|
+
run.log_code(
|
1186
|
+
"../",
|
1187
|
+
include_fn=lambda path: path.endswith(".py") or path.endswith(".ipynb"),
|
1188
|
+
exclude_fn=lambda path, root: os.path.relpath(path, root).startswith("cache/"),
|
1189
|
+
)
|
1190
|
+
```
|
1191
|
+
|
1192
|
+
Returns:
|
1193
|
+
An `Artifact` object if code was logged
|
1194
|
+
"""
|
1195
|
+
if name is None:
|
1196
|
+
if self.settings._jupyter:
|
1197
|
+
notebook_name = None
|
1198
|
+
if self.settings.notebook_name:
|
1199
|
+
notebook_name = self.settings.notebook_name
|
1200
|
+
elif self.settings._jupyter_path:
|
1201
|
+
if self.settings._jupyter_path.startswith("fileId="):
|
1202
|
+
notebook_name = self.settings._jupyter_name
|
1203
|
+
else:
|
1204
|
+
notebook_name = self.settings._jupyter_path
|
1205
|
+
name_string = f"{self._project}-{notebook_name}"
|
1206
|
+
else:
|
1207
|
+
name_string = f"{self._project}-{self._settings.program_relpath}"
|
1208
|
+
name = wandb.util.make_artifact_name_safe(f"source-{name_string}")
|
1209
|
+
art = wandb.Artifact(name, "code")
|
1210
|
+
files_added = False
|
1211
|
+
if root is not None:
|
1212
|
+
root = os.path.abspath(root)
|
1213
|
+
for file_path in filenames.filtered_dir(root, include_fn, exclude_fn):
|
1214
|
+
files_added = True
|
1215
|
+
save_name = os.path.relpath(file_path, root)
|
1216
|
+
art.add_file(file_path, name=save_name)
|
1217
|
+
# Add any manually staged files such as ipynb notebooks
|
1218
|
+
for dirpath, _, files in os.walk(self._settings._tmp_code_dir):
|
1219
|
+
for fname in files:
|
1220
|
+
file_path = os.path.join(dirpath, fname)
|
1221
|
+
save_name = os.path.relpath(file_path, self._settings._tmp_code_dir)
|
1222
|
+
files_added = True
|
1223
|
+
art.add_file(file_path, name=save_name)
|
1224
|
+
if not files_added:
|
1225
|
+
wandb.termwarn(
|
1226
|
+
"No relevant files were detected in the specified directory. No code will be logged to your run."
|
1227
|
+
)
|
1228
|
+
return None
|
1229
|
+
|
1230
|
+
return self._log_artifact(art)
|
1231
|
+
|
1232
|
+
def get_url(self) -> Optional[str]:
|
1233
|
+
"""Return the url for the W&B run, if there is one.
|
1234
|
+
|
1235
|
+
Offline runs will not have a url.
|
1236
|
+
"""
|
1237
|
+
if self._settings._offline:
|
1238
|
+
wandb.termwarn("URL not available in offline run")
|
1239
|
+
return None
|
1240
|
+
return self._settings.run_url
|
1241
|
+
|
1242
|
+
def get_project_url(self) -> Optional[str]:
|
1243
|
+
"""Return the url for the W&B project associated with the run, if there is one.
|
1244
|
+
|
1245
|
+
Offline runs will not have a project url.
|
1246
|
+
"""
|
1247
|
+
if self._settings._offline:
|
1248
|
+
wandb.termwarn("URL not available in offline run")
|
1249
|
+
return None
|
1250
|
+
return self._settings.project_url
|
1251
|
+
|
1252
|
+
def get_sweep_url(self) -> Optional[str]:
|
1253
|
+
"""Return the url for the sweep associated with the run, if there is one."""
|
1254
|
+
if self._settings._offline:
|
1255
|
+
wandb.termwarn("URL not available in offline run")
|
1256
|
+
return None
|
1257
|
+
return self._settings.sweep_url
|
1258
|
+
|
1259
|
+
@property
|
1260
|
+
@_run_decorator._attach
|
1261
|
+
def url(self) -> Optional[str]:
|
1262
|
+
"""The W&B url associated with the run."""
|
1263
|
+
return self.get_url()
|
1264
|
+
|
1265
|
+
@property
|
1266
|
+
@_run_decorator._attach
|
1267
|
+
def entity(self) -> str:
|
1268
|
+
"""The name of the W&B entity associated with the run.
|
1269
|
+
|
1270
|
+
Entity can be a username or the name of a team or organization.
|
1271
|
+
"""
|
1272
|
+
return self._entity or ""
|
1273
|
+
|
1274
|
+
def _label_internal(
|
1275
|
+
self,
|
1276
|
+
code: Optional[str] = None,
|
1277
|
+
repo: Optional[str] = None,
|
1278
|
+
code_version: Optional[str] = None,
|
1279
|
+
) -> None:
|
1280
|
+
with telemetry.context(run=self) as tel:
|
1281
|
+
if code and RE_LABEL.match(code):
|
1282
|
+
tel.label.code_string = code
|
1283
|
+
if repo and RE_LABEL.match(repo):
|
1284
|
+
tel.label.repo_string = repo
|
1285
|
+
if code_version and RE_LABEL.match(code_version):
|
1286
|
+
tel.label.code_version = code_version
|
1287
|
+
|
1288
|
+
def _label(
|
1289
|
+
self,
|
1290
|
+
code: Optional[str] = None,
|
1291
|
+
repo: Optional[str] = None,
|
1292
|
+
code_version: Optional[str] = None,
|
1293
|
+
**kwargs: str,
|
1294
|
+
) -> None:
|
1295
|
+
if self._settings.label_disable:
|
1296
|
+
return
|
1297
|
+
for k, v in (("code", code), ("repo", repo), ("code_version", code_version)):
|
1298
|
+
if v and not RE_LABEL.match(v):
|
1299
|
+
wandb.termwarn(
|
1300
|
+
f"Label added for '{k}' with invalid identifier '{v}' (ignored).",
|
1301
|
+
repeat=False,
|
1302
|
+
)
|
1303
|
+
for v in kwargs:
|
1304
|
+
wandb.termwarn(
|
1305
|
+
f"Label added for unsupported key {v!r} (ignored).",
|
1306
|
+
repeat=False,
|
1307
|
+
)
|
1308
|
+
|
1309
|
+
self._label_internal(code=code, repo=repo, code_version=code_version)
|
1310
|
+
|
1311
|
+
# update telemetry in the backend immediately for _label() callers
|
1312
|
+
self._telemetry_flush()
|
1313
|
+
|
1314
|
+
def _label_probe_lines(self, lines: List[str]) -> None:
|
1315
|
+
if not lines:
|
1316
|
+
return
|
1317
|
+
parsed = telemetry._parse_label_lines(lines)
|
1318
|
+
if not parsed:
|
1319
|
+
return
|
1320
|
+
label_dict = {}
|
1321
|
+
code = parsed.get("code") or parsed.get("c")
|
1322
|
+
if code:
|
1323
|
+
label_dict["code"] = code
|
1324
|
+
repo = parsed.get("repo") or parsed.get("r")
|
1325
|
+
if repo:
|
1326
|
+
label_dict["repo"] = repo
|
1327
|
+
code_ver = parsed.get("version") or parsed.get("v")
|
1328
|
+
if code_ver:
|
1329
|
+
label_dict["code_version"] = code_ver
|
1330
|
+
self._label_internal(**label_dict)
|
1331
|
+
|
1332
|
+
def _label_probe_main(self) -> None:
|
1333
|
+
m = sys.modules.get("__main__")
|
1334
|
+
if not m:
|
1335
|
+
return
|
1336
|
+
doc = getattr(m, "__doc__", None)
|
1337
|
+
if not doc:
|
1338
|
+
return
|
1339
|
+
|
1340
|
+
doclines = doc.splitlines()
|
1341
|
+
self._label_probe_lines(doclines)
|
1342
|
+
|
1343
|
+
# TODO: annotate jupyter Notebook class
|
1344
|
+
def _label_probe_notebook(self, notebook: Any) -> None:
|
1345
|
+
logger.info("probe notebook")
|
1346
|
+
lines = None
|
1347
|
+
try:
|
1348
|
+
data = notebook.probe_ipynb()
|
1349
|
+
cell0 = data.get("cells", [])[0]
|
1350
|
+
lines = cell0.get("source")
|
1351
|
+
# kaggle returns a string instead of a list
|
1352
|
+
if isinstance(lines, str):
|
1353
|
+
lines = lines.split()
|
1354
|
+
except Exception as e:
|
1355
|
+
logger.info(f"Unable to probe notebook: {e}")
|
1356
|
+
return
|
1357
|
+
if lines:
|
1358
|
+
self._label_probe_lines(lines)
|
1359
|
+
|
1360
|
+
@_run_decorator._attach
|
1361
|
+
def display(self, height: int = 420, hidden: bool = False) -> bool:
|
1362
|
+
"""Display this run in jupyter."""
|
1363
|
+
if self._settings._jupyter:
|
1364
|
+
ipython.display_html(self.to_html(height, hidden))
|
1365
|
+
return True
|
1366
|
+
else:
|
1367
|
+
wandb.termwarn(".display() only works in jupyter environments")
|
1368
|
+
return False
|
1369
|
+
|
1370
|
+
@_run_decorator._attach
|
1371
|
+
def to_html(self, height: int = 420, hidden: bool = False) -> str:
|
1372
|
+
"""Generate HTML containing an iframe displaying the current run."""
|
1373
|
+
url = self._settings.run_url + "?jupyter=true"
|
1374
|
+
style = f"border:none;width:100%;height:{height}px;"
|
1375
|
+
prefix = ""
|
1376
|
+
if hidden:
|
1377
|
+
style += "display:none;"
|
1378
|
+
prefix = ipython.toggle_button()
|
1379
|
+
return prefix + f"<iframe src={url!r} style={style!r}></iframe>"
|
1380
|
+
|
1381
|
+
def _repr_mimebundle_(
|
1382
|
+
self, include: Optional[Any] = None, exclude: Optional[Any] = None
|
1383
|
+
) -> Dict[str, str]:
|
1384
|
+
return {"text/html": self.to_html(hidden=True)}
|
1385
|
+
|
1386
|
+
@_run_decorator._noop_on_finish()
|
1387
|
+
def _config_callback(
|
1388
|
+
self,
|
1389
|
+
key: Optional[Union[Tuple[str, ...], str]] = None,
|
1390
|
+
val: Optional[Any] = None,
|
1391
|
+
data: Optional[Dict[str, object]] = None,
|
1392
|
+
) -> None:
|
1393
|
+
logger.info(f"config_cb {key} {val} {data}")
|
1394
|
+
if self._backend and self._backend.interface:
|
1395
|
+
self._backend.interface.publish_config(key=key, val=val, data=data)
|
1396
|
+
|
1397
|
+
def _config_artifact_callback(
|
1398
|
+
self, key: str, val: Union[str, Artifact, dict]
|
1399
|
+
) -> Artifact:
|
1400
|
+
# artifacts can look like dicts as they are passed into the run config
|
1401
|
+
# since the run config stores them on the backend as a dict with fields shown
|
1402
|
+
# in wandb.util.artifact_to_json
|
1403
|
+
if _is_artifact_version_weave_dict(val):
|
1404
|
+
assert isinstance(val, dict)
|
1405
|
+
public_api = self._public_api()
|
1406
|
+
artifact = Artifact._from_id(val["id"], public_api.client)
|
1407
|
+
return self.use_artifact(artifact, use_as=key)
|
1408
|
+
elif _is_artifact_string(val):
|
1409
|
+
# this will never fail, but is required to make mypy happy
|
1410
|
+
assert isinstance(val, str)
|
1411
|
+
artifact_string, base_url, is_id = parse_artifact_string(val)
|
1412
|
+
overrides = {}
|
1413
|
+
if base_url is not None:
|
1414
|
+
overrides = {"base_url": base_url}
|
1415
|
+
public_api = public.Api(overrides)
|
1416
|
+
else:
|
1417
|
+
public_api = self._public_api()
|
1418
|
+
if is_id:
|
1419
|
+
artifact = Artifact._from_id(artifact_string, public_api._client)
|
1420
|
+
else:
|
1421
|
+
artifact = public_api.artifact(name=artifact_string)
|
1422
|
+
# in the future we'll need to support using artifacts from
|
1423
|
+
# different instances of wandb.
|
1424
|
+
|
1425
|
+
return self.use_artifact(artifact, use_as=key)
|
1426
|
+
elif _is_artifact_object(val):
|
1427
|
+
return self.use_artifact(val, use_as=key)
|
1428
|
+
else:
|
1429
|
+
raise ValueError(
|
1430
|
+
f"Cannot call _config_artifact_callback on type {type(val)}"
|
1431
|
+
)
|
1432
|
+
|
1433
|
+
def _set_config_wandb(self, key: str, val: Any) -> None:
|
1434
|
+
self._config_callback(key=("_wandb", key), val=val)
|
1435
|
+
|
1436
|
+
@_run_decorator._noop_on_finish()
|
1437
|
+
def _summary_update_callback(self, summary_record: SummaryRecord) -> None:
|
1438
|
+
with telemetry.context(run=self) as tel:
|
1439
|
+
tel.feature.set_summary = True
|
1440
|
+
if self._backend and self._backend.interface:
|
1441
|
+
self._backend.interface.publish_summary(summary_record)
|
1442
|
+
|
1443
|
+
def _on_progress_get_summary(self, handle: MailboxProgress) -> None:
|
1444
|
+
pass
|
1445
|
+
# TODO(jhr): enable printing for get_summary in later mailbox dev phase
|
1446
|
+
# line = "Waiting for run.summary data..."
|
1447
|
+
# self._printer.display(line)
|
1448
|
+
|
1449
|
+
def _summary_get_current_summary_callback(self) -> Dict[str, Any]:
|
1450
|
+
if self._is_finished:
|
1451
|
+
# TODO: WB-18420: fetch summary from backend and stage it before run is finished
|
1452
|
+
wandb.termwarn("Summary data not available in finished run")
|
1453
|
+
return {}
|
1454
|
+
if not self._backend or not self._backend.interface:
|
1455
|
+
return {}
|
1456
|
+
handle = self._backend.interface.deliver_get_summary()
|
1457
|
+
result = handle.wait(
|
1458
|
+
timeout=self._settings.summary_timeout,
|
1459
|
+
on_progress=self._on_progress_get_summary,
|
1460
|
+
)
|
1461
|
+
if not result:
|
1462
|
+
return {}
|
1463
|
+
get_summary_response = result.response.get_summary_response
|
1464
|
+
return proto_util.dict_from_proto_list(get_summary_response.item)
|
1465
|
+
|
1466
|
+
def _metric_callback(self, metric_record: MetricRecord) -> None:
|
1467
|
+
if self._backend and self._backend.interface:
|
1468
|
+
self._backend.interface._publish_metric(metric_record)
|
1469
|
+
|
1470
|
+
def _datatypes_callback(self, fname: str) -> None:
|
1471
|
+
if not self._backend or not self._backend.interface:
|
1472
|
+
return
|
1473
|
+
files: FilesDict = dict(files=[(GlobStr(glob.escape(fname)), "now")])
|
1474
|
+
self._backend.interface.publish_files(files)
|
1475
|
+
|
1476
|
+
def _visualization_hack(self, row: Dict[str, Any]) -> Dict[str, Any]:
|
1477
|
+
# TODO(jhr): move visualize hack somewhere else
|
1478
|
+
chart_keys = set()
|
1479
|
+
split_table_set = set()
|
1480
|
+
for k in row:
|
1481
|
+
if isinstance(row[k], Visualize):
|
1482
|
+
key = row[k].get_config_key(k)
|
1483
|
+
value = row[k].get_config_value(k)
|
1484
|
+
row[k] = row[k]._data
|
1485
|
+
self._config_callback(val=value, key=key)
|
1486
|
+
elif isinstance(row[k], CustomChart):
|
1487
|
+
chart_keys.add(k)
|
1488
|
+
key = row[k].get_config_key(k)
|
1489
|
+
if row[k]._split_table:
|
1490
|
+
value = row[k].get_config_value(
|
1491
|
+
"Vega2", row[k].user_query(f"Custom Chart Tables/{k}_table")
|
1492
|
+
)
|
1493
|
+
split_table_set.add(k)
|
1494
|
+
else:
|
1495
|
+
value = row[k].get_config_value(
|
1496
|
+
"Vega2", row[k].user_query(f"{k}_table")
|
1497
|
+
)
|
1498
|
+
row[k] = row[k]._data
|
1499
|
+
self._config_callback(val=value, key=key)
|
1500
|
+
|
1501
|
+
for k in chart_keys:
|
1502
|
+
# remove the chart key from the row
|
1503
|
+
# TODO: is this really the right move? what if the user logs
|
1504
|
+
# a non-custom chart to this key?
|
1505
|
+
if k in split_table_set:
|
1506
|
+
row[f"Custom Chart Tables/{k}_table"] = row.pop(k)
|
1507
|
+
else:
|
1508
|
+
row[f"{k}_table"] = row.pop(k)
|
1509
|
+
return row
|
1510
|
+
|
1511
|
+
def _partial_history_callback(
|
1512
|
+
self,
|
1513
|
+
row: Dict[str, Any],
|
1514
|
+
step: Optional[int] = None,
|
1515
|
+
commit: Optional[bool] = None,
|
1516
|
+
) -> None:
|
1517
|
+
row = row.copy()
|
1518
|
+
if row:
|
1519
|
+
row = self._visualization_hack(row)
|
1520
|
+
|
1521
|
+
if self._backend and self._backend.interface:
|
1522
|
+
not_using_tensorboard = len(wandb.patched["tensorboard"]) == 0
|
1523
|
+
|
1524
|
+
self._backend.interface.publish_partial_history(
|
1525
|
+
row,
|
1526
|
+
user_step=self._step,
|
1527
|
+
step=step,
|
1528
|
+
flush=commit,
|
1529
|
+
publish_step=not_using_tensorboard,
|
1530
|
+
)
|
1531
|
+
|
1532
|
+
def _console_callback(self, name: str, data: str) -> None:
|
1533
|
+
# logger.info("console callback: %s, %s", name, data)
|
1534
|
+
if self._backend and self._backend.interface:
|
1535
|
+
self._backend.interface.publish_output(name, data)
|
1536
|
+
|
1537
|
+
@_run_decorator._noop_on_finish(only_warn=True)
|
1538
|
+
def _console_raw_callback(self, name: str, data: str) -> None:
|
1539
|
+
# logger.info("console callback: %s, %s", name, data)
|
1540
|
+
|
1541
|
+
# NOTE: console output is only allowed on the process which installed the callback
|
1542
|
+
# this will prevent potential corruption in the socket to the service. Other methods
|
1543
|
+
# are protected by the _attach run decorator, but this callback was installed on the
|
1544
|
+
# write function of stdout and stderr streams.
|
1545
|
+
console_pid = getattr(self, "_attach_pid", 0)
|
1546
|
+
if console_pid != os.getpid():
|
1547
|
+
return
|
1548
|
+
|
1549
|
+
if self._backend and self._backend.interface:
|
1550
|
+
self._backend.interface.publish_output_raw(name, data)
|
1551
|
+
|
1552
|
+
def _tensorboard_callback(
|
1553
|
+
self, logdir: str, save: bool = True, root_logdir: str = ""
|
1554
|
+
) -> None:
|
1555
|
+
logger.info("tensorboard callback: %s, %s", logdir, save)
|
1556
|
+
if self._backend and self._backend.interface:
|
1557
|
+
self._backend.interface.publish_tbdata(logdir, save, root_logdir)
|
1558
|
+
|
1559
|
+
def _set_library(self, library: _WandbSetup) -> None:
|
1560
|
+
self._wl = library
|
1561
|
+
|
1562
|
+
def _set_backend(self, backend: "wandb.sdk.backend.backend.Backend") -> None:
|
1563
|
+
self._backend = backend
|
1564
|
+
|
1565
|
+
def _set_internal_run_interface(
|
1566
|
+
self,
|
1567
|
+
interface: "wandb.sdk.interface.interface_queue.InterfaceQueue",
|
1568
|
+
) -> None:
|
1569
|
+
self._internal_run_interface = interface
|
1570
|
+
|
1571
|
+
def _set_reporter(self, reporter: Reporter) -> None:
|
1572
|
+
self._reporter = reporter
|
1573
|
+
|
1574
|
+
def _set_teardown_hooks(self, hooks: List[TeardownHook]) -> None:
|
1575
|
+
self._teardown_hooks = hooks
|
1576
|
+
|
1577
|
+
def _set_run_obj(self, run_obj: RunRecord) -> None:
|
1578
|
+
self._run_obj = run_obj
|
1579
|
+
if self.settings._offline:
|
1580
|
+
return
|
1581
|
+
|
1582
|
+
self._entity = run_obj.entity
|
1583
|
+
self._project = run_obj.project
|
1584
|
+
|
1585
|
+
# Grab the config from resuming
|
1586
|
+
if run_obj.config:
|
1587
|
+
c_dict = config_util.dict_no_value_from_proto_list(run_obj.config.update)
|
1588
|
+
# TODO: Windows throws a wild error when this is set...
|
1589
|
+
if "_wandb" in c_dict:
|
1590
|
+
del c_dict["_wandb"]
|
1591
|
+
# We update the config object here without triggering the callback
|
1592
|
+
self._config._update(c_dict, allow_val_change=True, ignore_locked=True)
|
1593
|
+
# Update the summary, this will trigger an un-needed graphql request :(
|
1594
|
+
if run_obj.summary:
|
1595
|
+
summary_dict = {}
|
1596
|
+
for orig in run_obj.summary.update:
|
1597
|
+
summary_dict[orig.key] = json.loads(orig.value_json)
|
1598
|
+
if summary_dict:
|
1599
|
+
self.summary.update(summary_dict)
|
1600
|
+
self._step = self._get_starting_step()
|
1601
|
+
|
1602
|
+
# update settings from run_obj
|
1603
|
+
self._settings._apply_run_start(message_to_dict(self._run_obj))
|
1604
|
+
self._update_settings(self._settings)
|
1605
|
+
|
1606
|
+
wandb._sentry.configure_scope(
|
1607
|
+
process_context="user",
|
1608
|
+
tags=dict(self._settings),
|
1609
|
+
)
|
1610
|
+
|
1611
|
+
def _add_singleton(
|
1612
|
+
self, data_type: str, key: str, value: Dict[Union[int, str], str]
|
1613
|
+
) -> None:
|
1614
|
+
"""Store a singleton item to wandb config.
|
1615
|
+
|
1616
|
+
A singleton in this context is a piece of data that is continually
|
1617
|
+
logged with the same value in each history step, but represented
|
1618
|
+
as a single item in the config.
|
1619
|
+
|
1620
|
+
We do this to avoid filling up history with a lot of repeated unnecessary data
|
1621
|
+
|
1622
|
+
Add singleton can be called many times in one run, and it will only be
|
1623
|
+
updated when the value changes. The last value logged will be the one
|
1624
|
+
persisted to the server.
|
1625
|
+
"""
|
1626
|
+
value_extra = {"type": data_type, "key": key, "value": value}
|
1627
|
+
|
1628
|
+
if data_type not in self._config["_wandb"]:
|
1629
|
+
self._config["_wandb"][data_type] = {}
|
1630
|
+
|
1631
|
+
if data_type in self._config["_wandb"][data_type]:
|
1632
|
+
old_value = self._config["_wandb"][data_type][key]
|
1633
|
+
else:
|
1634
|
+
old_value = None
|
1635
|
+
|
1636
|
+
if value_extra != old_value:
|
1637
|
+
self._config["_wandb"][data_type][key] = value_extra
|
1638
|
+
self._config.persist()
|
1639
|
+
|
1640
|
+
def _log(
|
1641
|
+
self,
|
1642
|
+
data: Dict[str, Any],
|
1643
|
+
step: Optional[int] = None,
|
1644
|
+
commit: Optional[bool] = None,
|
1645
|
+
) -> None:
|
1646
|
+
if not isinstance(data, Mapping):
|
1647
|
+
raise ValueError("wandb.log must be passed a dictionary")
|
1648
|
+
|
1649
|
+
if any(not isinstance(key, str) for key in data.keys()):
|
1650
|
+
raise ValueError("Key values passed to `wandb.log` must be strings.")
|
1651
|
+
|
1652
|
+
self._partial_history_callback(data, step, commit)
|
1653
|
+
|
1654
|
+
if step is not None:
|
1655
|
+
if os.getpid() != self._init_pid or self._is_attached:
|
1656
|
+
wandb.termwarn(
|
1657
|
+
"Note that setting step in multiprocessing can result in data loss. "
|
1658
|
+
"Please use `run.define_metric(...)` to define a custom metric "
|
1659
|
+
"to log your step values.",
|
1660
|
+
repeat=False,
|
1661
|
+
)
|
1662
|
+
# if step is passed in when tensorboard_sync is used we honor the step passed
|
1663
|
+
# to make decisions about how to close out the history record, but will strip
|
1664
|
+
# this history later on in publish_history()
|
1665
|
+
if len(wandb.patched["tensorboard"]) > 0:
|
1666
|
+
wandb.termwarn(
|
1667
|
+
"Step cannot be set when using tensorboard syncing. "
|
1668
|
+
"Please use `run.define_metric(...)` to define a custom metric "
|
1669
|
+
"to log your step values.",
|
1670
|
+
repeat=False,
|
1671
|
+
)
|
1672
|
+
if step > self._step:
|
1673
|
+
self._step = step
|
1674
|
+
|
1675
|
+
if (step is None and commit is None) or commit:
|
1676
|
+
self._step += 1
|
1677
|
+
|
1678
|
+
@_run_decorator._noop
|
1679
|
+
@_run_decorator._noop_on_finish()
|
1680
|
+
@_run_decorator._attach
|
1681
|
+
def log(
|
1682
|
+
self,
|
1683
|
+
data: Dict[str, Any],
|
1684
|
+
step: Optional[int] = None,
|
1685
|
+
commit: Optional[bool] = None,
|
1686
|
+
sync: Optional[bool] = None,
|
1687
|
+
) -> None:
|
1688
|
+
"""Upload run data.
|
1689
|
+
|
1690
|
+
Use `log` to log data from runs, such as scalars, images, video,
|
1691
|
+
histograms, plots, and tables.
|
1692
|
+
|
1693
|
+
See our [guides to logging](https://docs.wandb.ai/guides/track/log) for
|
1694
|
+
live examples, code snippets, best practices, and more.
|
1695
|
+
|
1696
|
+
The most basic usage is `run.log({"train-loss": 0.5, "accuracy": 0.9})`.
|
1697
|
+
This will save the loss and accuracy to the run's history and update
|
1698
|
+
the summary values for these metrics.
|
1699
|
+
|
1700
|
+
Visualize logged data in the workspace at [wandb.ai](https://wandb.ai),
|
1701
|
+
or locally on a [self-hosted instance](https://docs.wandb.ai/guides/hosting)
|
1702
|
+
of the W&B app, or export data to visualize and explore locally, e.g. in
|
1703
|
+
Jupyter notebooks, with [our API](https://docs.wandb.ai/guides/track/public-api-guide).
|
1704
|
+
|
1705
|
+
Logged values don't have to be scalars. Logging any wandb object is supported.
|
1706
|
+
For example `run.log({"example": wandb.Image("myimage.jpg")})` will log an
|
1707
|
+
example image which will be displayed nicely in the W&B UI.
|
1708
|
+
See the [reference documentation](https://docs.wandb.com/ref/python/data-types)
|
1709
|
+
for all of the different supported types or check out our
|
1710
|
+
[guides to logging](https://docs.wandb.ai/guides/track/log) for examples,
|
1711
|
+
from 3D molecular structures and segmentation masks to PR curves and histograms.
|
1712
|
+
You can use `wandb.Table` to log structured data. See our
|
1713
|
+
[guide to logging tables](https://docs.wandb.ai/guides/data-vis/log-tables)
|
1714
|
+
for details.
|
1715
|
+
|
1716
|
+
The W&B UI organizes metrics with a forward slash (`/`) in their name
|
1717
|
+
into sections named using the text before the final slash. For example,
|
1718
|
+
the following results in two sections named "train" and "validate":
|
1719
|
+
|
1720
|
+
```
|
1721
|
+
run.log({
|
1722
|
+
"train/accuracy": 0.9,
|
1723
|
+
"train/loss": 30,
|
1724
|
+
"validate/accuracy": 0.8,
|
1725
|
+
"validate/loss": 20,
|
1726
|
+
})
|
1727
|
+
```
|
1728
|
+
|
1729
|
+
Only one level of nesting is supported; `run.log({"a/b/c": 1})`
|
1730
|
+
produces a section named "a/b".
|
1731
|
+
|
1732
|
+
`run.log` is not intended to be called more than a few times per second.
|
1733
|
+
For optimal performance, limit your logging to once every N iterations,
|
1734
|
+
or collect data over multiple iterations and log it in a single step.
|
1735
|
+
|
1736
|
+
### The W&B step
|
1737
|
+
|
1738
|
+
With basic usage, each call to `log` creates a new "step".
|
1739
|
+
The step must always increase, and it is not possible to log
|
1740
|
+
to a previous step.
|
1741
|
+
|
1742
|
+
Note that you can use any metric as the X axis in charts.
|
1743
|
+
In many cases, it is better to treat the W&B step like
|
1744
|
+
you'd treat a timestamp rather than a training step.
|
1745
|
+
|
1746
|
+
```
|
1747
|
+
# Example: log an "epoch" metric for use as an X axis.
|
1748
|
+
run.log({"epoch": 40, "train-loss": 0.5})
|
1749
|
+
```
|
1750
|
+
|
1751
|
+
See also [define_metric](https://docs.wandb.ai/ref/python/run#define_metric).
|
1752
|
+
|
1753
|
+
It is possible to use multiple `log` invocations to log to
|
1754
|
+
the same step with the `step` and `commit` parameters.
|
1755
|
+
The following are all equivalent:
|
1756
|
+
|
1757
|
+
```
|
1758
|
+
# Normal usage:
|
1759
|
+
run.log({"train-loss": 0.5, "accuracy": 0.8})
|
1760
|
+
run.log({"train-loss": 0.4, "accuracy": 0.9})
|
1761
|
+
|
1762
|
+
# Implicit step without auto-incrementing:
|
1763
|
+
run.log({"train-loss": 0.5}, commit=False)
|
1764
|
+
run.log({"accuracy": 0.8})
|
1765
|
+
run.log({"train-loss": 0.4}, commit=False)
|
1766
|
+
run.log({"accuracy": 0.9})
|
1767
|
+
|
1768
|
+
# Explicit step:
|
1769
|
+
run.log({"train-loss": 0.5}, step=current_step)
|
1770
|
+
run.log({"accuracy": 0.8}, step=current_step)
|
1771
|
+
current_step += 1
|
1772
|
+
run.log({"train-loss": 0.4}, step=current_step)
|
1773
|
+
run.log({"accuracy": 0.9}, step=current_step)
|
1774
|
+
```
|
1775
|
+
|
1776
|
+
Arguments:
|
1777
|
+
data: A `dict` with `str` keys and values that are serializable
|
1778
|
+
Python objects including: `int`, `float` and `string`;
|
1779
|
+
any of the `wandb.data_types`; lists, tuples and NumPy arrays
|
1780
|
+
of serializable Python objects; other `dict`s of this
|
1781
|
+
structure.
|
1782
|
+
step: The step number to log. If `None`, then an implicit
|
1783
|
+
auto-incrementing step is used. See the notes in
|
1784
|
+
the description.
|
1785
|
+
commit: If true, finalize and upload the step. If false, then
|
1786
|
+
accumulate data for the step. See the notes in the description.
|
1787
|
+
If `step` is `None`, then the default is `commit=True`;
|
1788
|
+
otherwise, the default is `commit=False`.
|
1789
|
+
sync: This argument is deprecated and does nothing.
|
1790
|
+
|
1791
|
+
Examples:
|
1792
|
+
For more and more detailed examples, see
|
1793
|
+
[our guides to logging](https://docs.wandb.com/guides/track/log).
|
1794
|
+
|
1795
|
+
### Basic usage
|
1796
|
+
<!--yeadoc-test:init-and-log-basic-->
|
1797
|
+
```python
|
1798
|
+
import wandb
|
1799
|
+
|
1800
|
+
run = wandb.init()
|
1801
|
+
run.log({"accuracy": 0.9, "epoch": 5})
|
1802
|
+
```
|
1803
|
+
|
1804
|
+
### Incremental logging
|
1805
|
+
<!--yeadoc-test:init-and-log-incremental-->
|
1806
|
+
```python
|
1807
|
+
import wandb
|
1808
|
+
|
1809
|
+
run = wandb.init()
|
1810
|
+
run.log({"loss": 0.2}, commit=False)
|
1811
|
+
# Somewhere else when I'm ready to report this step:
|
1812
|
+
run.log({"accuracy": 0.8})
|
1813
|
+
```
|
1814
|
+
|
1815
|
+
### Histogram
|
1816
|
+
<!--yeadoc-test:init-and-log-histogram-->
|
1817
|
+
```python
|
1818
|
+
import numpy as np
|
1819
|
+
import wandb
|
1820
|
+
|
1821
|
+
# sample gradients at random from normal distribution
|
1822
|
+
gradients = np.random.randn(100, 100)
|
1823
|
+
run = wandb.init()
|
1824
|
+
run.log({"gradients": wandb.Histogram(gradients)})
|
1825
|
+
```
|
1826
|
+
|
1827
|
+
### Image from numpy
|
1828
|
+
<!--yeadoc-test:init-and-log-image-numpy-->
|
1829
|
+
```python
|
1830
|
+
import numpy as np
|
1831
|
+
import wandb
|
1832
|
+
|
1833
|
+
run = wandb.init()
|
1834
|
+
examples = []
|
1835
|
+
for i in range(3):
|
1836
|
+
pixels = np.random.randint(low=0, high=256, size=(100, 100, 3))
|
1837
|
+
image = wandb.Image(pixels, caption=f"random field {i}")
|
1838
|
+
examples.append(image)
|
1839
|
+
run.log({"examples": examples})
|
1840
|
+
```
|
1841
|
+
|
1842
|
+
### Image from PIL
|
1843
|
+
<!--yeadoc-test:init-and-log-image-pillow-->
|
1844
|
+
```python
|
1845
|
+
import numpy as np
|
1846
|
+
from PIL import Image as PILImage
|
1847
|
+
import wandb
|
1848
|
+
|
1849
|
+
run = wandb.init()
|
1850
|
+
examples = []
|
1851
|
+
for i in range(3):
|
1852
|
+
pixels = np.random.randint(low=0, high=256, size=(100, 100, 3), dtype=np.uint8)
|
1853
|
+
pil_image = PILImage.fromarray(pixels, mode="RGB")
|
1854
|
+
image = wandb.Image(pil_image, caption=f"random field {i}")
|
1855
|
+
examples.append(image)
|
1856
|
+
run.log({"examples": examples})
|
1857
|
+
```
|
1858
|
+
|
1859
|
+
### Video from numpy
|
1860
|
+
<!--yeadoc-test:init-and-log-video-numpy-->
|
1861
|
+
```python
|
1862
|
+
import numpy as np
|
1863
|
+
import wandb
|
1864
|
+
|
1865
|
+
run = wandb.init()
|
1866
|
+
# axes are (time, channel, height, width)
|
1867
|
+
frames = np.random.randint(low=0, high=256, size=(10, 3, 100, 100), dtype=np.uint8)
|
1868
|
+
run.log({"video": wandb.Video(frames, fps=4)})
|
1869
|
+
```
|
1870
|
+
|
1871
|
+
### Matplotlib Plot
|
1872
|
+
<!--yeadoc-test:init-and-log-matplotlib-->
|
1873
|
+
```python
|
1874
|
+
from matplotlib import pyplot as plt
|
1875
|
+
import numpy as np
|
1876
|
+
import wandb
|
1877
|
+
|
1878
|
+
run = wandb.init()
|
1879
|
+
fig, ax = plt.subplots()
|
1880
|
+
x = np.linspace(0, 10)
|
1881
|
+
y = x * x
|
1882
|
+
ax.plot(x, y) # plot y = x^2
|
1883
|
+
run.log({"chart": fig})
|
1884
|
+
```
|
1885
|
+
|
1886
|
+
### PR Curve
|
1887
|
+
```python
|
1888
|
+
import wandb
|
1889
|
+
|
1890
|
+
run = wandb.init()
|
1891
|
+
run.log({"pr": wandb.plot.pr_curve(y_test, y_probas, labels)})
|
1892
|
+
```
|
1893
|
+
|
1894
|
+
### 3D Object
|
1895
|
+
```python
|
1896
|
+
import wandb
|
1897
|
+
|
1898
|
+
run = wandb.init()
|
1899
|
+
run.log(
|
1900
|
+
{
|
1901
|
+
"generated_samples": [
|
1902
|
+
wandb.Object3D(open("sample.obj")),
|
1903
|
+
wandb.Object3D(open("sample.gltf")),
|
1904
|
+
wandb.Object3D(open("sample.glb")),
|
1905
|
+
]
|
1906
|
+
}
|
1907
|
+
)
|
1908
|
+
```
|
1909
|
+
|
1910
|
+
Raises:
|
1911
|
+
wandb.Error: if called before `wandb.init`
|
1912
|
+
ValueError: if invalid data is passed
|
1913
|
+
|
1914
|
+
"""
|
1915
|
+
if step is not None:
|
1916
|
+
with telemetry.context(run=self) as tel:
|
1917
|
+
tel.feature.set_step_log = True
|
1918
|
+
|
1919
|
+
if sync is not None:
|
1920
|
+
deprecate.deprecate(
|
1921
|
+
field_name=deprecate.Deprecated.run__log_sync,
|
1922
|
+
warning_message=(
|
1923
|
+
"`sync` argument is deprecated and does not affect the behaviour of `wandb.log`"
|
1924
|
+
),
|
1925
|
+
)
|
1926
|
+
if self._settings._shared and step is not None:
|
1927
|
+
wandb.termwarn(
|
1928
|
+
"In shared mode, the use of `wandb.log` with the step argument is not supported "
|
1929
|
+
f"and will be ignored. Please refer to {wburls.get('wandb_define_metric')} "
|
1930
|
+
"on how to customize your x-axis.",
|
1931
|
+
repeat=False,
|
1932
|
+
)
|
1933
|
+
self._log(data=data, step=step, commit=commit)
|
1934
|
+
|
1935
|
+
@_run_decorator._noop_on_finish()
|
1936
|
+
@_run_decorator._attach
|
1937
|
+
def save(
|
1938
|
+
self,
|
1939
|
+
glob_str: Optional[Union[str, os.PathLike]] = None,
|
1940
|
+
base_path: Optional[Union[str, os.PathLike]] = None,
|
1941
|
+
policy: PolicyName = "live",
|
1942
|
+
) -> Union[bool, List[str]]:
|
1943
|
+
"""Sync one or more files to W&B.
|
1944
|
+
|
1945
|
+
Relative paths are relative to the current working directory.
|
1946
|
+
|
1947
|
+
A Unix glob, such as "myfiles/*", is expanded at the time `save` is
|
1948
|
+
called regardless of the `policy`. In particular, new files are not
|
1949
|
+
picked up automatically.
|
1950
|
+
|
1951
|
+
A `base_path` may be provided to control the directory structure of
|
1952
|
+
uploaded files. It should be a prefix of `glob_str`, and the directory
|
1953
|
+
structure beneath it is preserved. It's best understood through
|
1954
|
+
examples:
|
1955
|
+
|
1956
|
+
```
|
1957
|
+
wandb.save("these/are/myfiles/*")
|
1958
|
+
# => Saves files in a "these/are/myfiles/" folder in the run.
|
1959
|
+
|
1960
|
+
wandb.save("these/are/myfiles/*", base_path="these")
|
1961
|
+
# => Saves files in an "are/myfiles/" folder in the run.
|
1962
|
+
|
1963
|
+
wandb.save("/User/username/Documents/run123/*.txt")
|
1964
|
+
# => Saves files in a "run123/" folder in the run. See note below.
|
1965
|
+
|
1966
|
+
wandb.save("/User/username/Documents/run123/*.txt", base_path="/User")
|
1967
|
+
# => Saves files in a "username/Documents/run123/" folder in the run.
|
1968
|
+
|
1969
|
+
wandb.save("files/*/saveme.txt")
|
1970
|
+
# => Saves each "saveme.txt" file in an appropriate subdirectory
|
1971
|
+
# of "files/".
|
1972
|
+
```
|
1973
|
+
|
1974
|
+
Note: when given an absolute path or glob and no `base_path`, one
|
1975
|
+
directory level is preserved as in the example above.
|
1976
|
+
|
1977
|
+
Arguments:
|
1978
|
+
glob_str: A relative or absolute path or Unix glob.
|
1979
|
+
base_path: A path to use to infer a directory structure; see examples.
|
1980
|
+
policy: One of `live`, `now`, or `end`.
|
1981
|
+
* live: upload the file as it changes, overwriting the previous version
|
1982
|
+
* now: upload the file once now
|
1983
|
+
* end: upload file when the run ends
|
1984
|
+
|
1985
|
+
Returns:
|
1986
|
+
Paths to the symlinks created for the matched files.
|
1987
|
+
|
1988
|
+
For historical reasons, this may return a boolean in legacy code.
|
1989
|
+
"""
|
1990
|
+
if glob_str is None:
|
1991
|
+
# noop for historical reasons, run.save() may be called in legacy code
|
1992
|
+
deprecate.deprecate(
|
1993
|
+
field_name=deprecate.Deprecated.run__save_no_args,
|
1994
|
+
warning_message=(
|
1995
|
+
"Calling wandb.run.save without any arguments is deprecated."
|
1996
|
+
"Changes to attributes are automatically persisted."
|
1997
|
+
),
|
1998
|
+
)
|
1999
|
+
return True
|
2000
|
+
|
2001
|
+
if isinstance(glob_str, bytes):
|
2002
|
+
# Preserved for backward compatibility: allow bytes inputs.
|
2003
|
+
glob_str = glob_str.decode("utf-8")
|
2004
|
+
if isinstance(glob_str, str) and (
|
2005
|
+
glob_str.startswith("gs://") or glob_str.startswith("s3://")
|
2006
|
+
):
|
2007
|
+
# Provide a better error message for a common misuse.
|
2008
|
+
wandb.termlog(f"{glob_str} is a cloud storage url, can't save file to W&B.")
|
2009
|
+
return []
|
2010
|
+
# NOTE: We use PurePath instead of Path because WindowsPath doesn't
|
2011
|
+
# like asterisks and errors out in resolve(). It also makes logical
|
2012
|
+
# sense: globs aren't real paths, they're just path-like strings.
|
2013
|
+
glob_path = pathlib.PurePath(glob_str)
|
2014
|
+
resolved_glob_path = pathlib.PurePath(os.path.abspath(glob_path))
|
2015
|
+
|
2016
|
+
if base_path is not None:
|
2017
|
+
base_path = pathlib.Path(base_path)
|
2018
|
+
elif not glob_path.is_absolute():
|
2019
|
+
base_path = pathlib.Path(".")
|
2020
|
+
else:
|
2021
|
+
# Absolute glob paths with no base path get special handling.
|
2022
|
+
wandb.termwarn(
|
2023
|
+
"Saving files without folders. If you want to preserve "
|
2024
|
+
"subdirectories pass base_path to wandb.save, i.e. "
|
2025
|
+
'wandb.save("/mnt/folder/file.h5", base_path="/mnt")',
|
2026
|
+
repeat=False,
|
2027
|
+
)
|
2028
|
+
base_path = resolved_glob_path.parent.parent
|
2029
|
+
|
2030
|
+
if policy not in ("live", "end", "now"):
|
2031
|
+
raise ValueError(
|
2032
|
+
'Only "live", "end" and "now" policies are currently supported.'
|
2033
|
+
)
|
2034
|
+
|
2035
|
+
resolved_base_path = pathlib.PurePath(os.path.abspath(base_path))
|
2036
|
+
|
2037
|
+
return self._save(
|
2038
|
+
resolved_glob_path,
|
2039
|
+
resolved_base_path,
|
2040
|
+
policy,
|
2041
|
+
)
|
2042
|
+
|
2043
|
+
def _save(
|
2044
|
+
self,
|
2045
|
+
glob_path: pathlib.PurePath,
|
2046
|
+
base_path: pathlib.PurePath,
|
2047
|
+
policy: "PolicyName",
|
2048
|
+
) -> List[str]:
|
2049
|
+
# Can't use is_relative_to() because that's added in Python 3.9,
|
2050
|
+
# but we support down to Python 3.7.
|
2051
|
+
if not str(glob_path).startswith(str(base_path)):
|
2052
|
+
raise ValueError("Glob may not walk above the base path")
|
2053
|
+
|
2054
|
+
if glob_path == base_path:
|
2055
|
+
raise ValueError("Glob cannot be the same as the base path")
|
2056
|
+
|
2057
|
+
relative_glob = glob_path.relative_to(base_path)
|
2058
|
+
if relative_glob.parts[0] == "*":
|
2059
|
+
raise ValueError("Glob may not start with '*' relative to the base path")
|
2060
|
+
relative_glob_str = GlobStr(str(relative_glob))
|
2061
|
+
|
2062
|
+
with telemetry.context(run=self) as tel:
|
2063
|
+
tel.feature.save = True
|
2064
|
+
|
2065
|
+
# Files in the files directory matched by the glob, including old and
|
2066
|
+
# new ones.
|
2067
|
+
globbed_files = set(
|
2068
|
+
pathlib.Path(
|
2069
|
+
self._settings.files_dir,
|
2070
|
+
).glob(relative_glob_str)
|
2071
|
+
)
|
2072
|
+
|
2073
|
+
had_symlinked_files = len(globbed_files) > 0
|
2074
|
+
is_star_glob = "*" in relative_glob_str
|
2075
|
+
|
2076
|
+
# The base_path may itself be a glob, so we can't do
|
2077
|
+
# base_path.glob(relative_glob_str)
|
2078
|
+
for path_str in glob.glob(str(base_path / relative_glob_str)):
|
2079
|
+
source_path = pathlib.Path(path_str).absolute()
|
2080
|
+
|
2081
|
+
# We can't use relative_to() because base_path may be a glob.
|
2082
|
+
relative_path = pathlib.Path(*source_path.parts[len(base_path.parts) :])
|
2083
|
+
|
2084
|
+
target_path = pathlib.Path(self._settings.files_dir, relative_path)
|
2085
|
+
globbed_files.add(target_path)
|
2086
|
+
|
2087
|
+
# If the file is already where it needs to be, don't create a symlink.
|
2088
|
+
if source_path.resolve() == target_path.resolve():
|
2089
|
+
continue
|
2090
|
+
|
2091
|
+
target_path.parent.mkdir(parents=True, exist_ok=True)
|
2092
|
+
|
2093
|
+
# Delete the symlink if it exists.
|
2094
|
+
try:
|
2095
|
+
target_path.unlink()
|
2096
|
+
except FileNotFoundError:
|
2097
|
+
# In Python 3.8, we would pass missing_ok=True, but as of now
|
2098
|
+
# we support down to Python 3.7.
|
2099
|
+
pass
|
2100
|
+
|
2101
|
+
target_path.symlink_to(source_path)
|
2102
|
+
|
2103
|
+
# Inform users that new files aren't detected automatically.
|
2104
|
+
if not had_symlinked_files and is_star_glob:
|
2105
|
+
file_str = f"{len(globbed_files)} file"
|
2106
|
+
if len(globbed_files) > 1:
|
2107
|
+
file_str += "s"
|
2108
|
+
wandb.termwarn(
|
2109
|
+
f"Symlinked {file_str} into the W&B run directory, "
|
2110
|
+
"call wandb.save again to sync new files."
|
2111
|
+
)
|
2112
|
+
|
2113
|
+
files_dict: FilesDict = {
|
2114
|
+
"files": [
|
2115
|
+
(
|
2116
|
+
GlobStr(str(f.relative_to(self._settings.files_dir))),
|
2117
|
+
policy,
|
2118
|
+
)
|
2119
|
+
for f in globbed_files
|
2120
|
+
]
|
2121
|
+
}
|
2122
|
+
if self._backend and self._backend.interface:
|
2123
|
+
self._backend.interface.publish_files(files_dict)
|
2124
|
+
|
2125
|
+
return [str(f) for f in globbed_files]
|
2126
|
+
|
2127
|
+
@_run_decorator._attach
|
2128
|
+
def restore(
|
2129
|
+
self,
|
2130
|
+
name: str,
|
2131
|
+
run_path: Optional[str] = None,
|
2132
|
+
replace: bool = False,
|
2133
|
+
root: Optional[str] = None,
|
2134
|
+
) -> Union[None, TextIO]:
|
2135
|
+
return restore(
|
2136
|
+
name,
|
2137
|
+
run_path or self._get_path(),
|
2138
|
+
replace,
|
2139
|
+
root or self._settings.files_dir,
|
2140
|
+
)
|
2141
|
+
|
2142
|
+
@_run_decorator._noop
|
2143
|
+
@_run_decorator._attach
|
2144
|
+
def finish(
|
2145
|
+
self, exit_code: Optional[int] = None, quiet: Optional[bool] = None
|
2146
|
+
) -> None:
|
2147
|
+
"""Mark a run as finished, and finish uploading all data.
|
2148
|
+
|
2149
|
+
This is used when creating multiple runs in the same process. We automatically
|
2150
|
+
call this method when your script exits or if you use the run context manager.
|
2151
|
+
|
2152
|
+
Arguments:
|
2153
|
+
exit_code: Set to something other than 0 to mark a run as failed
|
2154
|
+
quiet: Set to true to minimize log output
|
2155
|
+
"""
|
2156
|
+
return self._finish(exit_code, quiet)
|
2157
|
+
|
2158
|
+
def _finish(
|
2159
|
+
self,
|
2160
|
+
exit_code: Optional[int] = None,
|
2161
|
+
quiet: Optional[bool] = None,
|
2162
|
+
) -> None:
|
2163
|
+
logger.info(f"finishing run {self._get_path()}")
|
2164
|
+
with telemetry.context(run=self) as tel:
|
2165
|
+
tel.feature.finish = True
|
2166
|
+
|
2167
|
+
if quiet is not None:
|
2168
|
+
self._quiet = quiet
|
2169
|
+
|
2170
|
+
# Pop this run (hopefully) from the run stack, to support the "reinit"
|
2171
|
+
# functionality of wandb.init().
|
2172
|
+
#
|
2173
|
+
# TODO: It's not clear how _global_run_stack could have length other
|
2174
|
+
# than 1 at this point in the code. If you're reading this, consider
|
2175
|
+
# refactoring this thing.
|
2176
|
+
if self._wl and len(self._wl._global_run_stack) > 0:
|
2177
|
+
self._wl._global_run_stack.pop()
|
2178
|
+
|
2179
|
+
# Run hooks that need to happen before the last messages to the
|
2180
|
+
# internal service, like Jupyter hooks.
|
2181
|
+
for hook in self._teardown_hooks:
|
2182
|
+
if hook.stage == TeardownStage.EARLY:
|
2183
|
+
hook.call()
|
2184
|
+
|
2185
|
+
# Early-stage hooks may use methods that require _is_finished
|
2186
|
+
# to be False, so we set this after running those hooks.
|
2187
|
+
self._is_finished = True
|
2188
|
+
|
2189
|
+
try:
|
2190
|
+
self._atexit_cleanup(exit_code=exit_code)
|
2191
|
+
|
2192
|
+
# Run hooks that should happen after the last messages to the
|
2193
|
+
# internal service, like detaching the logger.
|
2194
|
+
for hook in self._teardown_hooks:
|
2195
|
+
if hook.stage == TeardownStage.LATE:
|
2196
|
+
hook.call()
|
2197
|
+
self._teardown_hooks = []
|
2198
|
+
|
2199
|
+
# Inform the service that we're done sending messages for this run.
|
2200
|
+
#
|
2201
|
+
# TODO: Why not do this in _atexit_cleanup()?
|
2202
|
+
service = self._wl and self._wl.service
|
2203
|
+
if service:
|
2204
|
+
service.inform_finish(run_id=self._run_id)
|
2205
|
+
|
2206
|
+
finally:
|
2207
|
+
module.unset_globals()
|
2208
|
+
wandb._sentry.end_session()
|
2209
|
+
|
2210
|
+
@_run_decorator._noop
|
2211
|
+
@_run_decorator._attach
|
2212
|
+
def join(self, exit_code: Optional[int] = None) -> None:
|
2213
|
+
"""Deprecated alias for `finish()` - use finish instead."""
|
2214
|
+
if hasattr(self, "_telemetry_obj"):
|
2215
|
+
deprecate.deprecate(
|
2216
|
+
field_name=deprecate.Deprecated.run__join,
|
2217
|
+
warning_message=(
|
2218
|
+
"wandb.run.join() is deprecated, please use wandb.run.finish()."
|
2219
|
+
),
|
2220
|
+
)
|
2221
|
+
self._finish(exit_code=exit_code)
|
2222
|
+
|
2223
|
+
@_run_decorator._noop_on_finish()
|
2224
|
+
@_run_decorator._attach
|
2225
|
+
def status(
|
2226
|
+
self,
|
2227
|
+
) -> RunStatus:
|
2228
|
+
"""Get sync info from the internal backend, about the current run's sync status."""
|
2229
|
+
if not self._backend or not self._backend.interface:
|
2230
|
+
return RunStatus()
|
2231
|
+
|
2232
|
+
handle_run_status = self._backend.interface.deliver_request_run_status()
|
2233
|
+
result = handle_run_status.wait(timeout=-1)
|
2234
|
+
assert result
|
2235
|
+
sync_data = result.response.run_status_response
|
2236
|
+
|
2237
|
+
sync_time = None
|
2238
|
+
if sync_data.sync_time.seconds:
|
2239
|
+
sync_time = datetime.fromtimestamp(
|
2240
|
+
sync_data.sync_time.seconds + sync_data.sync_time.nanos / 1e9
|
2241
|
+
)
|
2242
|
+
return RunStatus(
|
2243
|
+
sync_items_total=sync_data.sync_items_total,
|
2244
|
+
sync_items_pending=sync_data.sync_items_pending,
|
2245
|
+
sync_time=sync_time,
|
2246
|
+
)
|
2247
|
+
|
2248
|
+
@staticmethod
|
2249
|
+
def plot_table(
|
2250
|
+
vega_spec_name: str,
|
2251
|
+
data_table: "wandb.Table",
|
2252
|
+
fields: Dict[str, Any],
|
2253
|
+
string_fields: Optional[Dict[str, Any]] = None,
|
2254
|
+
split_table: Optional[bool] = False,
|
2255
|
+
) -> CustomChart:
|
2256
|
+
"""Create a custom plot on a table.
|
2257
|
+
|
2258
|
+
Arguments:
|
2259
|
+
vega_spec_name: the name of the spec for the plot
|
2260
|
+
data_table: a wandb.Table object containing the data to
|
2261
|
+
be used on the visualization
|
2262
|
+
fields: a dict mapping from table keys to fields that the custom
|
2263
|
+
visualization needs
|
2264
|
+
string_fields: a dict that provides values for any string constants
|
2265
|
+
the custom visualization needs
|
2266
|
+
"""
|
2267
|
+
return custom_chart(
|
2268
|
+
vega_spec_name, data_table, fields, string_fields or {}, split_table
|
2269
|
+
)
|
2270
|
+
|
2271
|
+
def _add_panel(
|
2272
|
+
self, visualize_key: str, panel_type: str, panel_config: dict
|
2273
|
+
) -> None:
|
2274
|
+
config = {
|
2275
|
+
"panel_type": panel_type,
|
2276
|
+
"panel_config": panel_config,
|
2277
|
+
}
|
2278
|
+
self._config_callback(val=config, key=("_wandb", "visualize", visualize_key))
|
2279
|
+
|
2280
|
+
def _set_globals(self) -> None:
|
2281
|
+
module.set_global(
|
2282
|
+
run=self,
|
2283
|
+
config=self.config,
|
2284
|
+
log=self.log,
|
2285
|
+
summary=self.summary,
|
2286
|
+
save=self.save,
|
2287
|
+
use_artifact=self.use_artifact,
|
2288
|
+
log_artifact=self.log_artifact,
|
2289
|
+
define_metric=self.define_metric,
|
2290
|
+
plot_table=self.plot_table,
|
2291
|
+
alert=self.alert,
|
2292
|
+
mark_preempting=self.mark_preempting,
|
2293
|
+
log_model=self.log_model,
|
2294
|
+
use_model=self.use_model,
|
2295
|
+
link_model=self.link_model,
|
2296
|
+
)
|
2297
|
+
|
2298
|
+
def _redirect(
|
2299
|
+
self,
|
2300
|
+
stdout_slave_fd: Optional[int],
|
2301
|
+
stderr_slave_fd: Optional[int],
|
2302
|
+
console: Optional[str] = None,
|
2303
|
+
) -> None:
|
2304
|
+
if console is None:
|
2305
|
+
console = self._settings.console
|
2306
|
+
# only use raw for service to minimize potential changes
|
2307
|
+
if console == "wrap":
|
2308
|
+
if not self._settings._disable_service:
|
2309
|
+
console = "wrap_raw"
|
2310
|
+
else:
|
2311
|
+
console = "wrap_emu"
|
2312
|
+
logger.info("redirect: %s", console)
|
2313
|
+
|
2314
|
+
out_redir: redirect.RedirectBase
|
2315
|
+
err_redir: redirect.RedirectBase
|
2316
|
+
|
2317
|
+
# raw output handles the output_log writing in the internal process
|
2318
|
+
if console in {"redirect", "wrap_emu"}:
|
2319
|
+
output_log_path = os.path.join(
|
2320
|
+
self._settings.files_dir, filenames.OUTPUT_FNAME
|
2321
|
+
)
|
2322
|
+
# output writer might have been set up, see wrap_fallback case
|
2323
|
+
if not self._output_writer:
|
2324
|
+
self._output_writer = filesystem.CRDedupedFile(
|
2325
|
+
open(output_log_path, "wb")
|
2326
|
+
)
|
2327
|
+
|
2328
|
+
if console == "redirect":
|
2329
|
+
logger.info("Redirecting console.")
|
2330
|
+
out_redir = redirect.Redirect(
|
2331
|
+
src="stdout",
|
2332
|
+
cbs=[
|
2333
|
+
lambda data: self._console_callback("stdout", data),
|
2334
|
+
self._output_writer.write, # type: ignore
|
2335
|
+
],
|
2336
|
+
)
|
2337
|
+
err_redir = redirect.Redirect(
|
2338
|
+
src="stderr",
|
2339
|
+
cbs=[
|
2340
|
+
lambda data: self._console_callback("stderr", data),
|
2341
|
+
self._output_writer.write, # type: ignore
|
2342
|
+
],
|
2343
|
+
)
|
2344
|
+
if os.name == "nt":
|
2345
|
+
|
2346
|
+
def wrap_fallback() -> None:
|
2347
|
+
if self._out_redir:
|
2348
|
+
self._out_redir.uninstall()
|
2349
|
+
if self._err_redir:
|
2350
|
+
self._err_redir.uninstall()
|
2351
|
+
msg = (
|
2352
|
+
"Tensorflow detected. Stream redirection is not supported "
|
2353
|
+
"on Windows when tensorflow is imported. Falling back to "
|
2354
|
+
"wrapping stdout/err."
|
2355
|
+
)
|
2356
|
+
wandb.termlog(msg)
|
2357
|
+
self._redirect(None, None, console="wrap")
|
2358
|
+
|
2359
|
+
add_import_hook("tensorflow", wrap_fallback)
|
2360
|
+
elif console == "wrap_emu":
|
2361
|
+
logger.info("Wrapping output streams.")
|
2362
|
+
out_redir = redirect.StreamWrapper(
|
2363
|
+
src="stdout",
|
2364
|
+
cbs=[
|
2365
|
+
lambda data: self._console_callback("stdout", data),
|
2366
|
+
self._output_writer.write, # type: ignore
|
2367
|
+
],
|
2368
|
+
)
|
2369
|
+
err_redir = redirect.StreamWrapper(
|
2370
|
+
src="stderr",
|
2371
|
+
cbs=[
|
2372
|
+
lambda data: self._console_callback("stderr", data),
|
2373
|
+
self._output_writer.write, # type: ignore
|
2374
|
+
],
|
2375
|
+
)
|
2376
|
+
elif console == "wrap_raw":
|
2377
|
+
logger.info("Wrapping output streams.")
|
2378
|
+
out_redir = redirect.StreamRawWrapper(
|
2379
|
+
src="stdout",
|
2380
|
+
cbs=[
|
2381
|
+
lambda data: self._console_raw_callback("stdout", data),
|
2382
|
+
],
|
2383
|
+
)
|
2384
|
+
err_redir = redirect.StreamRawWrapper(
|
2385
|
+
src="stderr",
|
2386
|
+
cbs=[
|
2387
|
+
lambda data: self._console_raw_callback("stderr", data),
|
2388
|
+
],
|
2389
|
+
)
|
2390
|
+
elif console == "off":
|
2391
|
+
return
|
2392
|
+
else:
|
2393
|
+
raise ValueError("unhandled console")
|
2394
|
+
try:
|
2395
|
+
# save stdout and stderr before installing new write functions
|
2396
|
+
out_redir.save()
|
2397
|
+
err_redir.save()
|
2398
|
+
out_redir.install()
|
2399
|
+
err_redir.install()
|
2400
|
+
self._out_redir = out_redir
|
2401
|
+
self._err_redir = err_redir
|
2402
|
+
logger.info("Redirects installed.")
|
2403
|
+
except Exception as e:
|
2404
|
+
print(e)
|
2405
|
+
logger.error("Failed to redirect.", exc_info=e)
|
2406
|
+
return
|
2407
|
+
|
2408
|
+
def _restore(self) -> None:
|
2409
|
+
logger.info("restore")
|
2410
|
+
# TODO(jhr): drain and shutdown all threads
|
2411
|
+
if self._out_redir:
|
2412
|
+
self._out_redir.uninstall()
|
2413
|
+
if self._err_redir:
|
2414
|
+
self._err_redir.uninstall()
|
2415
|
+
logger.info("restore done")
|
2416
|
+
|
2417
|
+
def _atexit_cleanup(self, exit_code: Optional[int] = None) -> None:
|
2418
|
+
if self._backend is None:
|
2419
|
+
logger.warning("process exited without backend configured")
|
2420
|
+
return
|
2421
|
+
if self._atexit_cleanup_called:
|
2422
|
+
return
|
2423
|
+
self._atexit_cleanup_called = True
|
2424
|
+
|
2425
|
+
exit_code = exit_code or (self._hooks and self._hooks.exit_code) or 0
|
2426
|
+
self._exit_code = exit_code
|
2427
|
+
logger.info(f"got exitcode: {exit_code}")
|
2428
|
+
|
2429
|
+
# Delete this run's "resume" file if the run finished successfully.
|
2430
|
+
#
|
2431
|
+
# This is used by the "auto" resume mode, which resumes from the last
|
2432
|
+
# failed (or unfinished/crashed) run. If we reach this line, then this
|
2433
|
+
# run shouldn't be a candidate for "auto" resume.
|
2434
|
+
if exit_code == 0:
|
2435
|
+
if os.path.exists(self._settings.resume_fname):
|
2436
|
+
os.remove(self._settings.resume_fname)
|
2437
|
+
|
2438
|
+
try:
|
2439
|
+
self._on_finish()
|
2440
|
+
|
2441
|
+
except KeyboardInterrupt:
|
2442
|
+
if not wandb.wandb_agent._is_running(): # type: ignore
|
2443
|
+
wandb.termerror("Control-C detected -- Run data was not synced")
|
2444
|
+
raise
|
2445
|
+
|
2446
|
+
except Exception as e:
|
2447
|
+
self._console_stop()
|
2448
|
+
logger.error("Problem finishing run", exc_info=e)
|
2449
|
+
wandb.termerror("Problem finishing run")
|
2450
|
+
raise
|
2451
|
+
|
2452
|
+
Run._footer(
|
2453
|
+
sampled_history=self._sampled_history,
|
2454
|
+
final_summary=self._final_summary,
|
2455
|
+
poll_exit_response=self._poll_exit_response,
|
2456
|
+
internal_messages_response=self._internal_messages_response,
|
2457
|
+
reporter=self._reporter,
|
2458
|
+
quiet=self._quiet,
|
2459
|
+
settings=self._settings,
|
2460
|
+
printer=self._printer,
|
2461
|
+
)
|
2462
|
+
|
2463
|
+
def _console_start(self) -> None:
|
2464
|
+
logger.info("atexit reg")
|
2465
|
+
self._hooks = ExitHooks()
|
2466
|
+
|
2467
|
+
service = self._wl and self._wl.service
|
2468
|
+
if not service:
|
2469
|
+
self._hooks.hook()
|
2470
|
+
# NB: manager will perform atexit hook like behavior for outstanding runs
|
2471
|
+
atexit.register(lambda: self._atexit_cleanup())
|
2472
|
+
|
2473
|
+
self._redirect(self._stdout_slave_fd, self._stderr_slave_fd)
|
2474
|
+
|
2475
|
+
def _console_stop(self) -> None:
|
2476
|
+
self._restore()
|
2477
|
+
if self._output_writer:
|
2478
|
+
self._output_writer.close()
|
2479
|
+
self._output_writer = None
|
2480
|
+
|
2481
|
+
def _on_init(self) -> None:
|
2482
|
+
if self._settings._offline:
|
2483
|
+
return
|
2484
|
+
|
2485
|
+
def _on_start(self) -> None:
|
2486
|
+
# would like to move _set_global to _on_ready to unify _on_start and _on_attach
|
2487
|
+
# (we want to do the set globals after attach)
|
2488
|
+
# TODO(console) However _console_start calls Redirect that uses `wandb.run` hence breaks
|
2489
|
+
# TODO(jupyter) However _header calls _header_run_info that uses wandb.jupyter that uses
|
2490
|
+
# `wandb.run` and hence breaks
|
2491
|
+
self._set_globals()
|
2492
|
+
self._header(settings=self._settings, printer=self._printer)
|
2493
|
+
|
2494
|
+
if self._settings.save_code and self._settings.code_dir is not None:
|
2495
|
+
self.log_code(self._settings.code_dir)
|
2496
|
+
|
2497
|
+
if self._settings._save_requirements:
|
2498
|
+
if self._backend and self._backend.interface:
|
2499
|
+
from wandb.util import working_set
|
2500
|
+
|
2501
|
+
logger.debug(
|
2502
|
+
"Saving list of pip packages installed into the current environment"
|
2503
|
+
)
|
2504
|
+
self._backend.interface.publish_python_packages(working_set())
|
2505
|
+
|
2506
|
+
if self._backend and self._backend.interface and not self._settings._offline:
|
2507
|
+
self._run_status_checker = RunStatusChecker(
|
2508
|
+
interface=self._backend.interface,
|
2509
|
+
)
|
2510
|
+
self._run_status_checker.start()
|
2511
|
+
|
2512
|
+
self._console_start()
|
2513
|
+
self._on_ready()
|
2514
|
+
|
2515
|
+
def _on_attach(self) -> None:
|
2516
|
+
"""Event triggered when run is attached to another run."""
|
2517
|
+
with telemetry.context(run=self) as tel:
|
2518
|
+
tel.feature.attach = True
|
2519
|
+
|
2520
|
+
self._set_globals()
|
2521
|
+
self._is_attached = True
|
2522
|
+
self._on_ready()
|
2523
|
+
|
2524
|
+
def _register_telemetry_import_hooks(
|
2525
|
+
self,
|
2526
|
+
) -> None:
|
2527
|
+
def _telemetry_import_hook(
|
2528
|
+
run: "Run",
|
2529
|
+
module: Any,
|
2530
|
+
) -> None:
|
2531
|
+
with telemetry.context(run=run) as tel:
|
2532
|
+
try:
|
2533
|
+
name = getattr(module, "__name__", None)
|
2534
|
+
if name is not None:
|
2535
|
+
setattr(tel.imports_finish, name, True)
|
2536
|
+
except AttributeError:
|
2537
|
+
return
|
2538
|
+
|
2539
|
+
import_telemetry_set = telemetry.list_telemetry_imports()
|
2540
|
+
import_hook_fn = functools.partial(_telemetry_import_hook, self)
|
2541
|
+
for module_name in import_telemetry_set:
|
2542
|
+
register_post_import_hook(
|
2543
|
+
import_hook_fn,
|
2544
|
+
self._run_id,
|
2545
|
+
module_name,
|
2546
|
+
)
|
2547
|
+
|
2548
|
+
def _on_ready(self) -> None:
|
2549
|
+
"""Event triggered when run is ready for the user."""
|
2550
|
+
self._register_telemetry_import_hooks()
|
2551
|
+
|
2552
|
+
# start reporting any telemetry changes
|
2553
|
+
self._telemetry_obj_active = True
|
2554
|
+
self._telemetry_flush()
|
2555
|
+
|
2556
|
+
self._detect_and_apply_job_inputs()
|
2557
|
+
|
2558
|
+
# object is about to be returned to the user, don't let them modify it
|
2559
|
+
self._freeze()
|
2560
|
+
|
2561
|
+
if not self._settings.resume:
|
2562
|
+
if os.path.exists(self._settings.resume_fname):
|
2563
|
+
os.remove(self._settings.resume_fname)
|
2564
|
+
|
2565
|
+
def _detect_and_apply_job_inputs(self) -> None:
|
2566
|
+
"""If the user has staged launch inputs, apply them to the run."""
|
2567
|
+
from wandb.sdk.launch.inputs.internal import StagedLaunchInputs
|
2568
|
+
|
2569
|
+
StagedLaunchInputs().apply(self)
|
2570
|
+
|
2571
|
+
def _make_job_source_reqs(self) -> Tuple[List[str], Dict[str, Any], Dict[str, Any]]:
|
2572
|
+
from wandb.util import working_set
|
2573
|
+
|
2574
|
+
installed_packages_list = sorted(f"{d.key}=={d.version}" for d in working_set())
|
2575
|
+
input_types = TypeRegistry.type_of(self.config.as_dict()).to_json()
|
2576
|
+
output_types = TypeRegistry.type_of(self.summary._as_dict()).to_json()
|
2577
|
+
|
2578
|
+
return installed_packages_list, input_types, output_types
|
2579
|
+
|
2580
|
+
def _construct_job_artifact(
|
2581
|
+
self,
|
2582
|
+
name: str,
|
2583
|
+
source_dict: "JobSourceDict",
|
2584
|
+
installed_packages_list: List[str],
|
2585
|
+
patch_path: Optional[os.PathLike] = None,
|
2586
|
+
) -> "Artifact":
|
2587
|
+
job_artifact = job_builder.JobArtifact(name)
|
2588
|
+
if patch_path and os.path.exists(patch_path):
|
2589
|
+
job_artifact.add_file(FilePathStr(str(patch_path)), "diff.patch")
|
2590
|
+
with job_artifact.new_file("requirements.frozen.txt") as f:
|
2591
|
+
f.write("\n".join(installed_packages_list))
|
2592
|
+
with job_artifact.new_file("wandb-job.json") as f:
|
2593
|
+
f.write(json.dumps(source_dict))
|
2594
|
+
|
2595
|
+
return job_artifact
|
2596
|
+
|
2597
|
+
def _create_image_job(
|
2598
|
+
self,
|
2599
|
+
input_types: Dict[str, Any],
|
2600
|
+
output_types: Dict[str, Any],
|
2601
|
+
installed_packages_list: List[str],
|
2602
|
+
docker_image_name: Optional[str] = None,
|
2603
|
+
args: Optional[List[str]] = None,
|
2604
|
+
) -> Optional["Artifact"]:
|
2605
|
+
docker_image_name = docker_image_name or os.getenv("WANDB_DOCKER")
|
2606
|
+
|
2607
|
+
if not docker_image_name:
|
2608
|
+
return None
|
2609
|
+
|
2610
|
+
name = wandb.util.make_artifact_name_safe(f"job-{docker_image_name}")
|
2611
|
+
s_args: Sequence[str] = args if args is not None else self._settings._args
|
2612
|
+
source_info: JobSourceDict = {
|
2613
|
+
"_version": "v0",
|
2614
|
+
"source_type": "image",
|
2615
|
+
"source": {"image": docker_image_name, "args": s_args},
|
2616
|
+
"input_types": input_types,
|
2617
|
+
"output_types": output_types,
|
2618
|
+
"runtime": self._settings._python,
|
2619
|
+
}
|
2620
|
+
job_artifact = self._construct_job_artifact(
|
2621
|
+
name, source_info, installed_packages_list
|
2622
|
+
)
|
2623
|
+
|
2624
|
+
return job_artifact
|
2625
|
+
|
2626
|
+
def _log_job_artifact_with_image(
|
2627
|
+
self, docker_image_name: str, args: Optional[List[str]] = None
|
2628
|
+
) -> Artifact:
|
2629
|
+
packages, in_types, out_types = self._make_job_source_reqs()
|
2630
|
+
job_artifact = self._create_image_job(
|
2631
|
+
in_types,
|
2632
|
+
out_types,
|
2633
|
+
packages,
|
2634
|
+
args=args,
|
2635
|
+
docker_image_name=docker_image_name,
|
2636
|
+
)
|
2637
|
+
|
2638
|
+
artifact = self.log_artifact(job_artifact)
|
2639
|
+
|
2640
|
+
if not artifact:
|
2641
|
+
raise wandb.Error(f"Job Artifact log unsuccessful: {artifact}")
|
2642
|
+
else:
|
2643
|
+
return artifact
|
2644
|
+
|
2645
|
+
def _on_probe_exit(self, probe_handle: MailboxProbe) -> None:
|
2646
|
+
handle = probe_handle.get_mailbox_handle()
|
2647
|
+
if handle:
|
2648
|
+
result = handle.wait(timeout=0, release=False)
|
2649
|
+
if not result:
|
2650
|
+
return
|
2651
|
+
probe_handle.set_probe_result(result)
|
2652
|
+
assert self._backend and self._backend.interface
|
2653
|
+
handle = self._backend.interface.deliver_poll_exit()
|
2654
|
+
probe_handle.set_mailbox_handle(handle)
|
2655
|
+
|
2656
|
+
def _on_progress_exit(self, progress_handle: MailboxProgress) -> None:
|
2657
|
+
probe_handles = progress_handle.get_probe_handles()
|
2658
|
+
assert probe_handles and len(probe_handles) == 1
|
2659
|
+
|
2660
|
+
result = probe_handles[0].get_probe_result()
|
2661
|
+
if not result:
|
2662
|
+
return
|
2663
|
+
self._footer_file_pusher_status_info(
|
2664
|
+
result.response.poll_exit_response, printer=self._printer
|
2665
|
+
)
|
2666
|
+
|
2667
|
+
def _on_finish(self) -> None:
|
2668
|
+
trigger.call("on_finished")
|
2669
|
+
|
2670
|
+
if self._run_status_checker is not None:
|
2671
|
+
self._run_status_checker.stop()
|
2672
|
+
|
2673
|
+
self._console_stop() # TODO: there's a race here with jupyter console logging
|
2674
|
+
|
2675
|
+
assert self._backend and self._backend.interface
|
2676
|
+
|
2677
|
+
exit_handle = self._backend.interface.deliver_exit(self._exit_code)
|
2678
|
+
exit_handle.add_probe(on_probe=self._on_probe_exit)
|
2679
|
+
|
2680
|
+
# wait for the exit to complete
|
2681
|
+
_ = exit_handle.wait(timeout=-1, on_progress=self._on_progress_exit)
|
2682
|
+
|
2683
|
+
poll_exit_handle = self._backend.interface.deliver_poll_exit()
|
2684
|
+
# wait for them, it's ok to do this serially but this can be improved
|
2685
|
+
result = poll_exit_handle.wait(timeout=-1)
|
2686
|
+
assert result
|
2687
|
+
self._footer_file_pusher_status_info(
|
2688
|
+
result.response.poll_exit_response, printer=self._printer
|
2689
|
+
)
|
2690
|
+
self._poll_exit_response = result.response.poll_exit_response
|
2691
|
+
internal_messages_handle = self._backend.interface.deliver_internal_messages()
|
2692
|
+
result = internal_messages_handle.wait(timeout=-1)
|
2693
|
+
assert result
|
2694
|
+
self._internal_messages_response = result.response.internal_messages_response
|
2695
|
+
|
2696
|
+
# dispatch all our final requests
|
2697
|
+
|
2698
|
+
final_summary_handle = self._backend.interface.deliver_get_summary()
|
2699
|
+
sampled_history_handle = (
|
2700
|
+
self._backend.interface.deliver_request_sampled_history()
|
2701
|
+
)
|
2702
|
+
|
2703
|
+
result = sampled_history_handle.wait(timeout=-1)
|
2704
|
+
assert result
|
2705
|
+
self._sampled_history = result.response.sampled_history_response
|
2706
|
+
|
2707
|
+
result = final_summary_handle.wait(timeout=-1)
|
2708
|
+
assert result
|
2709
|
+
self._final_summary = result.response.get_summary_response
|
2710
|
+
|
2711
|
+
if self._backend:
|
2712
|
+
self._backend.cleanup()
|
2713
|
+
|
2714
|
+
if self._run_status_checker:
|
2715
|
+
self._run_status_checker.join()
|
2716
|
+
|
2717
|
+
self._unregister_telemetry_import_hooks(self._run_id)
|
2718
|
+
|
2719
|
+
@staticmethod
|
2720
|
+
def _unregister_telemetry_import_hooks(run_id: str) -> None:
|
2721
|
+
import_telemetry_set = telemetry.list_telemetry_imports()
|
2722
|
+
for module_name in import_telemetry_set:
|
2723
|
+
unregister_post_import_hook(module_name, run_id)
|
2724
|
+
|
2725
|
+
@_run_decorator._noop_on_finish()
|
2726
|
+
@_run_decorator._attach
|
2727
|
+
def define_metric(
|
2728
|
+
self,
|
2729
|
+
name: str,
|
2730
|
+
step_metric: Union[str, wandb_metric.Metric, None] = None,
|
2731
|
+
step_sync: Optional[bool] = None,
|
2732
|
+
hidden: Optional[bool] = None,
|
2733
|
+
summary: Optional[str] = None,
|
2734
|
+
goal: Optional[str] = None,
|
2735
|
+
overwrite: Optional[bool] = None,
|
2736
|
+
) -> wandb_metric.Metric:
|
2737
|
+
"""Customize metrics logged with `wandb.log()`.
|
2738
|
+
|
2739
|
+
Arguments:
|
2740
|
+
name: The name of the metric to customize.
|
2741
|
+
step_metric: The name of another metric to serve as the X-axis
|
2742
|
+
for this metric in automatically generated charts.
|
2743
|
+
step_sync: Automatically insert the last value of step_metric into
|
2744
|
+
`run.log()` if it is not provided explicitly. Defaults to True
|
2745
|
+
if step_metric is specified.
|
2746
|
+
hidden: Hide this metric from automatic plots.
|
2747
|
+
summary: Specify aggregate metrics added to summary.
|
2748
|
+
Supported aggregations include "min", "max", "mean", "last",
|
2749
|
+
"best", "copy" and "none". "best" is used together with the
|
2750
|
+
goal parameter. "none" prevents a summary from being generated.
|
2751
|
+
"copy" is deprecated and should not be used.
|
2752
|
+
goal: Specify how to interpret the "best" summary type.
|
2753
|
+
Supported options are "minimize" and "maximize".
|
2754
|
+
overwrite: If false, then this call is merged with previous
|
2755
|
+
`define_metric` calls for the same metric by using their
|
2756
|
+
values for any unspecified parameters. If true, then
|
2757
|
+
unspecified parameters overwrite values specified by
|
2758
|
+
previous calls.
|
2759
|
+
|
2760
|
+
Returns:
|
2761
|
+
An object that represents this call but can otherwise be discarded.
|
2762
|
+
"""
|
2763
|
+
if summary and "copy" in summary:
|
2764
|
+
deprecate.deprecate(
|
2765
|
+
deprecate.Deprecated.run__define_metric_copy,
|
2766
|
+
"define_metric(summary='copy') is deprecated and will be removed.",
|
2767
|
+
self,
|
2768
|
+
)
|
2769
|
+
|
2770
|
+
if (summary and "best" in summary) or goal is not None:
|
2771
|
+
deprecate.deprecate(
|
2772
|
+
deprecate.Deprecated.run__define_metric_best_goal,
|
2773
|
+
"define_metric(summary='best', goal=...) is deprecated and will be removed. "
|
2774
|
+
"Use define_metric(summary='min') or define_metric(summary='max') instead.",
|
2775
|
+
self,
|
2776
|
+
)
|
2777
|
+
|
2778
|
+
return self._define_metric(
|
2779
|
+
name,
|
2780
|
+
step_metric,
|
2781
|
+
step_sync,
|
2782
|
+
hidden,
|
2783
|
+
summary,
|
2784
|
+
goal,
|
2785
|
+
overwrite,
|
2786
|
+
)
|
2787
|
+
|
2788
|
+
def _define_metric(
|
2789
|
+
self,
|
2790
|
+
name: str,
|
2791
|
+
step_metric: Union[str, wandb_metric.Metric, None] = None,
|
2792
|
+
step_sync: Optional[bool] = None,
|
2793
|
+
hidden: Optional[bool] = None,
|
2794
|
+
summary: Optional[str] = None,
|
2795
|
+
goal: Optional[str] = None,
|
2796
|
+
overwrite: Optional[bool] = None,
|
2797
|
+
) -> wandb_metric.Metric:
|
2798
|
+
if not name:
|
2799
|
+
raise wandb.Error("define_metric() requires non-empty name argument")
|
2800
|
+
if isinstance(step_metric, wandb_metric.Metric):
|
2801
|
+
step_metric = step_metric.name
|
2802
|
+
for arg_name, arg_val, exp_type in (
|
2803
|
+
("name", name, str),
|
2804
|
+
("step_metric", step_metric, str),
|
2805
|
+
("step_sync", step_sync, bool),
|
2806
|
+
("hidden", hidden, bool),
|
2807
|
+
("summary", summary, str),
|
2808
|
+
("goal", goal, str),
|
2809
|
+
("overwrite", overwrite, bool),
|
2810
|
+
):
|
2811
|
+
# NOTE: type checking is broken for isinstance and str
|
2812
|
+
if arg_val is not None and not isinstance(arg_val, exp_type):
|
2813
|
+
arg_type = type(arg_val).__name__
|
2814
|
+
raise wandb.Error(
|
2815
|
+
f"Unhandled define_metric() arg: {arg_name} type: {arg_type}"
|
2816
|
+
)
|
2817
|
+
stripped = name[:-1] if name.endswith("*") else name
|
2818
|
+
if "*" in stripped:
|
2819
|
+
raise wandb.Error(
|
2820
|
+
f"Unhandled define_metric() arg: name (glob suffixes only): {name}"
|
2821
|
+
)
|
2822
|
+
summary_ops: Optional[Sequence[str]] = None
|
2823
|
+
if summary:
|
2824
|
+
summary_items = [s.lower() for s in summary.split(",")]
|
2825
|
+
summary_ops = []
|
2826
|
+
valid = {"min", "max", "mean", "best", "last", "copy", "none"}
|
2827
|
+
# TODO: deprecate copy and best
|
2828
|
+
for i in summary_items:
|
2829
|
+
if i not in valid:
|
2830
|
+
raise wandb.Error(f"Unhandled define_metric() arg: summary op: {i}")
|
2831
|
+
summary_ops.append(i)
|
2832
|
+
with telemetry.context(run=self) as tel:
|
2833
|
+
tel.feature.metric_summary = True
|
2834
|
+
# TODO: deprecate goal
|
2835
|
+
goal_cleaned: Optional[str] = None
|
2836
|
+
if goal is not None:
|
2837
|
+
goal_cleaned = goal[:3].lower()
|
2838
|
+
valid_goal = {"min", "max"}
|
2839
|
+
if goal_cleaned not in valid_goal:
|
2840
|
+
raise wandb.Error(f"Unhandled define_metric() arg: goal: {goal}")
|
2841
|
+
with telemetry.context(run=self) as tel:
|
2842
|
+
tel.feature.metric_goal = True
|
2843
|
+
if hidden:
|
2844
|
+
with telemetry.context(run=self) as tel:
|
2845
|
+
tel.feature.metric_hidden = True
|
2846
|
+
if step_sync:
|
2847
|
+
with telemetry.context(run=self) as tel:
|
2848
|
+
tel.feature.metric_step_sync = True
|
2849
|
+
|
2850
|
+
with telemetry.context(run=self) as tel:
|
2851
|
+
tel.feature.metric = True
|
2852
|
+
|
2853
|
+
m = wandb_metric.Metric(
|
2854
|
+
name=name,
|
2855
|
+
step_metric=step_metric,
|
2856
|
+
step_sync=step_sync,
|
2857
|
+
summary=summary_ops,
|
2858
|
+
hidden=hidden,
|
2859
|
+
goal=goal_cleaned,
|
2860
|
+
overwrite=overwrite,
|
2861
|
+
)
|
2862
|
+
m._set_callback(self._metric_callback)
|
2863
|
+
m._commit()
|
2864
|
+
return m
|
2865
|
+
|
2866
|
+
# TODO(jhr): annotate this
|
2867
|
+
@_run_decorator._attach
|
2868
|
+
def watch( # type: ignore
|
2869
|
+
self,
|
2870
|
+
models,
|
2871
|
+
criterion=None,
|
2872
|
+
log="gradients",
|
2873
|
+
log_freq=100,
|
2874
|
+
idx=None,
|
2875
|
+
log_graph=False,
|
2876
|
+
) -> None:
|
2877
|
+
wandb.watch(models, criterion, log, log_freq, idx, log_graph) # type: ignore
|
2878
|
+
|
2879
|
+
# TODO(jhr): annotate this
|
2880
|
+
@_run_decorator._attach
|
2881
|
+
def unwatch(self, models=None) -> None: # type: ignore
|
2882
|
+
wandb.unwatch(models=models) # type: ignore
|
2883
|
+
|
2884
|
+
# TODO(kdg): remove all artifact swapping logic
|
2885
|
+
def _swap_artifact_name(self, artifact_name: str, use_as: Optional[str]) -> str:
|
2886
|
+
artifact_key_string = use_as or artifact_name
|
2887
|
+
replacement_artifact_info = self._launch_artifact_mapping.get(
|
2888
|
+
artifact_key_string
|
2889
|
+
)
|
2890
|
+
if replacement_artifact_info is not None:
|
2891
|
+
new_name = replacement_artifact_info.get("name")
|
2892
|
+
entity = replacement_artifact_info.get("entity")
|
2893
|
+
project = replacement_artifact_info.get("project")
|
2894
|
+
if new_name is None or entity is None or project is None:
|
2895
|
+
raise ValueError(
|
2896
|
+
"Misconfigured artifact in launch config. Must include name, project and entity keys."
|
2897
|
+
)
|
2898
|
+
return f"{entity}/{project}/{new_name}"
|
2899
|
+
elif replacement_artifact_info is None and use_as is None:
|
2900
|
+
sequence_name = artifact_name.split(":")[0].split("/")[-1]
|
2901
|
+
unique_artifact_replacement_info = (
|
2902
|
+
self._unique_launch_artifact_sequence_names.get(sequence_name)
|
2903
|
+
)
|
2904
|
+
if unique_artifact_replacement_info is not None:
|
2905
|
+
new_name = unique_artifact_replacement_info.get("name")
|
2906
|
+
entity = unique_artifact_replacement_info.get("entity")
|
2907
|
+
project = unique_artifact_replacement_info.get("project")
|
2908
|
+
if new_name is None or entity is None or project is None:
|
2909
|
+
raise ValueError(
|
2910
|
+
"Misconfigured artifact in launch config. Must include name, project and entity keys."
|
2911
|
+
)
|
2912
|
+
return f"{entity}/{project}/{new_name}"
|
2913
|
+
|
2914
|
+
else:
|
2915
|
+
return artifact_name
|
2916
|
+
|
2917
|
+
return artifact_name
|
2918
|
+
|
2919
|
+
def _detach(self) -> None:
|
2920
|
+
pass
|
2921
|
+
|
2922
|
+
@_run_decorator._noop_on_finish()
|
2923
|
+
@_run_decorator._attach
|
2924
|
+
def link_artifact(
|
2925
|
+
self,
|
2926
|
+
artifact: Artifact,
|
2927
|
+
target_path: str,
|
2928
|
+
aliases: Optional[List[str]] = None,
|
2929
|
+
) -> None:
|
2930
|
+
"""Link the given artifact to a portfolio (a promoted collection of artifacts).
|
2931
|
+
|
2932
|
+
The linked artifact will be visible in the UI for the specified portfolio.
|
2933
|
+
|
2934
|
+
Arguments:
|
2935
|
+
artifact: the (public or local) artifact which will be linked
|
2936
|
+
target_path: `str` - takes the following forms: {portfolio}, {project}/{portfolio},
|
2937
|
+
or {entity}/{project}/{portfolio}
|
2938
|
+
aliases: `List[str]` - optional alias(es) that will only be applied on this linked artifact
|
2939
|
+
inside the portfolio.
|
2940
|
+
The alias "latest" will always be applied to the latest version of an artifact that is linked.
|
2941
|
+
|
2942
|
+
Returns:
|
2943
|
+
None
|
2944
|
+
|
2945
|
+
"""
|
2946
|
+
portfolio, project, entity = wandb.util._parse_entity_project_item(target_path)
|
2947
|
+
if aliases is None:
|
2948
|
+
aliases = []
|
2949
|
+
|
2950
|
+
if self._backend and self._backend.interface:
|
2951
|
+
if artifact.is_draft() and not artifact._is_draft_save_started():
|
2952
|
+
artifact = self._log_artifact(artifact)
|
2953
|
+
if not self._settings._offline:
|
2954
|
+
handle = self._backend.interface.deliver_link_artifact(
|
2955
|
+
self,
|
2956
|
+
artifact,
|
2957
|
+
portfolio,
|
2958
|
+
aliases,
|
2959
|
+
entity,
|
2960
|
+
project,
|
2961
|
+
)
|
2962
|
+
if artifact._ttl_duration_seconds is not None:
|
2963
|
+
wandb.termwarn(
|
2964
|
+
"Artifact TTL will be disabled for source artifacts that are linked to portfolios."
|
2965
|
+
)
|
2966
|
+
result = handle.wait(timeout=-1)
|
2967
|
+
if result is None:
|
2968
|
+
handle.abandon()
|
2969
|
+
else:
|
2970
|
+
response = result.response.link_artifact_response
|
2971
|
+
if response.error_message:
|
2972
|
+
wandb.termerror(response.error_message)
|
2973
|
+
else:
|
2974
|
+
# TODO: implement offline mode + sync
|
2975
|
+
raise NotImplementedError
|
2976
|
+
|
2977
|
+
@_run_decorator._noop_on_finish()
|
2978
|
+
@_run_decorator._attach
|
2979
|
+
def use_artifact(
|
2980
|
+
self,
|
2981
|
+
artifact_or_name: Union[str, Artifact],
|
2982
|
+
type: Optional[str] = None,
|
2983
|
+
aliases: Optional[List[str]] = None,
|
2984
|
+
use_as: Optional[str] = None,
|
2985
|
+
) -> Artifact:
|
2986
|
+
"""Declare an artifact as an input to a run.
|
2987
|
+
|
2988
|
+
Call `download` or `file` on the returned object to get the contents locally.
|
2989
|
+
|
2990
|
+
Arguments:
|
2991
|
+
artifact_or_name: (str or Artifact) An artifact name.
|
2992
|
+
May be prefixed with entity/project/. Valid names
|
2993
|
+
can be in the following forms:
|
2994
|
+
- name:version
|
2995
|
+
- name:alias
|
2996
|
+
You can also pass an Artifact object created by calling `wandb.Artifact`
|
2997
|
+
type: (str, optional) The type of artifact to use.
|
2998
|
+
aliases: (list, optional) Aliases to apply to this artifact
|
2999
|
+
use_as: (string, optional) Optional string indicating what purpose the artifact was used with.
|
3000
|
+
Will be shown in UI.
|
3001
|
+
|
3002
|
+
Returns:
|
3003
|
+
An `Artifact` object.
|
3004
|
+
"""
|
3005
|
+
if self._settings._offline:
|
3006
|
+
raise TypeError("Cannot use artifact when in offline mode.")
|
3007
|
+
r = self._run_obj
|
3008
|
+
assert r is not None
|
3009
|
+
api = internal.Api(default_settings={"entity": r.entity, "project": r.project})
|
3010
|
+
api.set_current_run_id(self._run_id)
|
3011
|
+
|
3012
|
+
if isinstance(artifact_or_name, str):
|
3013
|
+
if self._launch_artifact_mapping:
|
3014
|
+
name = self._swap_artifact_name(artifact_or_name, use_as)
|
3015
|
+
else:
|
3016
|
+
name = artifact_or_name
|
3017
|
+
public_api = self._public_api()
|
3018
|
+
artifact = public_api.artifact(type=type, name=name)
|
3019
|
+
if type is not None and type != artifact.type:
|
3020
|
+
raise ValueError(
|
3021
|
+
"Supplied type {} does not match type {} of artifact {}".format(
|
3022
|
+
type, artifact.type, artifact.name
|
3023
|
+
)
|
3024
|
+
)
|
3025
|
+
artifact._use_as = use_as or artifact_or_name
|
3026
|
+
if use_as:
|
3027
|
+
if (
|
3028
|
+
use_as in self._used_artifact_slots.keys()
|
3029
|
+
and self._used_artifact_slots[use_as] != artifact.id
|
3030
|
+
):
|
3031
|
+
raise ValueError(
|
3032
|
+
"Cannot call use_artifact with the same use_as argument more than once"
|
3033
|
+
)
|
3034
|
+
elif ":" in use_as or "/" in use_as:
|
3035
|
+
raise ValueError(
|
3036
|
+
"use_as cannot contain special characters ':' or '/'"
|
3037
|
+
)
|
3038
|
+
self._used_artifact_slots[use_as] = artifact.id
|
3039
|
+
api.use_artifact(
|
3040
|
+
artifact.id,
|
3041
|
+
entity_name=r.entity,
|
3042
|
+
project_name=r.project,
|
3043
|
+
use_as=use_as or artifact_or_name,
|
3044
|
+
)
|
3045
|
+
else:
|
3046
|
+
artifact = artifact_or_name
|
3047
|
+
if aliases is None:
|
3048
|
+
aliases = []
|
3049
|
+
elif isinstance(aliases, str):
|
3050
|
+
aliases = [aliases]
|
3051
|
+
if isinstance(artifact_or_name, Artifact) and artifact.is_draft():
|
3052
|
+
if use_as is not None:
|
3053
|
+
wandb.termwarn(
|
3054
|
+
"Indicating use_as is not supported when using a draft artifact"
|
3055
|
+
)
|
3056
|
+
self._log_artifact(
|
3057
|
+
artifact,
|
3058
|
+
aliases=aliases,
|
3059
|
+
is_user_created=True,
|
3060
|
+
use_after_commit=True,
|
3061
|
+
)
|
3062
|
+
artifact.wait()
|
3063
|
+
artifact._use_as = use_as or artifact.name
|
3064
|
+
elif isinstance(artifact, Artifact) and not artifact.is_draft():
|
3065
|
+
if (
|
3066
|
+
self._launch_artifact_mapping
|
3067
|
+
and artifact.name in self._launch_artifact_mapping.keys()
|
3068
|
+
):
|
3069
|
+
wandb.termwarn(
|
3070
|
+
"Swapping artifacts is not supported when using a non-draft artifact. "
|
3071
|
+
f"Using {artifact.name}."
|
3072
|
+
)
|
3073
|
+
artifact._use_as = use_as or artifact.name
|
3074
|
+
api.use_artifact(
|
3075
|
+
artifact.id, use_as=use_as or artifact._use_as or artifact.name
|
3076
|
+
)
|
3077
|
+
else:
|
3078
|
+
raise ValueError(
|
3079
|
+
'You must pass an artifact name (e.g. "pedestrian-dataset:v1"), '
|
3080
|
+
"an instance of `wandb.Artifact`, or `wandb.Api().artifact()` to `use_artifact`"
|
3081
|
+
)
|
3082
|
+
if self._backend and self._backend.interface:
|
3083
|
+
self._backend.interface.publish_use_artifact(artifact)
|
3084
|
+
return artifact
|
3085
|
+
|
3086
|
+
@_run_decorator._noop_on_finish()
|
3087
|
+
@_run_decorator._attach
|
3088
|
+
def log_artifact(
|
3089
|
+
self,
|
3090
|
+
artifact_or_path: Union[Artifact, StrPath],
|
3091
|
+
name: Optional[str] = None,
|
3092
|
+
type: Optional[str] = None,
|
3093
|
+
aliases: Optional[List[str]] = None,
|
3094
|
+
tags: Optional[List[str]] = None,
|
3095
|
+
) -> Artifact:
|
3096
|
+
"""Declare an artifact as an output of a run.
|
3097
|
+
|
3098
|
+
Arguments:
|
3099
|
+
artifact_or_path: (str or Artifact) A path to the contents of this artifact,
|
3100
|
+
can be in the following forms:
|
3101
|
+
- `/local/directory`
|
3102
|
+
- `/local/directory/file.txt`
|
3103
|
+
- `s3://bucket/path`
|
3104
|
+
You can also pass an Artifact object created by calling
|
3105
|
+
`wandb.Artifact`.
|
3106
|
+
name: (str, optional) An artifact name. Valid names can be in the following forms:
|
3107
|
+
- name:version
|
3108
|
+
- name:alias
|
3109
|
+
- digest
|
3110
|
+
This will default to the basename of the path prepended with the current
|
3111
|
+
run id if not specified.
|
3112
|
+
type: (str) The type of artifact to log, examples include `dataset`, `model`
|
3113
|
+
aliases: (list, optional) Aliases to apply to this artifact,
|
3114
|
+
defaults to `["latest"]`
|
3115
|
+
tags: (list, optional) Tags to apply to this artifact, if any.
|
3116
|
+
|
3117
|
+
Returns:
|
3118
|
+
An `Artifact` object.
|
3119
|
+
"""
|
3120
|
+
return self._log_artifact(
|
3121
|
+
artifact_or_path,
|
3122
|
+
name=name,
|
3123
|
+
type=type,
|
3124
|
+
aliases=aliases,
|
3125
|
+
tags=tags,
|
3126
|
+
)
|
3127
|
+
|
3128
|
+
@_run_decorator._noop_on_finish()
|
3129
|
+
@_run_decorator._attach
|
3130
|
+
def upsert_artifact(
|
3131
|
+
self,
|
3132
|
+
artifact_or_path: Union[Artifact, str],
|
3133
|
+
name: Optional[str] = None,
|
3134
|
+
type: Optional[str] = None,
|
3135
|
+
aliases: Optional[List[str]] = None,
|
3136
|
+
distributed_id: Optional[str] = None,
|
3137
|
+
) -> Artifact:
|
3138
|
+
"""Declare (or append to) a non-finalized artifact as output of a run.
|
3139
|
+
|
3140
|
+
Note that you must call run.finish_artifact() to finalize the artifact.
|
3141
|
+
This is useful when distributed jobs need to all contribute to the same artifact.
|
3142
|
+
|
3143
|
+
Arguments:
|
3144
|
+
artifact_or_path: (str or Artifact) A path to the contents of this artifact,
|
3145
|
+
can be in the following forms:
|
3146
|
+
- `/local/directory`
|
3147
|
+
- `/local/directory/file.txt`
|
3148
|
+
- `s3://bucket/path`
|
3149
|
+
You can also pass an Artifact object created by calling
|
3150
|
+
`wandb.Artifact`.
|
3151
|
+
name: (str, optional) An artifact name. May be prefixed with entity/project.
|
3152
|
+
Valid names can be in the following forms:
|
3153
|
+
- name:version
|
3154
|
+
- name:alias
|
3155
|
+
- digest
|
3156
|
+
This will default to the basename of the path prepended with the current
|
3157
|
+
run id if not specified.
|
3158
|
+
type: (str) The type of artifact to log, examples include `dataset`, `model`
|
3159
|
+
aliases: (list, optional) Aliases to apply to this artifact,
|
3160
|
+
defaults to `["latest"]`
|
3161
|
+
distributed_id: (string, optional) Unique string that all distributed jobs share. If None,
|
3162
|
+
defaults to the run's group name.
|
3163
|
+
|
3164
|
+
Returns:
|
3165
|
+
An `Artifact` object.
|
3166
|
+
"""
|
3167
|
+
if self._get_group() == "" and distributed_id is None:
|
3168
|
+
raise TypeError(
|
3169
|
+
"Cannot upsert artifact unless run is in a group or distributed_id is provided"
|
3170
|
+
)
|
3171
|
+
if distributed_id is None:
|
3172
|
+
distributed_id = self._get_group()
|
3173
|
+
return self._log_artifact(
|
3174
|
+
artifact_or_path,
|
3175
|
+
name=name,
|
3176
|
+
type=type,
|
3177
|
+
aliases=aliases,
|
3178
|
+
distributed_id=distributed_id,
|
3179
|
+
finalize=False,
|
3180
|
+
)
|
3181
|
+
|
3182
|
+
@_run_decorator._noop_on_finish()
|
3183
|
+
@_run_decorator._attach
|
3184
|
+
def finish_artifact(
|
3185
|
+
self,
|
3186
|
+
artifact_or_path: Union[Artifact, str],
|
3187
|
+
name: Optional[str] = None,
|
3188
|
+
type: Optional[str] = None,
|
3189
|
+
aliases: Optional[List[str]] = None,
|
3190
|
+
distributed_id: Optional[str] = None,
|
3191
|
+
) -> Artifact:
|
3192
|
+
"""Finishes a non-finalized artifact as output of a run.
|
3193
|
+
|
3194
|
+
Subsequent "upserts" with the same distributed ID will result in a new version.
|
3195
|
+
|
3196
|
+
Arguments:
|
3197
|
+
artifact_or_path: (str or Artifact) A path to the contents of this artifact,
|
3198
|
+
can be in the following forms:
|
3199
|
+
- `/local/directory`
|
3200
|
+
- `/local/directory/file.txt`
|
3201
|
+
- `s3://bucket/path`
|
3202
|
+
You can also pass an Artifact object created by calling
|
3203
|
+
`wandb.Artifact`.
|
3204
|
+
name: (str, optional) An artifact name. May be prefixed with entity/project.
|
3205
|
+
Valid names can be in the following forms:
|
3206
|
+
- name:version
|
3207
|
+
- name:alias
|
3208
|
+
- digest
|
3209
|
+
This will default to the basename of the path prepended with the current
|
3210
|
+
run id if not specified.
|
3211
|
+
type: (str) The type of artifact to log, examples include `dataset`, `model`
|
3212
|
+
aliases: (list, optional) Aliases to apply to this artifact,
|
3213
|
+
defaults to `["latest"]`
|
3214
|
+
distributed_id: (string, optional) Unique string that all distributed jobs share. If None,
|
3215
|
+
defaults to the run's group name.
|
3216
|
+
|
3217
|
+
Returns:
|
3218
|
+
An `Artifact` object.
|
3219
|
+
"""
|
3220
|
+
if self._get_group() == "" and distributed_id is None:
|
3221
|
+
raise TypeError(
|
3222
|
+
"Cannot finish artifact unless run is in a group or distributed_id is provided"
|
3223
|
+
)
|
3224
|
+
if distributed_id is None:
|
3225
|
+
distributed_id = self._get_group()
|
3226
|
+
|
3227
|
+
return self._log_artifact(
|
3228
|
+
artifact_or_path,
|
3229
|
+
name,
|
3230
|
+
type,
|
3231
|
+
aliases,
|
3232
|
+
distributed_id=distributed_id,
|
3233
|
+
finalize=True,
|
3234
|
+
)
|
3235
|
+
|
3236
|
+
def _log_artifact(
|
3237
|
+
self,
|
3238
|
+
artifact_or_path: Union[Artifact, StrPath],
|
3239
|
+
name: Optional[str] = None,
|
3240
|
+
type: Optional[str] = None,
|
3241
|
+
aliases: Optional[List[str]] = None,
|
3242
|
+
tags: Optional[List[str]] = None,
|
3243
|
+
distributed_id: Optional[str] = None,
|
3244
|
+
finalize: bool = True,
|
3245
|
+
is_user_created: bool = False,
|
3246
|
+
use_after_commit: bool = False,
|
3247
|
+
) -> Artifact:
|
3248
|
+
api = internal.Api()
|
3249
|
+
if api.settings().get("anonymous") == "true":
|
3250
|
+
wandb.termwarn(
|
3251
|
+
"Artifacts logged anonymously cannot be claimed and expire after 7 days."
|
3252
|
+
)
|
3253
|
+
|
3254
|
+
if not finalize and distributed_id is None:
|
3255
|
+
raise TypeError("Must provide distributed_id if artifact is not finalize")
|
3256
|
+
|
3257
|
+
if aliases is not None:
|
3258
|
+
aliases = validate_aliases(aliases)
|
3259
|
+
|
3260
|
+
# Check if artifact tags are supported
|
3261
|
+
if tags is not None:
|
3262
|
+
tags = validate_tags(tags)
|
3263
|
+
|
3264
|
+
artifact, aliases = self._prepare_artifact(
|
3265
|
+
artifact_or_path, name, type, aliases
|
3266
|
+
)
|
3267
|
+
artifact.distributed_id = distributed_id
|
3268
|
+
self._assert_can_log_artifact(artifact)
|
3269
|
+
if self._backend and self._backend.interface:
|
3270
|
+
if not self._settings._offline:
|
3271
|
+
future = self._backend.interface.communicate_artifact(
|
3272
|
+
self,
|
3273
|
+
artifact,
|
3274
|
+
aliases,
|
3275
|
+
tags,
|
3276
|
+
self.step,
|
3277
|
+
finalize=finalize,
|
3278
|
+
is_user_created=is_user_created,
|
3279
|
+
use_after_commit=use_after_commit,
|
3280
|
+
)
|
3281
|
+
artifact._set_save_future(future, self._public_api().client)
|
3282
|
+
else:
|
3283
|
+
self._backend.interface.publish_artifact(
|
3284
|
+
self,
|
3285
|
+
artifact,
|
3286
|
+
aliases,
|
3287
|
+
tags,
|
3288
|
+
finalize=finalize,
|
3289
|
+
is_user_created=is_user_created,
|
3290
|
+
use_after_commit=use_after_commit,
|
3291
|
+
)
|
3292
|
+
elif self._internal_run_interface:
|
3293
|
+
self._internal_run_interface.publish_artifact(
|
3294
|
+
self,
|
3295
|
+
artifact,
|
3296
|
+
aliases,
|
3297
|
+
tags,
|
3298
|
+
finalize=finalize,
|
3299
|
+
is_user_created=is_user_created,
|
3300
|
+
use_after_commit=use_after_commit,
|
3301
|
+
)
|
3302
|
+
return artifact
|
3303
|
+
|
3304
|
+
def _public_api(self, overrides: Optional[Dict[str, str]] = None) -> PublicApi:
|
3305
|
+
overrides = {"run": self._run_id}
|
3306
|
+
if not (self._settings._offline or self._run_obj is None):
|
3307
|
+
overrides["entity"] = self._run_obj.entity
|
3308
|
+
overrides["project"] = self._run_obj.project
|
3309
|
+
return public.Api(overrides)
|
3310
|
+
|
3311
|
+
# TODO(jhr): annotate this
|
3312
|
+
def _assert_can_log_artifact(self, artifact) -> None: # type: ignore
|
3313
|
+
if self._settings._offline:
|
3314
|
+
return
|
3315
|
+
try:
|
3316
|
+
public_api = self._public_api()
|
3317
|
+
entity = public_api.settings["entity"]
|
3318
|
+
project = public_api.settings["project"]
|
3319
|
+
expected_type = Artifact._expected_type(
|
3320
|
+
entity, project, artifact.name, public_api.client
|
3321
|
+
)
|
3322
|
+
except requests.exceptions.RequestException:
|
3323
|
+
# Just return early if there is a network error. This is
|
3324
|
+
# ok, as this function is intended to help catch an invalid
|
3325
|
+
# type early, but not a hard requirement for valid operation.
|
3326
|
+
return
|
3327
|
+
if expected_type is not None and artifact.type != expected_type:
|
3328
|
+
raise ValueError(
|
3329
|
+
f"Artifact {artifact.name} already exists with type '{expected_type}'; "
|
3330
|
+
f"cannot create another with type '{artifact.type}'"
|
3331
|
+
)
|
3332
|
+
if entity and artifact._source_entity and entity != artifact._source_entity:
|
3333
|
+
raise ValueError(
|
3334
|
+
f"Artifact {artifact.name} is owned by entity "
|
3335
|
+
f"'{artifact._source_entity}'; it can't be moved to '{entity}'"
|
3336
|
+
)
|
3337
|
+
if project and artifact._source_project and project != artifact._source_project:
|
3338
|
+
raise ValueError(
|
3339
|
+
f"Artifact {artifact.name} exists in project "
|
3340
|
+
f"'{artifact._source_project}'; it can't be moved to '{project}'"
|
3341
|
+
)
|
3342
|
+
|
3343
|
+
def _prepare_artifact(
|
3344
|
+
self,
|
3345
|
+
artifact_or_path: Union[Artifact, StrPath],
|
3346
|
+
name: Optional[str] = None,
|
3347
|
+
type: Optional[str] = None,
|
3348
|
+
aliases: Optional[List[str]] = None,
|
3349
|
+
) -> Tuple[Artifact, List[str]]:
|
3350
|
+
if isinstance(artifact_or_path, (str, os.PathLike)):
|
3351
|
+
name = name or f"run-{self._run_id}-{os.path.basename(artifact_or_path)}"
|
3352
|
+
artifact = wandb.Artifact(name, type or "unspecified")
|
3353
|
+
if os.path.isfile(artifact_or_path):
|
3354
|
+
artifact.add_file(str(artifact_or_path))
|
3355
|
+
elif os.path.isdir(artifact_or_path):
|
3356
|
+
artifact.add_dir(str(artifact_or_path))
|
3357
|
+
elif "://" in str(artifact_or_path):
|
3358
|
+
artifact.add_reference(str(artifact_or_path))
|
3359
|
+
else:
|
3360
|
+
raise ValueError(
|
3361
|
+
"path must be a file, directory or external"
|
3362
|
+
"reference like s3://bucket/path"
|
3363
|
+
)
|
3364
|
+
else:
|
3365
|
+
artifact = artifact_or_path
|
3366
|
+
if not isinstance(artifact, wandb.Artifact):
|
3367
|
+
raise ValueError(
|
3368
|
+
"You must pass an instance of wandb.Artifact or a "
|
3369
|
+
"valid file path to log_artifact"
|
3370
|
+
)
|
3371
|
+
|
3372
|
+
artifact.finalize()
|
3373
|
+
return artifact, _resolve_aliases(aliases)
|
3374
|
+
|
3375
|
+
@_run_decorator._noop_on_finish()
|
3376
|
+
@_run_decorator._attach
|
3377
|
+
def log_model(
|
3378
|
+
self,
|
3379
|
+
path: StrPath,
|
3380
|
+
name: Optional[str] = None,
|
3381
|
+
aliases: Optional[List[str]] = None,
|
3382
|
+
) -> None:
|
3383
|
+
"""Logs a model artifact containing the contents inside the 'path' to a run and marks it as an output to this run.
|
3384
|
+
|
3385
|
+
Arguments:
|
3386
|
+
path: (str) A path to the contents of this model,
|
3387
|
+
can be in the following forms:
|
3388
|
+
- `/local/directory`
|
3389
|
+
- `/local/directory/file.txt`
|
3390
|
+
- `s3://bucket/path`
|
3391
|
+
name: (str, optional) A name to assign to the model artifact that the file contents will be added to.
|
3392
|
+
The string must contain only the following alphanumeric characters: dashes, underscores, and dots.
|
3393
|
+
This will default to the basename of the path prepended with the current
|
3394
|
+
run id if not specified.
|
3395
|
+
aliases: (list, optional) Aliases to apply to the created model artifact,
|
3396
|
+
defaults to `["latest"]`
|
3397
|
+
|
3398
|
+
Examples:
|
3399
|
+
```python
|
3400
|
+
run.log_model(
|
3401
|
+
path="/local/directory",
|
3402
|
+
name="my_model_artifact",
|
3403
|
+
aliases=["production"],
|
3404
|
+
)
|
3405
|
+
```
|
3406
|
+
|
3407
|
+
Invalid usage
|
3408
|
+
```python
|
3409
|
+
run.log_model(
|
3410
|
+
path="/local/directory",
|
3411
|
+
name="my_entity/my_project/my_model_artifact",
|
3412
|
+
aliases=["production"],
|
3413
|
+
)
|
3414
|
+
```
|
3415
|
+
|
3416
|
+
Raises:
|
3417
|
+
ValueError: if name has invalid special characters
|
3418
|
+
|
3419
|
+
Returns:
|
3420
|
+
None
|
3421
|
+
"""
|
3422
|
+
self._log_artifact(
|
3423
|
+
artifact_or_path=path, name=name, type="model", aliases=aliases
|
3424
|
+
)
|
3425
|
+
|
3426
|
+
@_run_decorator._noop_on_finish()
|
3427
|
+
@_run_decorator._attach
|
3428
|
+
def use_model(self, name: str) -> FilePathStr:
|
3429
|
+
"""Download the files logged in a model artifact 'name'.
|
3430
|
+
|
3431
|
+
Arguments:
|
3432
|
+
name: (str) A model artifact name. 'name' must match the name of an existing logged
|
3433
|
+
model artifact.
|
3434
|
+
May be prefixed with entity/project/. Valid names
|
3435
|
+
can be in the following forms:
|
3436
|
+
- model_artifact_name:version
|
3437
|
+
- model_artifact_name:alias
|
3438
|
+
|
3439
|
+
Examples:
|
3440
|
+
```python
|
3441
|
+
run.use_model(
|
3442
|
+
name="my_model_artifact:latest",
|
3443
|
+
)
|
3444
|
+
|
3445
|
+
run.use_model(
|
3446
|
+
name="my_project/my_model_artifact:v0",
|
3447
|
+
)
|
3448
|
+
|
3449
|
+
run.use_model(
|
3450
|
+
name="my_entity/my_project/my_model_artifact:<digest>",
|
3451
|
+
)
|
3452
|
+
```
|
3453
|
+
|
3454
|
+
Invalid usage
|
3455
|
+
```python
|
3456
|
+
run.use_model(
|
3457
|
+
name="my_entity/my_project/my_model_artifact",
|
3458
|
+
)
|
3459
|
+
```
|
3460
|
+
|
3461
|
+
Raises:
|
3462
|
+
AssertionError: if model artifact 'name' is of a type that does not contain the substring 'model'.
|
3463
|
+
|
3464
|
+
Returns:
|
3465
|
+
path: (str) path to downloaded model artifact file(s).
|
3466
|
+
"""
|
3467
|
+
artifact = self.use_artifact(artifact_or_name=name)
|
3468
|
+
assert (
|
3469
|
+
"model" in str(artifact.type.lower())
|
3470
|
+
), "You can only use this method for 'model' artifacts. For an artifact to be a 'model' artifact, its type property must contain the substring 'model'."
|
3471
|
+
path = artifact.download()
|
3472
|
+
|
3473
|
+
# If returned directory contains only one file, return path to that file
|
3474
|
+
dir_list = os.listdir(path)
|
3475
|
+
if len(dir_list) == 1:
|
3476
|
+
return FilePathStr(os.path.join(path, dir_list[0]))
|
3477
|
+
return path
|
3478
|
+
|
3479
|
+
@_run_decorator._noop_on_finish()
|
3480
|
+
@_run_decorator._attach
|
3481
|
+
def link_model(
|
3482
|
+
self,
|
3483
|
+
path: StrPath,
|
3484
|
+
registered_model_name: str,
|
3485
|
+
name: Optional[str] = None,
|
3486
|
+
aliases: Optional[List[str]] = None,
|
3487
|
+
) -> None:
|
3488
|
+
"""Log a model artifact version and link it to a registered model in the model registry.
|
3489
|
+
|
3490
|
+
The linked model version will be visible in the UI for the specified registered model.
|
3491
|
+
|
3492
|
+
Steps:
|
3493
|
+
- Check if 'name' model artifact has been logged. If so, use the artifact version that matches the files
|
3494
|
+
located at 'path' or log a new version. Otherwise log files under 'path' as a new model artifact, 'name'
|
3495
|
+
of type 'model'.
|
3496
|
+
- Check if registered model with name 'registered_model_name' exists in the 'model-registry' project.
|
3497
|
+
If not, create a new registered model with name 'registered_model_name'.
|
3498
|
+
- Link version of model artifact 'name' to registered model, 'registered_model_name'.
|
3499
|
+
- Attach aliases from 'aliases' list to the newly linked model artifact version.
|
3500
|
+
|
3501
|
+
Arguments:
|
3502
|
+
path: (str) A path to the contents of this model,
|
3503
|
+
can be in the following forms:
|
3504
|
+
- `/local/directory`
|
3505
|
+
- `/local/directory/file.txt`
|
3506
|
+
- `s3://bucket/path`
|
3507
|
+
registered_model_name: (str) - the name of the registered model that the model is to be linked to.
|
3508
|
+
A registered model is a collection of model versions linked to the model registry, typically representing a
|
3509
|
+
team's specific ML Task. The entity that this registered model belongs to will be derived from the run
|
3510
|
+
name: (str, optional) - the name of the model artifact that files in 'path' will be logged to. This will
|
3511
|
+
default to the basename of the path prepended with the current run id if not specified.
|
3512
|
+
aliases: (List[str], optional) - alias(es) that will only be applied on this linked artifact
|
3513
|
+
inside the registered model.
|
3514
|
+
The alias "latest" will always be applied to the latest version of an artifact that is linked.
|
3515
|
+
|
3516
|
+
Examples:
|
3517
|
+
```python
|
3518
|
+
run.link_model(
|
3519
|
+
path="/local/directory",
|
3520
|
+
registered_model_name="my_reg_model",
|
3521
|
+
name="my_model_artifact",
|
3522
|
+
aliases=["production"],
|
3523
|
+
)
|
3524
|
+
```
|
3525
|
+
|
3526
|
+
Invalid usage
|
3527
|
+
```python
|
3528
|
+
run.link_model(
|
3529
|
+
path="/local/directory",
|
3530
|
+
registered_model_name="my_entity/my_project/my_reg_model",
|
3531
|
+
name="my_model_artifact",
|
3532
|
+
aliases=["production"],
|
3533
|
+
)
|
3534
|
+
|
3535
|
+
run.link_model(
|
3536
|
+
path="/local/directory",
|
3537
|
+
registered_model_name="my_reg_model",
|
3538
|
+
name="my_entity/my_project/my_model_artifact",
|
3539
|
+
aliases=["production"],
|
3540
|
+
)
|
3541
|
+
```
|
3542
|
+
|
3543
|
+
Raises:
|
3544
|
+
AssertionError: if registered_model_name is a path or
|
3545
|
+
if model artifact 'name' is of a type that does not contain the substring 'model'
|
3546
|
+
ValueError: if name has invalid special characters
|
3547
|
+
|
3548
|
+
Returns:
|
3549
|
+
None
|
3550
|
+
"""
|
3551
|
+
name_parts = registered_model_name.split("/")
|
3552
|
+
assert (
|
3553
|
+
len(name_parts) == 1
|
3554
|
+
), "Please provide only the name of the registered model. Do not append the entity or project name."
|
3555
|
+
project = "model-registry"
|
3556
|
+
target_path = self.entity + "/" + project + "/" + registered_model_name
|
3557
|
+
|
3558
|
+
public_api = self._public_api()
|
3559
|
+
try:
|
3560
|
+
artifact = public_api.artifact(name=f"{name}:latest")
|
3561
|
+
assert (
|
3562
|
+
"model" in str(artifact.type.lower())
|
3563
|
+
), "You can only use this method for 'model' artifacts. For an artifact to be a 'model' artifact, its type property must contain the substring 'model'."
|
3564
|
+
artifact = self._log_artifact(
|
3565
|
+
artifact_or_path=path, name=name, type=artifact.type
|
3566
|
+
)
|
3567
|
+
except (ValueError, CommError):
|
3568
|
+
artifact = self._log_artifact(
|
3569
|
+
artifact_or_path=path, name=name, type="model"
|
3570
|
+
)
|
3571
|
+
self.link_artifact(artifact=artifact, target_path=target_path, aliases=aliases)
|
3572
|
+
|
3573
|
+
@_run_decorator._noop_on_finish()
|
3574
|
+
@_run_decorator._attach
|
3575
|
+
def alert(
|
3576
|
+
self,
|
3577
|
+
title: str,
|
3578
|
+
text: str,
|
3579
|
+
level: Optional[Union[str, "AlertLevel"]] = None,
|
3580
|
+
wait_duration: Union[int, float, timedelta, None] = None,
|
3581
|
+
) -> None:
|
3582
|
+
"""Launch an alert with the given title and text.
|
3583
|
+
|
3584
|
+
Arguments:
|
3585
|
+
title: (str) The title of the alert, must be less than 64 characters long.
|
3586
|
+
text: (str) The text body of the alert.
|
3587
|
+
level: (str or AlertLevel, optional) The alert level to use, either: `INFO`, `WARN`, or `ERROR`.
|
3588
|
+
wait_duration: (int, float, or timedelta, optional) The time to wait (in seconds) before sending another
|
3589
|
+
alert with this title.
|
3590
|
+
"""
|
3591
|
+
level = level or AlertLevel.INFO
|
3592
|
+
level_str: str = level.value if isinstance(level, AlertLevel) else level
|
3593
|
+
if level_str not in {lev.value for lev in AlertLevel}:
|
3594
|
+
raise ValueError("level must be one of 'INFO', 'WARN', or 'ERROR'")
|
3595
|
+
|
3596
|
+
wait_duration = wait_duration or timedelta(minutes=1)
|
3597
|
+
if isinstance(wait_duration, int) or isinstance(wait_duration, float):
|
3598
|
+
wait_duration = timedelta(seconds=wait_duration)
|
3599
|
+
elif not callable(getattr(wait_duration, "total_seconds", None)):
|
3600
|
+
raise ValueError(
|
3601
|
+
"wait_duration must be an int, float, or datetime.timedelta"
|
3602
|
+
)
|
3603
|
+
wait_duration = int(wait_duration.total_seconds() * 1000)
|
3604
|
+
|
3605
|
+
if self._backend and self._backend.interface:
|
3606
|
+
self._backend.interface.publish_alert(title, text, level_str, wait_duration)
|
3607
|
+
|
3608
|
+
def __enter__(self) -> "Run":
|
3609
|
+
return self
|
3610
|
+
|
3611
|
+
def __exit__(
|
3612
|
+
self,
|
3613
|
+
exc_type: Type[BaseException],
|
3614
|
+
exc_val: BaseException,
|
3615
|
+
exc_tb: TracebackType,
|
3616
|
+
) -> bool:
|
3617
|
+
exception_raised = exc_type is not None
|
3618
|
+
if exception_raised:
|
3619
|
+
traceback.print_exception(exc_type, exc_val, exc_tb)
|
3620
|
+
exit_code = 1 if exception_raised else 0
|
3621
|
+
self._finish(exit_code=exit_code)
|
3622
|
+
return not exception_raised
|
3623
|
+
|
3624
|
+
@_run_decorator._noop_on_finish()
|
3625
|
+
@_run_decorator._attach
|
3626
|
+
def mark_preempting(self) -> None:
|
3627
|
+
"""Mark this run as preempting.
|
3628
|
+
|
3629
|
+
Also tells the internal process to immediately report this to server.
|
3630
|
+
"""
|
3631
|
+
if self._backend and self._backend.interface:
|
3632
|
+
self._backend.interface.publish_preempting()
|
3633
|
+
|
3634
|
+
@property
|
3635
|
+
@_run_decorator._noop_on_finish()
|
3636
|
+
@_run_decorator._attach
|
3637
|
+
def _system_metrics(self) -> Dict[str, List[Tuple[datetime, float]]]:
|
3638
|
+
"""Returns a dictionary of system metrics.
|
3639
|
+
|
3640
|
+
Returns:
|
3641
|
+
A dictionary of system metrics.
|
3642
|
+
"""
|
3643
|
+
|
3644
|
+
def pb_to_dict(
|
3645
|
+
system_metrics_pb: wandb.proto.wandb_internal_pb2.GetSystemMetricsResponse,
|
3646
|
+
) -> Dict[str, List[Tuple[datetime, float]]]:
|
3647
|
+
res = {}
|
3648
|
+
|
3649
|
+
for metric, records in system_metrics_pb.system_metrics.items():
|
3650
|
+
measurements = []
|
3651
|
+
for record in records.record:
|
3652
|
+
# Convert timestamp to datetime
|
3653
|
+
dt = datetime.fromtimestamp(
|
3654
|
+
record.timestamp.seconds, tz=timezone.utc
|
3655
|
+
)
|
3656
|
+
dt = dt.replace(microsecond=record.timestamp.nanos // 1000)
|
3657
|
+
|
3658
|
+
measurements.append((dt, record.value))
|
3659
|
+
|
3660
|
+
res[metric] = measurements
|
3661
|
+
|
3662
|
+
return res
|
3663
|
+
|
3664
|
+
if not self._backend or not self._backend.interface:
|
3665
|
+
return {}
|
3666
|
+
|
3667
|
+
handle = self._backend.interface.deliver_get_system_metrics()
|
3668
|
+
result = handle.wait(timeout=1)
|
3669
|
+
|
3670
|
+
if result:
|
3671
|
+
try:
|
3672
|
+
response = result.response.get_system_metrics_response
|
3673
|
+
if response:
|
3674
|
+
return pb_to_dict(response)
|
3675
|
+
except Exception as e:
|
3676
|
+
logger.error("Error getting system metrics: %s", e)
|
3677
|
+
return {}
|
3678
|
+
|
3679
|
+
# ------------------------------------------------------------------------------
|
3680
|
+
# HEADER
|
3681
|
+
# ------------------------------------------------------------------------------
|
3682
|
+
# Note: All the header methods are static methods since we want to share the printing logic
|
3683
|
+
# with the service execution path that doesn't have access to the run instance
|
3684
|
+
@staticmethod
|
3685
|
+
def _header(
|
3686
|
+
*,
|
3687
|
+
settings: "Settings",
|
3688
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3689
|
+
) -> None:
|
3690
|
+
Run._header_wandb_version_info(settings=settings, printer=printer)
|
3691
|
+
Run._header_sync_info(settings=settings, printer=printer)
|
3692
|
+
Run._header_run_info(settings=settings, printer=printer)
|
3693
|
+
|
3694
|
+
@staticmethod
|
3695
|
+
def _header_wandb_version_info(
|
3696
|
+
*,
|
3697
|
+
settings: "Settings",
|
3698
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3699
|
+
) -> None:
|
3700
|
+
if settings.quiet or settings.silent:
|
3701
|
+
return
|
3702
|
+
|
3703
|
+
# TODO: add this to a higher verbosity level
|
3704
|
+
printer.display(
|
3705
|
+
f"Tracking run with wandb version {wandb.__version__}", off=False
|
3706
|
+
)
|
3707
|
+
|
3708
|
+
@staticmethod
|
3709
|
+
def _header_sync_info(
|
3710
|
+
*,
|
3711
|
+
settings: "Settings",
|
3712
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3713
|
+
) -> None:
|
3714
|
+
if settings._offline:
|
3715
|
+
printer.display(
|
3716
|
+
[
|
3717
|
+
f"W&B syncing is set to {printer.code('`offline`')} in this directory. ",
|
3718
|
+
f"Run {printer.code('`wandb online`')} or set {printer.code('WANDB_MODE=online')} "
|
3719
|
+
"to enable cloud syncing.",
|
3720
|
+
]
|
3721
|
+
)
|
3722
|
+
else:
|
3723
|
+
info = [f"Run data is saved locally in {printer.files(settings.sync_dir)}"]
|
3724
|
+
if not printer._html:
|
3725
|
+
info.append(
|
3726
|
+
f"Run {printer.code('`wandb offline`')} to turn off syncing."
|
3727
|
+
)
|
3728
|
+
printer.display(info, off=settings.quiet or settings.silent)
|
3729
|
+
|
3730
|
+
@staticmethod
|
3731
|
+
def _header_run_info(
|
3732
|
+
*,
|
3733
|
+
settings: "Settings",
|
3734
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3735
|
+
) -> None:
|
3736
|
+
if settings._offline or settings.silent:
|
3737
|
+
return
|
3738
|
+
|
3739
|
+
run_url = settings.run_url
|
3740
|
+
project_url = settings.project_url
|
3741
|
+
sweep_url = settings.sweep_url
|
3742
|
+
|
3743
|
+
run_state_str = (
|
3744
|
+
"Resuming run"
|
3745
|
+
if settings.resumed or settings.resume_from
|
3746
|
+
else "Syncing run"
|
3747
|
+
)
|
3748
|
+
run_name = settings.run_name
|
3749
|
+
if not run_name:
|
3750
|
+
return
|
3751
|
+
|
3752
|
+
if printer._html:
|
3753
|
+
if not wandb.jupyter.maybe_display(): # type: ignore
|
3754
|
+
run_line = f"<strong>{printer.link(run_url, run_name)}</strong>"
|
3755
|
+
project_line, sweep_line = "", ""
|
3756
|
+
|
3757
|
+
# TODO(settings): make settings the source of truth
|
3758
|
+
if not wandb.jupyter.quiet(): # type: ignore
|
3759
|
+
doc_html = printer.link(wburls.get("doc_run"), "docs")
|
3760
|
+
|
3761
|
+
project_html = printer.link(project_url, "Weights & Biases")
|
3762
|
+
project_line = f"to {project_html} ({doc_html})"
|
3763
|
+
|
3764
|
+
if sweep_url:
|
3765
|
+
sweep_line = f"Sweep page: {printer.link(sweep_url, sweep_url)}"
|
3766
|
+
|
3767
|
+
printer.display(
|
3768
|
+
[f"{run_state_str} {run_line} {project_line}", sweep_line],
|
3769
|
+
)
|
3770
|
+
|
3771
|
+
else:
|
3772
|
+
printer.display(
|
3773
|
+
f"{run_state_str} {printer.name(run_name)}", off=not run_name
|
3774
|
+
)
|
3775
|
+
|
3776
|
+
if not settings.quiet:
|
3777
|
+
# TODO: add verbosity levels and add this to higher levels
|
3778
|
+
printer.display(
|
3779
|
+
f'{printer.emoji("star")} View project at {printer.link(project_url)}'
|
3780
|
+
)
|
3781
|
+
if sweep_url:
|
3782
|
+
printer.display(
|
3783
|
+
f'{printer.emoji("broom")} View sweep at {printer.link(sweep_url)}'
|
3784
|
+
)
|
3785
|
+
printer.display(
|
3786
|
+
f'{printer.emoji("rocket")} View run at {printer.link(run_url)}',
|
3787
|
+
)
|
3788
|
+
|
3789
|
+
# TODO(settings) use `wandb_settings` (if self.settings.anonymous == "true":)
|
3790
|
+
if Api().api.settings().get("anonymous") == "true":
|
3791
|
+
printer.display(
|
3792
|
+
"Do NOT share these links with anyone. They can be used to claim your runs.",
|
3793
|
+
level="warn",
|
3794
|
+
off=not run_name,
|
3795
|
+
)
|
3796
|
+
|
3797
|
+
# ------------------------------------------------------------------------------
|
3798
|
+
# FOOTER
|
3799
|
+
# ------------------------------------------------------------------------------
|
3800
|
+
# Note: All the footer methods are static methods since we want to share the printing logic
|
3801
|
+
# with the service execution path that doesn't have access to the run instance
|
3802
|
+
@staticmethod
|
3803
|
+
def _footer(
|
3804
|
+
sampled_history: Optional["SampledHistoryResponse"] = None,
|
3805
|
+
final_summary: Optional["GetSummaryResponse"] = None,
|
3806
|
+
poll_exit_response: Optional[PollExitResponse] = None,
|
3807
|
+
internal_messages_response: Optional["InternalMessagesResponse"] = None,
|
3808
|
+
reporter: Optional[Reporter] = None,
|
3809
|
+
quiet: Optional[bool] = None,
|
3810
|
+
*,
|
3811
|
+
settings: "Settings",
|
3812
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3813
|
+
) -> None:
|
3814
|
+
Run._footer_history_summary_info(
|
3815
|
+
history=sampled_history,
|
3816
|
+
summary=final_summary,
|
3817
|
+
quiet=quiet,
|
3818
|
+
settings=settings,
|
3819
|
+
printer=printer,
|
3820
|
+
)
|
3821
|
+
|
3822
|
+
Run._footer_sync_info(
|
3823
|
+
poll_exit_response=poll_exit_response,
|
3824
|
+
quiet=quiet,
|
3825
|
+
settings=settings,
|
3826
|
+
printer=printer,
|
3827
|
+
)
|
3828
|
+
Run._footer_log_dir_info(quiet=quiet, settings=settings, printer=printer)
|
3829
|
+
Run._footer_notify_wandb_core(
|
3830
|
+
quiet=quiet,
|
3831
|
+
settings=settings,
|
3832
|
+
printer=printer,
|
3833
|
+
)
|
3834
|
+
Run._footer_internal_messages(
|
3835
|
+
internal_messages_response=internal_messages_response,
|
3836
|
+
quiet=quiet,
|
3837
|
+
settings=settings,
|
3838
|
+
printer=printer,
|
3839
|
+
)
|
3840
|
+
Run._footer_reporter_warn_err(
|
3841
|
+
reporter=reporter, quiet=quiet, settings=settings, printer=printer
|
3842
|
+
)
|
3843
|
+
|
3844
|
+
# fixme: Temporary hack until we move to rich which allows multiple spinners
|
3845
|
+
@staticmethod
|
3846
|
+
def _footer_file_pusher_status_info(
|
3847
|
+
poll_exit_responses: Optional[
|
3848
|
+
Union[PollExitResponse, List[Optional[PollExitResponse]]]
|
3849
|
+
] = None,
|
3850
|
+
*,
|
3851
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3852
|
+
) -> None:
|
3853
|
+
if not poll_exit_responses:
|
3854
|
+
return
|
3855
|
+
if isinstance(poll_exit_responses, PollExitResponse):
|
3856
|
+
Run._footer_single_run_file_pusher_status_info(
|
3857
|
+
poll_exit_responses, printer=printer
|
3858
|
+
)
|
3859
|
+
elif isinstance(poll_exit_responses, list):
|
3860
|
+
poll_exit_responses_list = poll_exit_responses
|
3861
|
+
assert all(
|
3862
|
+
response is None or isinstance(response, PollExitResponse)
|
3863
|
+
for response in poll_exit_responses_list
|
3864
|
+
)
|
3865
|
+
if len(poll_exit_responses_list) == 0:
|
3866
|
+
return
|
3867
|
+
elif len(poll_exit_responses_list) == 1:
|
3868
|
+
Run._footer_single_run_file_pusher_status_info(
|
3869
|
+
poll_exit_responses_list[0], printer=printer
|
3870
|
+
)
|
3871
|
+
else:
|
3872
|
+
Run._footer_multiple_runs_file_pusher_status_info(
|
3873
|
+
poll_exit_responses_list, printer=printer
|
3874
|
+
)
|
3875
|
+
else:
|
3876
|
+
logger.error(
|
3877
|
+
f"Got the type `{type(poll_exit_responses)}` for `poll_exit_responses`. "
|
3878
|
+
"Expected either None, PollExitResponse or a List[Union[PollExitResponse, None]]"
|
3879
|
+
)
|
3880
|
+
|
3881
|
+
@staticmethod
|
3882
|
+
def _footer_single_run_file_pusher_status_info(
|
3883
|
+
poll_exit_response: Optional[PollExitResponse] = None,
|
3884
|
+
*,
|
3885
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3886
|
+
) -> None:
|
3887
|
+
# todo: is this same as settings._offline?
|
3888
|
+
if not poll_exit_response:
|
3889
|
+
return
|
3890
|
+
|
3891
|
+
stats = poll_exit_response.pusher_stats
|
3892
|
+
|
3893
|
+
megabyte = wandb.util.POW_2_BYTES[2][1]
|
3894
|
+
line = (
|
3895
|
+
f"{stats.uploaded_bytes / megabyte:.3f} MB"
|
3896
|
+
f" of {stats.total_bytes / megabyte:.3f} MB uploaded"
|
3897
|
+
)
|
3898
|
+
if stats.deduped_bytes > 0:
|
3899
|
+
line += f" ({stats.deduped_bytes / megabyte:.3f} MB deduped)"
|
3900
|
+
line += "\r"
|
3901
|
+
|
3902
|
+
if stats.total_bytes > 0:
|
3903
|
+
printer.progress_update(line, stats.uploaded_bytes / stats.total_bytes)
|
3904
|
+
else:
|
3905
|
+
printer.progress_update(line, 1.0)
|
3906
|
+
|
3907
|
+
if poll_exit_response.done:
|
3908
|
+
printer.progress_close()
|
3909
|
+
|
3910
|
+
if stats.total_bytes > 0:
|
3911
|
+
dedupe_fraction = stats.deduped_bytes / float(stats.total_bytes)
|
3912
|
+
else:
|
3913
|
+
dedupe_fraction = 0
|
3914
|
+
|
3915
|
+
if stats.deduped_bytes > 0.01:
|
3916
|
+
printer.display(
|
3917
|
+
f"W&B sync reduced upload amount by {dedupe_fraction:.1%}"
|
3918
|
+
)
|
3919
|
+
|
3920
|
+
@staticmethod
|
3921
|
+
def _footer_multiple_runs_file_pusher_status_info(
|
3922
|
+
poll_exit_responses: List[Optional[PollExitResponse]],
|
3923
|
+
*,
|
3924
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3925
|
+
) -> None:
|
3926
|
+
# todo: is this same as settings._offline?
|
3927
|
+
if not all(poll_exit_responses):
|
3928
|
+
return
|
3929
|
+
|
3930
|
+
megabyte = wandb.util.POW_2_BYTES[2][1]
|
3931
|
+
total_files: int = sum(
|
3932
|
+
sum(
|
3933
|
+
[
|
3934
|
+
response.file_counts.wandb_count,
|
3935
|
+
response.file_counts.media_count,
|
3936
|
+
response.file_counts.artifact_count,
|
3937
|
+
response.file_counts.other_count,
|
3938
|
+
]
|
3939
|
+
)
|
3940
|
+
for response in poll_exit_responses
|
3941
|
+
if response is not None and response.file_counts is not None
|
3942
|
+
)
|
3943
|
+
uploaded = sum(
|
3944
|
+
response.pusher_stats.uploaded_bytes
|
3945
|
+
for response in poll_exit_responses
|
3946
|
+
if response is not None and response.pusher_stats is not None
|
3947
|
+
)
|
3948
|
+
total = sum(
|
3949
|
+
response.pusher_stats.total_bytes
|
3950
|
+
for response in poll_exit_responses
|
3951
|
+
if response is not None and response.pusher_stats is not None
|
3952
|
+
)
|
3953
|
+
|
3954
|
+
line = (
|
3955
|
+
f"Processing {len(poll_exit_responses)} runs with {total_files} files "
|
3956
|
+
f"({uploaded/megabyte :.2f} MB/{total/megabyte :.2f} MB)\r"
|
3957
|
+
)
|
3958
|
+
# line = "{}{:<{max_len}}\r".format(line, " ", max_len=(80 - len(line)))
|
3959
|
+
printer.progress_update(line) # type:ignore[call-arg]
|
3960
|
+
|
3961
|
+
done = all(
|
3962
|
+
[
|
3963
|
+
poll_exit_response.done
|
3964
|
+
for poll_exit_response in poll_exit_responses
|
3965
|
+
if poll_exit_response
|
3966
|
+
]
|
3967
|
+
)
|
3968
|
+
if done:
|
3969
|
+
printer.progress_close()
|
3970
|
+
|
3971
|
+
@staticmethod
|
3972
|
+
def _footer_sync_info(
|
3973
|
+
poll_exit_response: Optional[PollExitResponse] = None,
|
3974
|
+
quiet: Optional[bool] = None,
|
3975
|
+
*,
|
3976
|
+
settings: "Settings",
|
3977
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
3978
|
+
) -> None:
|
3979
|
+
if settings.silent:
|
3980
|
+
return
|
3981
|
+
|
3982
|
+
if settings._offline:
|
3983
|
+
printer.display(
|
3984
|
+
[
|
3985
|
+
"You can sync this run to the cloud by running:",
|
3986
|
+
printer.code(f"wandb sync {settings.sync_dir}"),
|
3987
|
+
],
|
3988
|
+
off=(quiet or settings.quiet),
|
3989
|
+
)
|
3990
|
+
else:
|
3991
|
+
info = []
|
3992
|
+
if settings.run_name and settings.run_url:
|
3993
|
+
info.append(
|
3994
|
+
f"{printer.emoji('rocket')} View run {printer.name(settings.run_name)} at: {printer.link(settings.run_url)}"
|
3995
|
+
)
|
3996
|
+
if settings.project_url:
|
3997
|
+
info.append(
|
3998
|
+
f"{printer.emoji('star')} View project at: {printer.link(settings.project_url)}"
|
3999
|
+
)
|
4000
|
+
if poll_exit_response and poll_exit_response.file_counts:
|
4001
|
+
logger.info("logging synced files")
|
4002
|
+
file_counts = poll_exit_response.file_counts
|
4003
|
+
info.append(
|
4004
|
+
f"Synced {file_counts.wandb_count} W&B file(s), {file_counts.media_count} media file(s), "
|
4005
|
+
f"{file_counts.artifact_count} artifact file(s) and {file_counts.other_count} other file(s)",
|
4006
|
+
)
|
4007
|
+
printer.display(info)
|
4008
|
+
|
4009
|
+
@staticmethod
|
4010
|
+
def _footer_log_dir_info(
|
4011
|
+
quiet: Optional[bool] = None,
|
4012
|
+
*,
|
4013
|
+
settings: "Settings",
|
4014
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
4015
|
+
) -> None:
|
4016
|
+
if (quiet or settings.quiet) or settings.silent:
|
4017
|
+
return
|
4018
|
+
|
4019
|
+
log_dir = settings.log_user or settings.log_internal
|
4020
|
+
if log_dir:
|
4021
|
+
log_dir = os.path.dirname(log_dir.replace(os.getcwd(), "."))
|
4022
|
+
printer.display(
|
4023
|
+
f"Find logs at: {printer.files(log_dir)}",
|
4024
|
+
)
|
4025
|
+
|
4026
|
+
@staticmethod
|
4027
|
+
def _footer_history_summary_info(
|
4028
|
+
history: Optional["SampledHistoryResponse"] = None,
|
4029
|
+
summary: Optional["GetSummaryResponse"] = None,
|
4030
|
+
quiet: Optional[bool] = None,
|
4031
|
+
*,
|
4032
|
+
settings: "Settings",
|
4033
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
4034
|
+
) -> None:
|
4035
|
+
if (quiet or settings.quiet) or settings.silent:
|
4036
|
+
return
|
4037
|
+
|
4038
|
+
panel = []
|
4039
|
+
|
4040
|
+
# Render history if available
|
4041
|
+
if history:
|
4042
|
+
logger.info("rendering history")
|
4043
|
+
|
4044
|
+
sampled_history = {
|
4045
|
+
item.key: wandb.util.downsample(
|
4046
|
+
item.values_float or item.values_int, 40
|
4047
|
+
)
|
4048
|
+
for item in history.item
|
4049
|
+
if not item.key.startswith("_")
|
4050
|
+
}
|
4051
|
+
|
4052
|
+
history_rows = []
|
4053
|
+
for key, values in sorted(sampled_history.items()):
|
4054
|
+
if any(not isinstance(value, numbers.Number) for value in values):
|
4055
|
+
continue
|
4056
|
+
sparkline = printer.sparklines(values)
|
4057
|
+
if sparkline:
|
4058
|
+
history_rows.append([key, sparkline])
|
4059
|
+
if history_rows:
|
4060
|
+
history_grid = printer.grid(
|
4061
|
+
history_rows,
|
4062
|
+
"Run history:",
|
4063
|
+
)
|
4064
|
+
panel.append(history_grid)
|
4065
|
+
|
4066
|
+
# Render summary if available
|
4067
|
+
if summary:
|
4068
|
+
final_summary = {}
|
4069
|
+
for item in summary.item:
|
4070
|
+
if item.key.startswith("_") or len(item.nested_key) > 0:
|
4071
|
+
continue
|
4072
|
+
final_summary[item.key] = json.loads(item.value_json)
|
4073
|
+
|
4074
|
+
logger.info("rendering summary")
|
4075
|
+
summary_rows = []
|
4076
|
+
for key, value in sorted(final_summary.items()):
|
4077
|
+
# arrays etc. might be too large. for now, we just don't print them
|
4078
|
+
if isinstance(value, str):
|
4079
|
+
value = value[:20] + "..." * (len(value) >= 20)
|
4080
|
+
summary_rows.append([key, value])
|
4081
|
+
elif isinstance(value, numbers.Number):
|
4082
|
+
value = round(value, 5) if isinstance(value, float) else value
|
4083
|
+
summary_rows.append([key, str(value)])
|
4084
|
+
else:
|
4085
|
+
continue
|
4086
|
+
|
4087
|
+
if summary_rows:
|
4088
|
+
summary_grid = printer.grid(
|
4089
|
+
summary_rows,
|
4090
|
+
"Run summary:",
|
4091
|
+
)
|
4092
|
+
panel.append(summary_grid)
|
4093
|
+
|
4094
|
+
if panel:
|
4095
|
+
printer.display(printer.panel(panel))
|
4096
|
+
|
4097
|
+
@staticmethod
|
4098
|
+
def _footer_internal_messages(
|
4099
|
+
internal_messages_response: Optional["InternalMessagesResponse"] = None,
|
4100
|
+
quiet: Optional[bool] = None,
|
4101
|
+
*,
|
4102
|
+
settings: "Settings",
|
4103
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
4104
|
+
) -> None:
|
4105
|
+
if (quiet or settings.quiet) or settings.silent:
|
4106
|
+
return
|
4107
|
+
|
4108
|
+
if not internal_messages_response:
|
4109
|
+
return
|
4110
|
+
|
4111
|
+
for message in internal_messages_response.messages.warning:
|
4112
|
+
printer.display(message, level="warn")
|
4113
|
+
|
4114
|
+
@staticmethod
|
4115
|
+
def _footer_notify_wandb_core(
|
4116
|
+
*,
|
4117
|
+
quiet: Optional[bool] = None,
|
4118
|
+
settings: "Settings",
|
4119
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
4120
|
+
) -> None:
|
4121
|
+
"""Prints a message advertising the upcoming core release."""
|
4122
|
+
if quiet or not settings._require_legacy_service:
|
4123
|
+
return
|
4124
|
+
|
4125
|
+
printer.display(
|
4126
|
+
"The legacy backend is deprecated. In future versions, `wandb-core` will become "
|
4127
|
+
"the sole backend service, and the `wandb.require('legacy-service')` flag will be removed. "
|
4128
|
+
"For more information, visit https://wandb.me/wandb-core",
|
4129
|
+
level="warn",
|
4130
|
+
)
|
4131
|
+
|
4132
|
+
@staticmethod
|
4133
|
+
def _footer_reporter_warn_err(
|
4134
|
+
reporter: Optional[Reporter] = None,
|
4135
|
+
quiet: Optional[bool] = None,
|
4136
|
+
*,
|
4137
|
+
settings: "Settings",
|
4138
|
+
printer: Union["PrinterTerm", "PrinterJupyter"],
|
4139
|
+
) -> None:
|
4140
|
+
if (quiet or settings.quiet) or settings.silent:
|
4141
|
+
return
|
4142
|
+
|
4143
|
+
if not reporter:
|
4144
|
+
return
|
4145
|
+
|
4146
|
+
warning_lines = reporter.warning_lines
|
4147
|
+
if warning_lines:
|
4148
|
+
warnings = ["Warnings:"] + [f"{line}" for line in warning_lines]
|
4149
|
+
if len(warning_lines) < reporter.warning_count:
|
4150
|
+
warnings.append("More warnings...")
|
4151
|
+
printer.display(warnings)
|
4152
|
+
|
4153
|
+
error_lines = reporter.error_lines
|
4154
|
+
if error_lines:
|
4155
|
+
errors = ["Errors:"] + [f"{line}" for line in error_lines]
|
4156
|
+
if len(error_lines) < reporter.error_count:
|
4157
|
+
errors.append("More errors...")
|
4158
|
+
printer.display(errors)
|
4159
|
+
|
4160
|
+
|
4161
|
+
# We define this outside of the run context to support restoring before init
|
4162
|
+
def restore(
|
4163
|
+
name: str,
|
4164
|
+
run_path: Optional[str] = None,
|
4165
|
+
replace: bool = False,
|
4166
|
+
root: Optional[str] = None,
|
4167
|
+
) -> Union[None, TextIO]:
|
4168
|
+
"""Download the specified file from cloud storage.
|
4169
|
+
|
4170
|
+
File is placed into the current directory or run directory.
|
4171
|
+
By default, will only download the file if it doesn't already exist.
|
4172
|
+
|
4173
|
+
Arguments:
|
4174
|
+
name: the name of the file
|
4175
|
+
run_path: optional path to a run to pull files from, i.e. `username/project_name/run_id`
|
4176
|
+
if wandb.init has not been called, this is required.
|
4177
|
+
replace: whether to download the file even if it already exists locally
|
4178
|
+
root: the directory to download the file to. Defaults to the current
|
4179
|
+
directory or the run directory if wandb.init was called.
|
4180
|
+
|
4181
|
+
Returns:
|
4182
|
+
None if it can't find the file, otherwise a file object open for reading
|
4183
|
+
|
4184
|
+
Raises:
|
4185
|
+
wandb.CommError: if we can't connect to the wandb backend
|
4186
|
+
ValueError: if the file is not found or can't find run_path
|
4187
|
+
"""
|
4188
|
+
is_disabled = wandb.run is not None and wandb.run.disabled
|
4189
|
+
run = None if is_disabled else wandb.run
|
4190
|
+
if run_path is None:
|
4191
|
+
if run is not None:
|
4192
|
+
run_path = run.path
|
4193
|
+
else:
|
4194
|
+
raise ValueError(
|
4195
|
+
"run_path required when calling wandb.restore before wandb.init"
|
4196
|
+
)
|
4197
|
+
if root is None:
|
4198
|
+
if run is not None:
|
4199
|
+
root = run.dir
|
4200
|
+
api = public.Api()
|
4201
|
+
api_run = api.run(run_path)
|
4202
|
+
if root is None:
|
4203
|
+
root = os.getcwd()
|
4204
|
+
path = os.path.join(root, name)
|
4205
|
+
if os.path.exists(path) and replace is False:
|
4206
|
+
return open(path)
|
4207
|
+
if is_disabled:
|
4208
|
+
return None
|
4209
|
+
files = api_run.files([name])
|
4210
|
+
if len(files) == 0:
|
4211
|
+
return None
|
4212
|
+
# if the file does not exist, the file has an md5 of 0
|
4213
|
+
if files[0].md5 == "0":
|
4214
|
+
raise ValueError(f"File {name} not found in {run_path or root}.")
|
4215
|
+
return files[0].download(root=root, replace=True)
|
4216
|
+
|
4217
|
+
|
4218
|
+
# propagate our doc string to the runs restore method
|
4219
|
+
try:
|
4220
|
+
Run.restore.__doc__ = restore.__doc__
|
4221
|
+
except AttributeError:
|
4222
|
+
pass
|
4223
|
+
|
4224
|
+
|
4225
|
+
def finish(exit_code: Optional[int] = None, quiet: Optional[bool] = None) -> None:
|
4226
|
+
"""Mark a run as finished, and finish uploading all data.
|
4227
|
+
|
4228
|
+
This is used when creating multiple runs in the same process.
|
4229
|
+
We automatically call this method when your script exits.
|
4230
|
+
|
4231
|
+
Arguments:
|
4232
|
+
exit_code: Set to something other than 0 to mark a run as failed
|
4233
|
+
quiet: Set to true to minimize log output
|
4234
|
+
"""
|
4235
|
+
if wandb.run:
|
4236
|
+
wandb.run.finish(exit_code=exit_code, quiet=quiet)
|