wandb 0.19.1__py3-none-musllinux_1_2_aarch64.whl

Sign up to get free protection for your applications and to get access to all the features.
Files changed (822) hide show
  1. package_readme.md +97 -0
  2. wandb/__init__.py +246 -0
  3. wandb/__init__.pyi +1197 -0
  4. wandb/__main__.py +3 -0
  5. wandb/_globals.py +19 -0
  6. wandb/agents/__init__.py +0 -0
  7. wandb/agents/pyagent.py +363 -0
  8. wandb/analytics/__init__.py +3 -0
  9. wandb/analytics/sentry.py +263 -0
  10. wandb/apis/__init__.py +48 -0
  11. wandb/apis/attrs.py +51 -0
  12. wandb/apis/importers/__init__.py +1 -0
  13. wandb/apis/importers/internals/internal.py +385 -0
  14. wandb/apis/importers/internals/protocols.py +103 -0
  15. wandb/apis/importers/internals/util.py +78 -0
  16. wandb/apis/importers/mlflow.py +254 -0
  17. wandb/apis/importers/validation.py +108 -0
  18. wandb/apis/importers/wandb.py +1603 -0
  19. wandb/apis/internal.py +232 -0
  20. wandb/apis/normalize.py +73 -0
  21. wandb/apis/paginator.py +81 -0
  22. wandb/apis/public/__init__.py +34 -0
  23. wandb/apis/public/api.py +1387 -0
  24. wandb/apis/public/artifacts.py +1095 -0
  25. wandb/apis/public/const.py +4 -0
  26. wandb/apis/public/files.py +263 -0
  27. wandb/apis/public/history.py +149 -0
  28. wandb/apis/public/jobs.py +653 -0
  29. wandb/apis/public/projects.py +154 -0
  30. wandb/apis/public/query_generator.py +166 -0
  31. wandb/apis/public/reports.py +458 -0
  32. wandb/apis/public/runs.py +1012 -0
  33. wandb/apis/public/sweeps.py +240 -0
  34. wandb/apis/public/teams.py +198 -0
  35. wandb/apis/public/users.py +136 -0
  36. wandb/apis/public/utils.py +68 -0
  37. wandb/apis/reports/__init__.py +1 -0
  38. wandb/apis/reports/v1/__init__.py +8 -0
  39. wandb/apis/reports/v2/__init__.py +8 -0
  40. wandb/apis/workspaces/__init__.py +8 -0
  41. wandb/beta/workflows.py +288 -0
  42. wandb/bin/gpu_stats +0 -0
  43. wandb/bin/wandb-core +0 -0
  44. wandb/cli/__init__.py +0 -0
  45. wandb/cli/beta.py +178 -0
  46. wandb/cli/cli.py +2812 -0
  47. wandb/data_types.py +66 -0
  48. wandb/docker/__init__.py +343 -0
  49. wandb/docker/auth.py +435 -0
  50. wandb/docker/wandb-entrypoint.sh +33 -0
  51. wandb/docker/www_authenticate.py +94 -0
  52. wandb/env.py +513 -0
  53. wandb/errors/__init__.py +17 -0
  54. wandb/errors/errors.py +37 -0
  55. wandb/errors/links.py +73 -0
  56. wandb/errors/term.py +415 -0
  57. wandb/errors/util.py +57 -0
  58. wandb/errors/warnings.py +2 -0
  59. wandb/filesync/__init__.py +0 -0
  60. wandb/filesync/dir_watcher.py +403 -0
  61. wandb/filesync/stats.py +100 -0
  62. wandb/filesync/step_checksum.py +142 -0
  63. wandb/filesync/step_prepare.py +179 -0
  64. wandb/filesync/step_upload.py +287 -0
  65. wandb/filesync/upload_job.py +142 -0
  66. wandb/integration/__init__.py +0 -0
  67. wandb/integration/catboost/__init__.py +5 -0
  68. wandb/integration/catboost/catboost.py +178 -0
  69. wandb/integration/cohere/__init__.py +3 -0
  70. wandb/integration/cohere/cohere.py +21 -0
  71. wandb/integration/cohere/resolver.py +347 -0
  72. wandb/integration/diffusers/__init__.py +3 -0
  73. wandb/integration/diffusers/autologger.py +76 -0
  74. wandb/integration/diffusers/pipeline_resolver.py +50 -0
  75. wandb/integration/diffusers/resolvers/__init__.py +9 -0
  76. wandb/integration/diffusers/resolvers/multimodal.py +882 -0
  77. wandb/integration/diffusers/resolvers/utils.py +102 -0
  78. wandb/integration/fastai/__init__.py +245 -0
  79. wandb/integration/gym/__init__.py +99 -0
  80. wandb/integration/huggingface/__init__.py +3 -0
  81. wandb/integration/huggingface/huggingface.py +18 -0
  82. wandb/integration/huggingface/resolver.py +213 -0
  83. wandb/integration/keras/__init__.py +11 -0
  84. wandb/integration/keras/callbacks/__init__.py +5 -0
  85. wandb/integration/keras/callbacks/metrics_logger.py +129 -0
  86. wandb/integration/keras/callbacks/model_checkpoint.py +188 -0
  87. wandb/integration/keras/callbacks/tables_builder.py +228 -0
  88. wandb/integration/keras/keras.py +1089 -0
  89. wandb/integration/kfp/__init__.py +6 -0
  90. wandb/integration/kfp/helpers.py +28 -0
  91. wandb/integration/kfp/kfp_patch.py +334 -0
  92. wandb/integration/kfp/wandb_logging.py +182 -0
  93. wandb/integration/langchain/__init__.py +3 -0
  94. wandb/integration/langchain/wandb_tracer.py +48 -0
  95. wandb/integration/lightgbm/__init__.py +239 -0
  96. wandb/integration/lightning/__init__.py +0 -0
  97. wandb/integration/lightning/fabric/__init__.py +3 -0
  98. wandb/integration/lightning/fabric/logger.py +764 -0
  99. wandb/integration/metaflow/__init__.py +3 -0
  100. wandb/integration/metaflow/metaflow.py +383 -0
  101. wandb/integration/openai/__init__.py +3 -0
  102. wandb/integration/openai/fine_tuning.py +480 -0
  103. wandb/integration/openai/openai.py +22 -0
  104. wandb/integration/openai/resolver.py +240 -0
  105. wandb/integration/prodigy/__init__.py +3 -0
  106. wandb/integration/prodigy/prodigy.py +299 -0
  107. wandb/integration/sacred/__init__.py +117 -0
  108. wandb/integration/sagemaker/__init__.py +12 -0
  109. wandb/integration/sagemaker/auth.py +28 -0
  110. wandb/integration/sagemaker/config.py +49 -0
  111. wandb/integration/sagemaker/files.py +3 -0
  112. wandb/integration/sagemaker/resources.py +34 -0
  113. wandb/integration/sb3/__init__.py +3 -0
  114. wandb/integration/sb3/sb3.py +147 -0
  115. wandb/integration/sklearn/__init__.py +37 -0
  116. wandb/integration/sklearn/calculate/__init__.py +32 -0
  117. wandb/integration/sklearn/calculate/calibration_curves.py +125 -0
  118. wandb/integration/sklearn/calculate/class_proportions.py +68 -0
  119. wandb/integration/sklearn/calculate/confusion_matrix.py +93 -0
  120. wandb/integration/sklearn/calculate/decision_boundaries.py +40 -0
  121. wandb/integration/sklearn/calculate/elbow_curve.py +55 -0
  122. wandb/integration/sklearn/calculate/feature_importances.py +67 -0
  123. wandb/integration/sklearn/calculate/learning_curve.py +64 -0
  124. wandb/integration/sklearn/calculate/outlier_candidates.py +69 -0
  125. wandb/integration/sklearn/calculate/residuals.py +86 -0
  126. wandb/integration/sklearn/calculate/silhouette.py +118 -0
  127. wandb/integration/sklearn/calculate/summary_metrics.py +62 -0
  128. wandb/integration/sklearn/plot/__init__.py +35 -0
  129. wandb/integration/sklearn/plot/classifier.py +329 -0
  130. wandb/integration/sklearn/plot/clusterer.py +146 -0
  131. wandb/integration/sklearn/plot/regressor.py +121 -0
  132. wandb/integration/sklearn/plot/shared.py +91 -0
  133. wandb/integration/sklearn/utils.py +183 -0
  134. wandb/integration/tensorboard/__init__.py +10 -0
  135. wandb/integration/tensorboard/log.py +354 -0
  136. wandb/integration/tensorboard/monkeypatch.py +186 -0
  137. wandb/integration/tensorflow/__init__.py +5 -0
  138. wandb/integration/tensorflow/estimator_hook.py +54 -0
  139. wandb/integration/torch/__init__.py +0 -0
  140. wandb/integration/torch/wandb_torch.py +554 -0
  141. wandb/integration/ultralytics/__init__.py +11 -0
  142. wandb/integration/ultralytics/bbox_utils.py +215 -0
  143. wandb/integration/ultralytics/callback.py +524 -0
  144. wandb/integration/ultralytics/classification_utils.py +83 -0
  145. wandb/integration/ultralytics/mask_utils.py +202 -0
  146. wandb/integration/ultralytics/pose_utils.py +103 -0
  147. wandb/integration/xgboost/__init__.py +11 -0
  148. wandb/integration/xgboost/xgboost.py +189 -0
  149. wandb/integration/yolov8/__init__.py +0 -0
  150. wandb/integration/yolov8/yolov8.py +284 -0
  151. wandb/jupyter.py +513 -0
  152. wandb/mpmain/__init__.py +0 -0
  153. wandb/mpmain/__main__.py +1 -0
  154. wandb/old/__init__.py +0 -0
  155. wandb/old/core.py +53 -0
  156. wandb/old/settings.py +173 -0
  157. wandb/old/summary.py +440 -0
  158. wandb/plot/__init__.py +28 -0
  159. wandb/plot/bar.py +70 -0
  160. wandb/plot/confusion_matrix.py +181 -0
  161. wandb/plot/custom_chart.py +124 -0
  162. wandb/plot/histogram.py +65 -0
  163. wandb/plot/line.py +74 -0
  164. wandb/plot/line_series.py +176 -0
  165. wandb/plot/pr_curve.py +185 -0
  166. wandb/plot/roc_curve.py +163 -0
  167. wandb/plot/scatter.py +66 -0
  168. wandb/plot/utils.py +183 -0
  169. wandb/plot/viz.py +41 -0
  170. wandb/proto/__init__.py +0 -0
  171. wandb/proto/v3/__init__.py +0 -0
  172. wandb/proto/v3/wandb_base_pb2.py +55 -0
  173. wandb/proto/v3/wandb_internal_pb2.py +1658 -0
  174. wandb/proto/v3/wandb_server_pb2.py +228 -0
  175. wandb/proto/v3/wandb_settings_pb2.py +122 -0
  176. wandb/proto/v3/wandb_telemetry_pb2.py +106 -0
  177. wandb/proto/v4/__init__.py +0 -0
  178. wandb/proto/v4/wandb_base_pb2.py +30 -0
  179. wandb/proto/v4/wandb_internal_pb2.py +370 -0
  180. wandb/proto/v4/wandb_server_pb2.py +67 -0
  181. wandb/proto/v4/wandb_settings_pb2.py +47 -0
  182. wandb/proto/v4/wandb_telemetry_pb2.py +41 -0
  183. wandb/proto/v5/wandb_base_pb2.py +31 -0
  184. wandb/proto/v5/wandb_internal_pb2.py +371 -0
  185. wandb/proto/v5/wandb_server_pb2.py +68 -0
  186. wandb/proto/v5/wandb_settings_pb2.py +48 -0
  187. wandb/proto/v5/wandb_telemetry_pb2.py +42 -0
  188. wandb/proto/wandb_base_pb2.py +10 -0
  189. wandb/proto/wandb_deprecated.py +45 -0
  190. wandb/proto/wandb_generate_deprecated.py +30 -0
  191. wandb/proto/wandb_generate_proto.py +49 -0
  192. wandb/proto/wandb_internal_pb2.py +16 -0
  193. wandb/proto/wandb_server_pb2.py +10 -0
  194. wandb/proto/wandb_settings_pb2.py +10 -0
  195. wandb/proto/wandb_telemetry_pb2.py +10 -0
  196. wandb/py.typed +0 -0
  197. wandb/sdk/__init__.py +37 -0
  198. wandb/sdk/artifacts/__init__.py +0 -0
  199. wandb/sdk/artifacts/_validators.py +121 -0
  200. wandb/sdk/artifacts/artifact.py +2364 -0
  201. wandb/sdk/artifacts/artifact_download_logger.py +43 -0
  202. wandb/sdk/artifacts/artifact_file_cache.py +249 -0
  203. wandb/sdk/artifacts/artifact_instance_cache.py +17 -0
  204. wandb/sdk/artifacts/artifact_manifest.py +75 -0
  205. wandb/sdk/artifacts/artifact_manifest_entry.py +249 -0
  206. wandb/sdk/artifacts/artifact_manifests/__init__.py +0 -0
  207. wandb/sdk/artifacts/artifact_manifests/artifact_manifest_v1.py +92 -0
  208. wandb/sdk/artifacts/artifact_saver.py +265 -0
  209. wandb/sdk/artifacts/artifact_state.py +11 -0
  210. wandb/sdk/artifacts/artifact_ttl.py +7 -0
  211. wandb/sdk/artifacts/exceptions.py +57 -0
  212. wandb/sdk/artifacts/staging.py +25 -0
  213. wandb/sdk/artifacts/storage_handler.py +62 -0
  214. wandb/sdk/artifacts/storage_handlers/__init__.py +0 -0
  215. wandb/sdk/artifacts/storage_handlers/azure_handler.py +213 -0
  216. wandb/sdk/artifacts/storage_handlers/gcs_handler.py +224 -0
  217. wandb/sdk/artifacts/storage_handlers/http_handler.py +114 -0
  218. wandb/sdk/artifacts/storage_handlers/local_file_handler.py +139 -0
  219. wandb/sdk/artifacts/storage_handlers/multi_handler.py +56 -0
  220. wandb/sdk/artifacts/storage_handlers/s3_handler.py +298 -0
  221. wandb/sdk/artifacts/storage_handlers/tracking_handler.py +72 -0
  222. wandb/sdk/artifacts/storage_handlers/wb_artifact_handler.py +135 -0
  223. wandb/sdk/artifacts/storage_handlers/wb_local_artifact_handler.py +74 -0
  224. wandb/sdk/artifacts/storage_layout.py +6 -0
  225. wandb/sdk/artifacts/storage_policies/__init__.py +4 -0
  226. wandb/sdk/artifacts/storage_policies/register.py +1 -0
  227. wandb/sdk/artifacts/storage_policies/wandb_storage_policy.py +378 -0
  228. wandb/sdk/artifacts/storage_policy.py +72 -0
  229. wandb/sdk/backend/__init__.py +0 -0
  230. wandb/sdk/backend/backend.py +221 -0
  231. wandb/sdk/data_types/__init__.py +0 -0
  232. wandb/sdk/data_types/_dtypes.py +918 -0
  233. wandb/sdk/data_types/_private.py +10 -0
  234. wandb/sdk/data_types/audio.py +165 -0
  235. wandb/sdk/data_types/base_types/__init__.py +0 -0
  236. wandb/sdk/data_types/base_types/json_metadata.py +55 -0
  237. wandb/sdk/data_types/base_types/media.py +376 -0
  238. wandb/sdk/data_types/base_types/wb_value.py +282 -0
  239. wandb/sdk/data_types/bokeh.py +70 -0
  240. wandb/sdk/data_types/graph.py +405 -0
  241. wandb/sdk/data_types/helper_types/__init__.py +0 -0
  242. wandb/sdk/data_types/helper_types/bounding_boxes_2d.py +305 -0
  243. wandb/sdk/data_types/helper_types/classes.py +159 -0
  244. wandb/sdk/data_types/helper_types/image_mask.py +241 -0
  245. wandb/sdk/data_types/histogram.py +94 -0
  246. wandb/sdk/data_types/html.py +115 -0
  247. wandb/sdk/data_types/image.py +847 -0
  248. wandb/sdk/data_types/molecule.py +241 -0
  249. wandb/sdk/data_types/object_3d.py +470 -0
  250. wandb/sdk/data_types/plotly.py +82 -0
  251. wandb/sdk/data_types/saved_model.py +445 -0
  252. wandb/sdk/data_types/table.py +1204 -0
  253. wandb/sdk/data_types/trace_tree.py +438 -0
  254. wandb/sdk/data_types/utils.py +228 -0
  255. wandb/sdk/data_types/video.py +268 -0
  256. wandb/sdk/integration_utils/__init__.py +0 -0
  257. wandb/sdk/integration_utils/auto_logging.py +232 -0
  258. wandb/sdk/integration_utils/data_logging.py +475 -0
  259. wandb/sdk/interface/__init__.py +0 -0
  260. wandb/sdk/interface/constants.py +4 -0
  261. wandb/sdk/interface/interface.py +1010 -0
  262. wandb/sdk/interface/interface_queue.py +53 -0
  263. wandb/sdk/interface/interface_relay.py +53 -0
  264. wandb/sdk/interface/interface_shared.py +546 -0
  265. wandb/sdk/interface/interface_sock.py +61 -0
  266. wandb/sdk/interface/message_future.py +27 -0
  267. wandb/sdk/interface/message_future_poll.py +50 -0
  268. wandb/sdk/interface/router.py +115 -0
  269. wandb/sdk/interface/router_queue.py +41 -0
  270. wandb/sdk/interface/router_relay.py +37 -0
  271. wandb/sdk/interface/router_sock.py +36 -0
  272. wandb/sdk/interface/summary_record.py +67 -0
  273. wandb/sdk/internal/__init__.py +0 -0
  274. wandb/sdk/internal/context.py +89 -0
  275. wandb/sdk/internal/datastore.py +297 -0
  276. wandb/sdk/internal/file_pusher.py +181 -0
  277. wandb/sdk/internal/file_stream.py +695 -0
  278. wandb/sdk/internal/flow_control.py +263 -0
  279. wandb/sdk/internal/handler.py +905 -0
  280. wandb/sdk/internal/internal.py +403 -0
  281. wandb/sdk/internal/internal_api.py +4587 -0
  282. wandb/sdk/internal/internal_util.py +97 -0
  283. wandb/sdk/internal/job_builder.py +638 -0
  284. wandb/sdk/internal/profiler.py +78 -0
  285. wandb/sdk/internal/progress.py +79 -0
  286. wandb/sdk/internal/run.py +25 -0
  287. wandb/sdk/internal/sample.py +70 -0
  288. wandb/sdk/internal/sender.py +1696 -0
  289. wandb/sdk/internal/sender_config.py +197 -0
  290. wandb/sdk/internal/settings_static.py +97 -0
  291. wandb/sdk/internal/system/__init__.py +0 -0
  292. wandb/sdk/internal/system/assets/__init__.py +25 -0
  293. wandb/sdk/internal/system/assets/aggregators.py +37 -0
  294. wandb/sdk/internal/system/assets/asset_registry.py +20 -0
  295. wandb/sdk/internal/system/assets/cpu.py +163 -0
  296. wandb/sdk/internal/system/assets/disk.py +210 -0
  297. wandb/sdk/internal/system/assets/gpu.py +416 -0
  298. wandb/sdk/internal/system/assets/gpu_amd.py +233 -0
  299. wandb/sdk/internal/system/assets/interfaces.py +205 -0
  300. wandb/sdk/internal/system/assets/ipu.py +177 -0
  301. wandb/sdk/internal/system/assets/memory.py +166 -0
  302. wandb/sdk/internal/system/assets/network.py +125 -0
  303. wandb/sdk/internal/system/assets/open_metrics.py +293 -0
  304. wandb/sdk/internal/system/assets/tpu.py +154 -0
  305. wandb/sdk/internal/system/assets/trainium.py +393 -0
  306. wandb/sdk/internal/system/env_probe_helpers.py +13 -0
  307. wandb/sdk/internal/system/system_info.py +250 -0
  308. wandb/sdk/internal/system/system_monitor.py +222 -0
  309. wandb/sdk/internal/tb_watcher.py +519 -0
  310. wandb/sdk/internal/thread_local_settings.py +18 -0
  311. wandb/sdk/internal/writer.py +204 -0
  312. wandb/sdk/launch/__init__.py +15 -0
  313. wandb/sdk/launch/_launch.py +331 -0
  314. wandb/sdk/launch/_launch_add.py +255 -0
  315. wandb/sdk/launch/_project_spec.py +566 -0
  316. wandb/sdk/launch/agent/__init__.py +5 -0
  317. wandb/sdk/launch/agent/agent.py +924 -0
  318. wandb/sdk/launch/agent/config.py +296 -0
  319. wandb/sdk/launch/agent/job_status_tracker.py +53 -0
  320. wandb/sdk/launch/agent/run_queue_item_file_saver.py +39 -0
  321. wandb/sdk/launch/builder/__init__.py +0 -0
  322. wandb/sdk/launch/builder/abstract.py +156 -0
  323. wandb/sdk/launch/builder/build.py +297 -0
  324. wandb/sdk/launch/builder/context_manager.py +235 -0
  325. wandb/sdk/launch/builder/docker_builder.py +177 -0
  326. wandb/sdk/launch/builder/kaniko_builder.py +594 -0
  327. wandb/sdk/launch/builder/noop.py +58 -0
  328. wandb/sdk/launch/builder/templates/_wandb_bootstrap.py +188 -0
  329. wandb/sdk/launch/builder/templates/dockerfile.py +92 -0
  330. wandb/sdk/launch/create_job.py +528 -0
  331. wandb/sdk/launch/environment/abstract.py +29 -0
  332. wandb/sdk/launch/environment/aws_environment.py +322 -0
  333. wandb/sdk/launch/environment/azure_environment.py +105 -0
  334. wandb/sdk/launch/environment/gcp_environment.py +335 -0
  335. wandb/sdk/launch/environment/local_environment.py +65 -0
  336. wandb/sdk/launch/errors.py +13 -0
  337. wandb/sdk/launch/git_reference.py +109 -0
  338. wandb/sdk/launch/inputs/files.py +148 -0
  339. wandb/sdk/launch/inputs/internal.py +315 -0
  340. wandb/sdk/launch/inputs/manage.py +113 -0
  341. wandb/sdk/launch/inputs/schema.py +39 -0
  342. wandb/sdk/launch/loader.py +249 -0
  343. wandb/sdk/launch/registry/abstract.py +48 -0
  344. wandb/sdk/launch/registry/anon.py +29 -0
  345. wandb/sdk/launch/registry/azure_container_registry.py +124 -0
  346. wandb/sdk/launch/registry/elastic_container_registry.py +192 -0
  347. wandb/sdk/launch/registry/google_artifact_registry.py +219 -0
  348. wandb/sdk/launch/registry/local_registry.py +65 -0
  349. wandb/sdk/launch/runner/__init__.py +0 -0
  350. wandb/sdk/launch/runner/abstract.py +185 -0
  351. wandb/sdk/launch/runner/kubernetes_monitor.py +472 -0
  352. wandb/sdk/launch/runner/kubernetes_runner.py +963 -0
  353. wandb/sdk/launch/runner/local_container.py +301 -0
  354. wandb/sdk/launch/runner/local_process.py +78 -0
  355. wandb/sdk/launch/runner/sagemaker_runner.py +426 -0
  356. wandb/sdk/launch/runner/vertex_runner.py +230 -0
  357. wandb/sdk/launch/sweeps/__init__.py +37 -0
  358. wandb/sdk/launch/sweeps/scheduler.py +740 -0
  359. wandb/sdk/launch/sweeps/scheduler_sweep.py +90 -0
  360. wandb/sdk/launch/sweeps/utils.py +316 -0
  361. wandb/sdk/launch/utils.py +747 -0
  362. wandb/sdk/launch/wandb_reference.py +138 -0
  363. wandb/sdk/lib/__init__.py +5 -0
  364. wandb/sdk/lib/apikey.py +269 -0
  365. wandb/sdk/lib/capped_dict.py +26 -0
  366. wandb/sdk/lib/config_util.py +101 -0
  367. wandb/sdk/lib/credentials.py +141 -0
  368. wandb/sdk/lib/deprecate.py +42 -0
  369. wandb/sdk/lib/disabled.py +29 -0
  370. wandb/sdk/lib/exit_hooks.py +54 -0
  371. wandb/sdk/lib/file_stream_utils.py +118 -0
  372. wandb/sdk/lib/filenames.py +64 -0
  373. wandb/sdk/lib/filesystem.py +372 -0
  374. wandb/sdk/lib/fsm.py +180 -0
  375. wandb/sdk/lib/gitlib.py +239 -0
  376. wandb/sdk/lib/gql_request.py +65 -0
  377. wandb/sdk/lib/handler_util.py +21 -0
  378. wandb/sdk/lib/hashutil.py +84 -0
  379. wandb/sdk/lib/import_hooks.py +275 -0
  380. wandb/sdk/lib/ipython.py +126 -0
  381. wandb/sdk/lib/json_util.py +80 -0
  382. wandb/sdk/lib/lazyloader.py +63 -0
  383. wandb/sdk/lib/mailbox.py +456 -0
  384. wandb/sdk/lib/module.py +78 -0
  385. wandb/sdk/lib/paths.py +106 -0
  386. wandb/sdk/lib/preinit.py +42 -0
  387. wandb/sdk/lib/printer.py +548 -0
  388. wandb/sdk/lib/progress.py +279 -0
  389. wandb/sdk/lib/proto_util.py +90 -0
  390. wandb/sdk/lib/redirect.py +845 -0
  391. wandb/sdk/lib/retry.py +289 -0
  392. wandb/sdk/lib/run_moment.py +72 -0
  393. wandb/sdk/lib/runid.py +12 -0
  394. wandb/sdk/lib/server.py +38 -0
  395. wandb/sdk/lib/service_connection.py +216 -0
  396. wandb/sdk/lib/service_token.py +94 -0
  397. wandb/sdk/lib/sock_client.py +290 -0
  398. wandb/sdk/lib/sparkline.py +44 -0
  399. wandb/sdk/lib/telemetry.py +100 -0
  400. wandb/sdk/lib/timed_input.py +133 -0
  401. wandb/sdk/lib/timer.py +19 -0
  402. wandb/sdk/service/__init__.py +0 -0
  403. wandb/sdk/service/_startup_debug.py +22 -0
  404. wandb/sdk/service/port_file.py +53 -0
  405. wandb/sdk/service/server.py +107 -0
  406. wandb/sdk/service/server_sock.py +274 -0
  407. wandb/sdk/service/service.py +242 -0
  408. wandb/sdk/service/streams.py +425 -0
  409. wandb/sdk/verify/__init__.py +0 -0
  410. wandb/sdk/verify/verify.py +501 -0
  411. wandb/sdk/wandb_alerts.py +12 -0
  412. wandb/sdk/wandb_config.py +322 -0
  413. wandb/sdk/wandb_helper.py +54 -0
  414. wandb/sdk/wandb_init.py +1313 -0
  415. wandb/sdk/wandb_login.py +339 -0
  416. wandb/sdk/wandb_metric.py +110 -0
  417. wandb/sdk/wandb_require.py +94 -0
  418. wandb/sdk/wandb_require_helpers.py +44 -0
  419. wandb/sdk/wandb_run.py +4066 -0
  420. wandb/sdk/wandb_settings.py +1309 -0
  421. wandb/sdk/wandb_setup.py +402 -0
  422. wandb/sdk/wandb_summary.py +150 -0
  423. wandb/sdk/wandb_sweep.py +119 -0
  424. wandb/sdk/wandb_sync.py +82 -0
  425. wandb/sdk/wandb_watch.py +150 -0
  426. wandb/sklearn.py +35 -0
  427. wandb/sync/__init__.py +3 -0
  428. wandb/sync/sync.py +442 -0
  429. wandb/trigger.py +29 -0
  430. wandb/util.py +1955 -0
  431. wandb/vendor/__init__.py +0 -0
  432. wandb/vendor/gql-0.2.0/setup.py +40 -0
  433. wandb/vendor/gql-0.2.0/tests/__init__.py +0 -0
  434. wandb/vendor/gql-0.2.0/tests/starwars/__init__.py +0 -0
  435. wandb/vendor/gql-0.2.0/tests/starwars/fixtures.py +96 -0
  436. wandb/vendor/gql-0.2.0/tests/starwars/schema.py +146 -0
  437. wandb/vendor/gql-0.2.0/tests/starwars/test_dsl.py +293 -0
  438. wandb/vendor/gql-0.2.0/tests/starwars/test_query.py +355 -0
  439. wandb/vendor/gql-0.2.0/tests/starwars/test_validation.py +171 -0
  440. wandb/vendor/gql-0.2.0/tests/test_client.py +31 -0
  441. wandb/vendor/gql-0.2.0/tests/test_transport.py +89 -0
  442. wandb/vendor/gql-0.2.0/wandb_gql/__init__.py +4 -0
  443. wandb/vendor/gql-0.2.0/wandb_gql/client.py +75 -0
  444. wandb/vendor/gql-0.2.0/wandb_gql/dsl.py +152 -0
  445. wandb/vendor/gql-0.2.0/wandb_gql/gql.py +10 -0
  446. wandb/vendor/gql-0.2.0/wandb_gql/transport/__init__.py +0 -0
  447. wandb/vendor/gql-0.2.0/wandb_gql/transport/http.py +6 -0
  448. wandb/vendor/gql-0.2.0/wandb_gql/transport/local_schema.py +15 -0
  449. wandb/vendor/gql-0.2.0/wandb_gql/transport/requests.py +46 -0
  450. wandb/vendor/gql-0.2.0/wandb_gql/utils.py +21 -0
  451. wandb/vendor/graphql-core-1.1/setup.py +86 -0
  452. wandb/vendor/graphql-core-1.1/wandb_graphql/__init__.py +287 -0
  453. wandb/vendor/graphql-core-1.1/wandb_graphql/error/__init__.py +6 -0
  454. wandb/vendor/graphql-core-1.1/wandb_graphql/error/base.py +42 -0
  455. wandb/vendor/graphql-core-1.1/wandb_graphql/error/format_error.py +11 -0
  456. wandb/vendor/graphql-core-1.1/wandb_graphql/error/located_error.py +29 -0
  457. wandb/vendor/graphql-core-1.1/wandb_graphql/error/syntax_error.py +36 -0
  458. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/__init__.py +26 -0
  459. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/base.py +311 -0
  460. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executor.py +398 -0
  461. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/__init__.py +0 -0
  462. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/asyncio.py +53 -0
  463. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/gevent.py +22 -0
  464. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/process.py +32 -0
  465. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/sync.py +7 -0
  466. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/thread.py +35 -0
  467. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/executors/utils.py +6 -0
  468. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/__init__.py +0 -0
  469. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/executor.py +66 -0
  470. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/fragment.py +252 -0
  471. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/resolver.py +151 -0
  472. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/experimental/utils.py +7 -0
  473. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/middleware.py +57 -0
  474. wandb/vendor/graphql-core-1.1/wandb_graphql/execution/values.py +145 -0
  475. wandb/vendor/graphql-core-1.1/wandb_graphql/graphql.py +60 -0
  476. wandb/vendor/graphql-core-1.1/wandb_graphql/language/__init__.py +0 -0
  477. wandb/vendor/graphql-core-1.1/wandb_graphql/language/ast.py +1349 -0
  478. wandb/vendor/graphql-core-1.1/wandb_graphql/language/base.py +19 -0
  479. wandb/vendor/graphql-core-1.1/wandb_graphql/language/lexer.py +435 -0
  480. wandb/vendor/graphql-core-1.1/wandb_graphql/language/location.py +30 -0
  481. wandb/vendor/graphql-core-1.1/wandb_graphql/language/parser.py +779 -0
  482. wandb/vendor/graphql-core-1.1/wandb_graphql/language/printer.py +193 -0
  483. wandb/vendor/graphql-core-1.1/wandb_graphql/language/source.py +18 -0
  484. wandb/vendor/graphql-core-1.1/wandb_graphql/language/visitor.py +222 -0
  485. wandb/vendor/graphql-core-1.1/wandb_graphql/language/visitor_meta.py +82 -0
  486. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/__init__.py +0 -0
  487. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/cached_property.py +17 -0
  488. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/contain_subset.py +28 -0
  489. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/default_ordered_dict.py +40 -0
  490. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/ordereddict.py +8 -0
  491. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/pair_set.py +43 -0
  492. wandb/vendor/graphql-core-1.1/wandb_graphql/pyutils/version.py +78 -0
  493. wandb/vendor/graphql-core-1.1/wandb_graphql/type/__init__.py +67 -0
  494. wandb/vendor/graphql-core-1.1/wandb_graphql/type/definition.py +619 -0
  495. wandb/vendor/graphql-core-1.1/wandb_graphql/type/directives.py +132 -0
  496. wandb/vendor/graphql-core-1.1/wandb_graphql/type/introspection.py +440 -0
  497. wandb/vendor/graphql-core-1.1/wandb_graphql/type/scalars.py +131 -0
  498. wandb/vendor/graphql-core-1.1/wandb_graphql/type/schema.py +100 -0
  499. wandb/vendor/graphql-core-1.1/wandb_graphql/type/typemap.py +145 -0
  500. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/__init__.py +0 -0
  501. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/assert_valid_name.py +9 -0
  502. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_from_value.py +65 -0
  503. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_to_code.py +49 -0
  504. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/ast_to_dict.py +24 -0
  505. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/base.py +75 -0
  506. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/build_ast_schema.py +291 -0
  507. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/build_client_schema.py +250 -0
  508. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/concat_ast.py +9 -0
  509. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/extend_schema.py +357 -0
  510. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/get_field_def.py +27 -0
  511. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/get_operation_ast.py +21 -0
  512. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/introspection_query.py +90 -0
  513. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/is_valid_literal_value.py +67 -0
  514. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/is_valid_value.py +66 -0
  515. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/quoted_or_list.py +21 -0
  516. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/schema_printer.py +168 -0
  517. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/suggestion_list.py +56 -0
  518. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_comparators.py +69 -0
  519. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_from_ast.py +21 -0
  520. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/type_info.py +149 -0
  521. wandb/vendor/graphql-core-1.1/wandb_graphql/utils/value_from_ast.py +69 -0
  522. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/__init__.py +4 -0
  523. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/__init__.py +79 -0
  524. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/arguments_of_correct_type.py +24 -0
  525. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/base.py +8 -0
  526. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/default_values_of_correct_type.py +44 -0
  527. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/fields_on_correct_type.py +113 -0
  528. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/fragments_on_composite_types.py +33 -0
  529. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_argument_names.py +70 -0
  530. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_directives.py +97 -0
  531. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_fragment_names.py +19 -0
  532. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/known_type_names.py +43 -0
  533. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/lone_anonymous_operation.py +23 -0
  534. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_fragment_cycles.py +59 -0
  535. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_undefined_variables.py +36 -0
  536. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_unused_fragments.py +38 -0
  537. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/no_unused_variables.py +37 -0
  538. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/overlapping_fields_can_be_merged.py +529 -0
  539. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/possible_fragment_spreads.py +44 -0
  540. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/provided_non_null_arguments.py +46 -0
  541. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/scalar_leafs.py +33 -0
  542. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_argument_names.py +32 -0
  543. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_fragment_names.py +28 -0
  544. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_input_field_names.py +33 -0
  545. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_operation_names.py +31 -0
  546. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/unique_variable_names.py +27 -0
  547. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/variables_are_input_types.py +21 -0
  548. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/rules/variables_in_allowed_position.py +53 -0
  549. wandb/vendor/graphql-core-1.1/wandb_graphql/validation/validation.py +158 -0
  550. wandb/vendor/promise-2.3.0/conftest.py +30 -0
  551. wandb/vendor/promise-2.3.0/setup.py +64 -0
  552. wandb/vendor/promise-2.3.0/tests/__init__.py +0 -0
  553. wandb/vendor/promise-2.3.0/tests/conftest.py +8 -0
  554. wandb/vendor/promise-2.3.0/tests/test_awaitable.py +32 -0
  555. wandb/vendor/promise-2.3.0/tests/test_awaitable_35.py +47 -0
  556. wandb/vendor/promise-2.3.0/tests/test_benchmark.py +116 -0
  557. wandb/vendor/promise-2.3.0/tests/test_complex_threads.py +23 -0
  558. wandb/vendor/promise-2.3.0/tests/test_dataloader.py +452 -0
  559. wandb/vendor/promise-2.3.0/tests/test_dataloader_awaitable_35.py +99 -0
  560. wandb/vendor/promise-2.3.0/tests/test_dataloader_extra.py +65 -0
  561. wandb/vendor/promise-2.3.0/tests/test_extra.py +670 -0
  562. wandb/vendor/promise-2.3.0/tests/test_issues.py +132 -0
  563. wandb/vendor/promise-2.3.0/tests/test_promise_list.py +70 -0
  564. wandb/vendor/promise-2.3.0/tests/test_spec.py +584 -0
  565. wandb/vendor/promise-2.3.0/tests/test_thread_safety.py +115 -0
  566. wandb/vendor/promise-2.3.0/tests/utils.py +3 -0
  567. wandb/vendor/promise-2.3.0/wandb_promise/__init__.py +38 -0
  568. wandb/vendor/promise-2.3.0/wandb_promise/async_.py +135 -0
  569. wandb/vendor/promise-2.3.0/wandb_promise/compat.py +32 -0
  570. wandb/vendor/promise-2.3.0/wandb_promise/dataloader.py +326 -0
  571. wandb/vendor/promise-2.3.0/wandb_promise/iterate_promise.py +12 -0
  572. wandb/vendor/promise-2.3.0/wandb_promise/promise.py +848 -0
  573. wandb/vendor/promise-2.3.0/wandb_promise/promise_list.py +151 -0
  574. wandb/vendor/promise-2.3.0/wandb_promise/pyutils/__init__.py +0 -0
  575. wandb/vendor/promise-2.3.0/wandb_promise/pyutils/version.py +83 -0
  576. wandb/vendor/promise-2.3.0/wandb_promise/schedulers/__init__.py +0 -0
  577. wandb/vendor/promise-2.3.0/wandb_promise/schedulers/asyncio.py +22 -0
  578. wandb/vendor/promise-2.3.0/wandb_promise/schedulers/gevent.py +21 -0
  579. wandb/vendor/promise-2.3.0/wandb_promise/schedulers/immediate.py +27 -0
  580. wandb/vendor/promise-2.3.0/wandb_promise/schedulers/thread.py +18 -0
  581. wandb/vendor/promise-2.3.0/wandb_promise/utils.py +56 -0
  582. wandb/vendor/pygments/__init__.py +90 -0
  583. wandb/vendor/pygments/cmdline.py +568 -0
  584. wandb/vendor/pygments/console.py +74 -0
  585. wandb/vendor/pygments/filter.py +74 -0
  586. wandb/vendor/pygments/filters/__init__.py +350 -0
  587. wandb/vendor/pygments/formatter.py +95 -0
  588. wandb/vendor/pygments/formatters/__init__.py +153 -0
  589. wandb/vendor/pygments/formatters/_mapping.py +85 -0
  590. wandb/vendor/pygments/formatters/bbcode.py +109 -0
  591. wandb/vendor/pygments/formatters/html.py +851 -0
  592. wandb/vendor/pygments/formatters/img.py +600 -0
  593. wandb/vendor/pygments/formatters/irc.py +182 -0
  594. wandb/vendor/pygments/formatters/latex.py +482 -0
  595. wandb/vendor/pygments/formatters/other.py +160 -0
  596. wandb/vendor/pygments/formatters/rtf.py +147 -0
  597. wandb/vendor/pygments/formatters/svg.py +153 -0
  598. wandb/vendor/pygments/formatters/terminal.py +136 -0
  599. wandb/vendor/pygments/formatters/terminal256.py +309 -0
  600. wandb/vendor/pygments/lexer.py +871 -0
  601. wandb/vendor/pygments/lexers/__init__.py +329 -0
  602. wandb/vendor/pygments/lexers/_asy_builtins.py +1645 -0
  603. wandb/vendor/pygments/lexers/_cl_builtins.py +232 -0
  604. wandb/vendor/pygments/lexers/_cocoa_builtins.py +72 -0
  605. wandb/vendor/pygments/lexers/_csound_builtins.py +1346 -0
  606. wandb/vendor/pygments/lexers/_lasso_builtins.py +5327 -0
  607. wandb/vendor/pygments/lexers/_lua_builtins.py +295 -0
  608. wandb/vendor/pygments/lexers/_mapping.py +500 -0
  609. wandb/vendor/pygments/lexers/_mql_builtins.py +1172 -0
  610. wandb/vendor/pygments/lexers/_openedge_builtins.py +2547 -0
  611. wandb/vendor/pygments/lexers/_php_builtins.py +4756 -0
  612. wandb/vendor/pygments/lexers/_postgres_builtins.py +621 -0
  613. wandb/vendor/pygments/lexers/_scilab_builtins.py +3094 -0
  614. wandb/vendor/pygments/lexers/_sourcemod_builtins.py +1163 -0
  615. wandb/vendor/pygments/lexers/_stan_builtins.py +532 -0
  616. wandb/vendor/pygments/lexers/_stata_builtins.py +419 -0
  617. wandb/vendor/pygments/lexers/_tsql_builtins.py +1004 -0
  618. wandb/vendor/pygments/lexers/_vim_builtins.py +1939 -0
  619. wandb/vendor/pygments/lexers/actionscript.py +240 -0
  620. wandb/vendor/pygments/lexers/agile.py +24 -0
  621. wandb/vendor/pygments/lexers/algebra.py +221 -0
  622. wandb/vendor/pygments/lexers/ambient.py +76 -0
  623. wandb/vendor/pygments/lexers/ampl.py +87 -0
  624. wandb/vendor/pygments/lexers/apl.py +101 -0
  625. wandb/vendor/pygments/lexers/archetype.py +318 -0
  626. wandb/vendor/pygments/lexers/asm.py +641 -0
  627. wandb/vendor/pygments/lexers/automation.py +374 -0
  628. wandb/vendor/pygments/lexers/basic.py +500 -0
  629. wandb/vendor/pygments/lexers/bibtex.py +160 -0
  630. wandb/vendor/pygments/lexers/business.py +612 -0
  631. wandb/vendor/pygments/lexers/c_cpp.py +252 -0
  632. wandb/vendor/pygments/lexers/c_like.py +541 -0
  633. wandb/vendor/pygments/lexers/capnproto.py +78 -0
  634. wandb/vendor/pygments/lexers/chapel.py +102 -0
  635. wandb/vendor/pygments/lexers/clean.py +288 -0
  636. wandb/vendor/pygments/lexers/compiled.py +34 -0
  637. wandb/vendor/pygments/lexers/configs.py +833 -0
  638. wandb/vendor/pygments/lexers/console.py +114 -0
  639. wandb/vendor/pygments/lexers/crystal.py +393 -0
  640. wandb/vendor/pygments/lexers/csound.py +366 -0
  641. wandb/vendor/pygments/lexers/css.py +689 -0
  642. wandb/vendor/pygments/lexers/d.py +251 -0
  643. wandb/vendor/pygments/lexers/dalvik.py +125 -0
  644. wandb/vendor/pygments/lexers/data.py +555 -0
  645. wandb/vendor/pygments/lexers/diff.py +165 -0
  646. wandb/vendor/pygments/lexers/dotnet.py +691 -0
  647. wandb/vendor/pygments/lexers/dsls.py +878 -0
  648. wandb/vendor/pygments/lexers/dylan.py +289 -0
  649. wandb/vendor/pygments/lexers/ecl.py +125 -0
  650. wandb/vendor/pygments/lexers/eiffel.py +65 -0
  651. wandb/vendor/pygments/lexers/elm.py +121 -0
  652. wandb/vendor/pygments/lexers/erlang.py +533 -0
  653. wandb/vendor/pygments/lexers/esoteric.py +277 -0
  654. wandb/vendor/pygments/lexers/ezhil.py +69 -0
  655. wandb/vendor/pygments/lexers/factor.py +344 -0
  656. wandb/vendor/pygments/lexers/fantom.py +250 -0
  657. wandb/vendor/pygments/lexers/felix.py +273 -0
  658. wandb/vendor/pygments/lexers/forth.py +177 -0
  659. wandb/vendor/pygments/lexers/fortran.py +205 -0
  660. wandb/vendor/pygments/lexers/foxpro.py +428 -0
  661. wandb/vendor/pygments/lexers/functional.py +21 -0
  662. wandb/vendor/pygments/lexers/go.py +101 -0
  663. wandb/vendor/pygments/lexers/grammar_notation.py +213 -0
  664. wandb/vendor/pygments/lexers/graph.py +80 -0
  665. wandb/vendor/pygments/lexers/graphics.py +553 -0
  666. wandb/vendor/pygments/lexers/haskell.py +843 -0
  667. wandb/vendor/pygments/lexers/haxe.py +936 -0
  668. wandb/vendor/pygments/lexers/hdl.py +382 -0
  669. wandb/vendor/pygments/lexers/hexdump.py +103 -0
  670. wandb/vendor/pygments/lexers/html.py +602 -0
  671. wandb/vendor/pygments/lexers/idl.py +270 -0
  672. wandb/vendor/pygments/lexers/igor.py +288 -0
  673. wandb/vendor/pygments/lexers/inferno.py +96 -0
  674. wandb/vendor/pygments/lexers/installers.py +322 -0
  675. wandb/vendor/pygments/lexers/int_fiction.py +1343 -0
  676. wandb/vendor/pygments/lexers/iolang.py +63 -0
  677. wandb/vendor/pygments/lexers/j.py +146 -0
  678. wandb/vendor/pygments/lexers/javascript.py +1525 -0
  679. wandb/vendor/pygments/lexers/julia.py +333 -0
  680. wandb/vendor/pygments/lexers/jvm.py +1573 -0
  681. wandb/vendor/pygments/lexers/lisp.py +2621 -0
  682. wandb/vendor/pygments/lexers/make.py +202 -0
  683. wandb/vendor/pygments/lexers/markup.py +595 -0
  684. wandb/vendor/pygments/lexers/math.py +21 -0
  685. wandb/vendor/pygments/lexers/matlab.py +663 -0
  686. wandb/vendor/pygments/lexers/ml.py +769 -0
  687. wandb/vendor/pygments/lexers/modeling.py +358 -0
  688. wandb/vendor/pygments/lexers/modula2.py +1561 -0
  689. wandb/vendor/pygments/lexers/monte.py +204 -0
  690. wandb/vendor/pygments/lexers/ncl.py +894 -0
  691. wandb/vendor/pygments/lexers/nimrod.py +159 -0
  692. wandb/vendor/pygments/lexers/nit.py +64 -0
  693. wandb/vendor/pygments/lexers/nix.py +136 -0
  694. wandb/vendor/pygments/lexers/oberon.py +105 -0
  695. wandb/vendor/pygments/lexers/objective.py +504 -0
  696. wandb/vendor/pygments/lexers/ooc.py +85 -0
  697. wandb/vendor/pygments/lexers/other.py +41 -0
  698. wandb/vendor/pygments/lexers/parasail.py +79 -0
  699. wandb/vendor/pygments/lexers/parsers.py +835 -0
  700. wandb/vendor/pygments/lexers/pascal.py +644 -0
  701. wandb/vendor/pygments/lexers/pawn.py +199 -0
  702. wandb/vendor/pygments/lexers/perl.py +620 -0
  703. wandb/vendor/pygments/lexers/php.py +267 -0
  704. wandb/vendor/pygments/lexers/praat.py +294 -0
  705. wandb/vendor/pygments/lexers/prolog.py +306 -0
  706. wandb/vendor/pygments/lexers/python.py +939 -0
  707. wandb/vendor/pygments/lexers/qvt.py +152 -0
  708. wandb/vendor/pygments/lexers/r.py +453 -0
  709. wandb/vendor/pygments/lexers/rdf.py +270 -0
  710. wandb/vendor/pygments/lexers/rebol.py +431 -0
  711. wandb/vendor/pygments/lexers/resource.py +85 -0
  712. wandb/vendor/pygments/lexers/rnc.py +67 -0
  713. wandb/vendor/pygments/lexers/roboconf.py +82 -0
  714. wandb/vendor/pygments/lexers/robotframework.py +560 -0
  715. wandb/vendor/pygments/lexers/ruby.py +519 -0
  716. wandb/vendor/pygments/lexers/rust.py +220 -0
  717. wandb/vendor/pygments/lexers/sas.py +228 -0
  718. wandb/vendor/pygments/lexers/scripting.py +1222 -0
  719. wandb/vendor/pygments/lexers/shell.py +794 -0
  720. wandb/vendor/pygments/lexers/smalltalk.py +195 -0
  721. wandb/vendor/pygments/lexers/smv.py +79 -0
  722. wandb/vendor/pygments/lexers/snobol.py +83 -0
  723. wandb/vendor/pygments/lexers/special.py +103 -0
  724. wandb/vendor/pygments/lexers/sql.py +681 -0
  725. wandb/vendor/pygments/lexers/stata.py +108 -0
  726. wandb/vendor/pygments/lexers/supercollider.py +90 -0
  727. wandb/vendor/pygments/lexers/tcl.py +145 -0
  728. wandb/vendor/pygments/lexers/templates.py +2283 -0
  729. wandb/vendor/pygments/lexers/testing.py +207 -0
  730. wandb/vendor/pygments/lexers/text.py +25 -0
  731. wandb/vendor/pygments/lexers/textedit.py +169 -0
  732. wandb/vendor/pygments/lexers/textfmts.py +297 -0
  733. wandb/vendor/pygments/lexers/theorem.py +458 -0
  734. wandb/vendor/pygments/lexers/trafficscript.py +54 -0
  735. wandb/vendor/pygments/lexers/typoscript.py +226 -0
  736. wandb/vendor/pygments/lexers/urbi.py +133 -0
  737. wandb/vendor/pygments/lexers/varnish.py +190 -0
  738. wandb/vendor/pygments/lexers/verification.py +111 -0
  739. wandb/vendor/pygments/lexers/web.py +24 -0
  740. wandb/vendor/pygments/lexers/webmisc.py +988 -0
  741. wandb/vendor/pygments/lexers/whiley.py +116 -0
  742. wandb/vendor/pygments/lexers/x10.py +69 -0
  743. wandb/vendor/pygments/modeline.py +44 -0
  744. wandb/vendor/pygments/plugin.py +68 -0
  745. wandb/vendor/pygments/regexopt.py +92 -0
  746. wandb/vendor/pygments/scanner.py +105 -0
  747. wandb/vendor/pygments/sphinxext.py +158 -0
  748. wandb/vendor/pygments/style.py +155 -0
  749. wandb/vendor/pygments/styles/__init__.py +80 -0
  750. wandb/vendor/pygments/styles/abap.py +29 -0
  751. wandb/vendor/pygments/styles/algol.py +63 -0
  752. wandb/vendor/pygments/styles/algol_nu.py +63 -0
  753. wandb/vendor/pygments/styles/arduino.py +98 -0
  754. wandb/vendor/pygments/styles/autumn.py +65 -0
  755. wandb/vendor/pygments/styles/borland.py +51 -0
  756. wandb/vendor/pygments/styles/bw.py +49 -0
  757. wandb/vendor/pygments/styles/colorful.py +81 -0
  758. wandb/vendor/pygments/styles/default.py +73 -0
  759. wandb/vendor/pygments/styles/emacs.py +72 -0
  760. wandb/vendor/pygments/styles/friendly.py +72 -0
  761. wandb/vendor/pygments/styles/fruity.py +42 -0
  762. wandb/vendor/pygments/styles/igor.py +29 -0
  763. wandb/vendor/pygments/styles/lovelace.py +97 -0
  764. wandb/vendor/pygments/styles/manni.py +75 -0
  765. wandb/vendor/pygments/styles/monokai.py +106 -0
  766. wandb/vendor/pygments/styles/murphy.py +80 -0
  767. wandb/vendor/pygments/styles/native.py +65 -0
  768. wandb/vendor/pygments/styles/paraiso_dark.py +125 -0
  769. wandb/vendor/pygments/styles/paraiso_light.py +125 -0
  770. wandb/vendor/pygments/styles/pastie.py +75 -0
  771. wandb/vendor/pygments/styles/perldoc.py +69 -0
  772. wandb/vendor/pygments/styles/rainbow_dash.py +89 -0
  773. wandb/vendor/pygments/styles/rrt.py +33 -0
  774. wandb/vendor/pygments/styles/sas.py +44 -0
  775. wandb/vendor/pygments/styles/stata.py +40 -0
  776. wandb/vendor/pygments/styles/tango.py +141 -0
  777. wandb/vendor/pygments/styles/trac.py +63 -0
  778. wandb/vendor/pygments/styles/vim.py +63 -0
  779. wandb/vendor/pygments/styles/vs.py +38 -0
  780. wandb/vendor/pygments/styles/xcode.py +51 -0
  781. wandb/vendor/pygments/token.py +213 -0
  782. wandb/vendor/pygments/unistring.py +217 -0
  783. wandb/vendor/pygments/util.py +388 -0
  784. wandb/vendor/pynvml/__init__.py +0 -0
  785. wandb/vendor/pynvml/pynvml.py +4779 -0
  786. wandb/vendor/watchdog_0_9_0/wandb_watchdog/__init__.py +17 -0
  787. wandb/vendor/watchdog_0_9_0/wandb_watchdog/events.py +615 -0
  788. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/__init__.py +98 -0
  789. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/api.py +369 -0
  790. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/fsevents.py +172 -0
  791. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/fsevents2.py +239 -0
  792. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify.py +218 -0
  793. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify_buffer.py +81 -0
  794. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/inotify_c.py +575 -0
  795. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/kqueue.py +730 -0
  796. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/polling.py +145 -0
  797. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/read_directory_changes.py +133 -0
  798. wandb/vendor/watchdog_0_9_0/wandb_watchdog/observers/winapi.py +348 -0
  799. wandb/vendor/watchdog_0_9_0/wandb_watchdog/patterns.py +265 -0
  800. wandb/vendor/watchdog_0_9_0/wandb_watchdog/tricks/__init__.py +174 -0
  801. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/__init__.py +151 -0
  802. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/bricks.py +249 -0
  803. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/compat.py +29 -0
  804. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/decorators.py +198 -0
  805. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/delayed_queue.py +88 -0
  806. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/dirsnapshot.py +293 -0
  807. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/echo.py +157 -0
  808. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/event_backport.py +41 -0
  809. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/importlib2.py +40 -0
  810. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/platform.py +57 -0
  811. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/unicode_paths.py +64 -0
  812. wandb/vendor/watchdog_0_9_0/wandb_watchdog/utils/win32stat.py +123 -0
  813. wandb/vendor/watchdog_0_9_0/wandb_watchdog/version.py +28 -0
  814. wandb/vendor/watchdog_0_9_0/wandb_watchdog/watchmedo.py +577 -0
  815. wandb/wandb_agent.py +588 -0
  816. wandb/wandb_controller.py +719 -0
  817. wandb/wandb_run.py +9 -0
  818. wandb-0.19.1.dist-info/METADATA +223 -0
  819. wandb-0.19.1.dist-info/RECORD +822 -0
  820. wandb-0.19.1.dist-info/WHEEL +5 -0
  821. wandb-0.19.1.dist-info/entry_points.txt +3 -0
  822. wandb-0.19.1.dist-info/licenses/LICENSE +21 -0
@@ -0,0 +1,740 @@
1
+ """Abstract Scheduler class."""
2
+
3
+ import asyncio
4
+ import base64
5
+ import copy
6
+ import logging
7
+ import os
8
+ import socket
9
+ import threading
10
+ import time
11
+ import traceback
12
+ from abc import ABC, abstractmethod
13
+ from dataclasses import dataclass
14
+ from enum import Enum
15
+ from typing import TYPE_CHECKING, Any, Dict, Iterator, List, Optional, Tuple, Union
16
+
17
+ import click
18
+ import yaml
19
+
20
+ import wandb
21
+ from wandb.errors import CommError
22
+ from wandb.sdk.launch._launch_add import launch_add
23
+ from wandb.sdk.launch.errors import LaunchError
24
+ from wandb.sdk.launch.sweeps import SchedulerError
25
+ from wandb.sdk.launch.sweeps.utils import (
26
+ create_sweep_command_args,
27
+ make_launch_sweep_entrypoint,
28
+ )
29
+ from wandb.sdk.launch.utils import (
30
+ event_loop_thread_exec,
31
+ strip_resource_args_and_template_vars,
32
+ )
33
+ from wandb.sdk.lib.runid import generate_id
34
+
35
+ if TYPE_CHECKING:
36
+ import wandb.apis.public as public
37
+ from wandb.apis.internal import Api
38
+ from wandb.apis.public import QueuedRun, Run
39
+ from wandb.sdk.wandb_run import Run as SdkRun
40
+
41
+
42
+ _logger = logging.getLogger(__name__)
43
+ LOG_PREFIX = f"{click.style('sched:', fg='cyan')} "
44
+
45
+ DEFAULT_POLLING_SLEEP = 5.0
46
+
47
+
48
+ class SchedulerState(Enum):
49
+ PENDING = 0
50
+ STARTING = 1
51
+ RUNNING = 2
52
+ FLUSH_RUNS = 3
53
+ COMPLETED = 4
54
+ FAILED = 5
55
+ STOPPED = 6
56
+ CANCELLED = 7
57
+
58
+
59
+ class RunState(Enum):
60
+ RUNNING = "running", "alive"
61
+ PENDING = "pending", "alive"
62
+ PREEMPTING = "preempting", "alive"
63
+ CRASHED = "crashed", "dead"
64
+ FAILED = "failed", "dead"
65
+ KILLED = "killed", "dead"
66
+ FINISHED = "finished", "dead"
67
+ PREEMPTED = "preempted", "dead"
68
+ # unknown when api.get_run_state fails or returns unexpected state
69
+ # assumed alive, unless we get unknown 2x then move to failed (dead)
70
+ UNKNOWN = "unknown", "alive"
71
+
72
+ def __new__(cls: Any, *args: List, **kwds: Any) -> "RunState":
73
+ obj: RunState = object.__new__(cls)
74
+ obj._value_ = args[0]
75
+ return obj
76
+
77
+ def __init__(self, _: str, life: str = "unknown") -> None:
78
+ self._life = life
79
+
80
+ @property
81
+ def is_alive(self) -> bool:
82
+ return self._life == "alive"
83
+
84
+
85
+ @dataclass
86
+ class _Worker:
87
+ agent_config: Dict[str, Any]
88
+ agent_id: str
89
+
90
+
91
+ @dataclass
92
+ class SweepRun:
93
+ id: str
94
+ worker_id: int
95
+ state: RunState = RunState.RUNNING
96
+ queued_run: Optional["public.QueuedRun"] = None
97
+ args: Optional[Dict[str, Any]] = None
98
+ logs: Optional[List[str]] = None
99
+
100
+
101
+ class Scheduler(ABC):
102
+ """A controller/agent that populates a Launch RunQueue from a hyperparameter sweep."""
103
+
104
+ PLACEHOLDER_URI = "placeholder-uri-scheduler"
105
+ SWEEP_JOB_TYPE = "sweep-controller"
106
+ ENTRYPOINT = ["wandb", "scheduler", "WANDB_SWEEP_ID"]
107
+
108
+ def __init__(
109
+ self,
110
+ api: "Api",
111
+ *args: Optional[Any],
112
+ polling_sleep: Optional[float] = None,
113
+ sweep_id: Optional[str] = None,
114
+ entity: Optional[str] = None,
115
+ project: Optional[str] = None,
116
+ project_queue: Optional[str] = None,
117
+ num_workers: Optional[Union[int, str]] = None,
118
+ **kwargs: Optional[Any],
119
+ ):
120
+ from wandb.apis.public import Api as PublicApi
121
+
122
+ self._api = api
123
+ self._public_api = PublicApi()
124
+ self._entity = (
125
+ entity
126
+ or os.environ.get("WANDB_ENTITY")
127
+ or api.settings("entity")
128
+ or api.default_entity
129
+ )
130
+ self._project = (
131
+ project or os.environ.get("WANDB_PROJECT") or api.settings("project")
132
+ )
133
+ self._sweep_id: str = sweep_id or "empty-sweep-id"
134
+ self._state: SchedulerState = SchedulerState.PENDING
135
+
136
+ # Make sure the provided sweep_id corresponds to a valid sweep
137
+ try:
138
+ resp = self._api.sweep(
139
+ sweep_id, "{}", entity=self._entity, project=self._project
140
+ )
141
+ if resp.get("state") == SchedulerState.CANCELLED.name:
142
+ self._state = SchedulerState.CANCELLED
143
+ self._sweep_config = yaml.safe_load(resp["config"])
144
+ self._num_runs_launched: int = self._get_num_runs_launched(resp["runs"])
145
+ if self._num_runs_launched > 0:
146
+ wandb.termlog(
147
+ f"{LOG_PREFIX}Found {self._num_runs_launched} previous valid runs for sweep {self._sweep_id}"
148
+ )
149
+ except Exception as e:
150
+ raise SchedulerError(
151
+ f"{LOG_PREFIX}Exception when finding sweep ({sweep_id}) {e}"
152
+ )
153
+
154
+ # Scheduler may receive additional kwargs which will be piped into the launch command
155
+ self._kwargs: Dict[str, Any] = kwargs
156
+
157
+ # Dictionary of the runs being managed by the scheduler
158
+ self._runs: Dict[str, SweepRun] = {}
159
+ # Threading lock to ensure thread-safe access to the runs dictionary
160
+ self._threading_lock: threading.Lock = threading.Lock()
161
+ self._polling_sleep = (
162
+ polling_sleep if polling_sleep is not None else DEFAULT_POLLING_SLEEP
163
+ )
164
+ self._project_queue = project_queue
165
+ # Optionally run multiple workers in (pseudo-)parallel. Workers do not
166
+ # actually run training workloads, they simply send heartbeat messages
167
+ # (emulating a real agent) and add new runs to the launch queue. The
168
+ # launch agent is the one that actually runs the training workloads.
169
+ self._workers: Dict[int, _Worker] = {}
170
+
171
+ # Init wandb scheduler run
172
+ self._wandb_run = self._init_wandb_run()
173
+
174
+ # Grab params from scheduler wandb run config
175
+ num_workers = num_workers or self._wandb_run.config.get("scheduler", {}).get(
176
+ "num_workers"
177
+ )
178
+ self._num_workers = int(num_workers) if str(num_workers).isdigit() else 8
179
+ self._settings_config: Dict[str, Any] = self._wandb_run.config.get(
180
+ "settings", {}
181
+ )
182
+
183
+ @abstractmethod
184
+ def _get_next_sweep_run(self, worker_id: int) -> Optional[SweepRun]:
185
+ """Called when worker available."""
186
+
187
+ @abstractmethod
188
+ def _poll(self) -> None:
189
+ """Called every polling loop."""
190
+
191
+ @abstractmethod
192
+ def _exit(self) -> None:
193
+ pass
194
+
195
+ @abstractmethod
196
+ def _load_state(self) -> None:
197
+ pass
198
+
199
+ @abstractmethod
200
+ def _save_state(self) -> None:
201
+ pass
202
+
203
+ @property
204
+ def state(self) -> SchedulerState:
205
+ _logger.debug(f"{LOG_PREFIX}Scheduler state is {self._state.name}")
206
+ return self._state
207
+
208
+ @state.setter
209
+ def state(self, value: SchedulerState) -> None:
210
+ _logger.debug(f"{LOG_PREFIX}Scheduler was {self.state.name} is {value.name}")
211
+ self._state = value
212
+
213
+ @property
214
+ def is_alive(self) -> bool:
215
+ if self.state in [
216
+ SchedulerState.COMPLETED,
217
+ SchedulerState.FAILED,
218
+ SchedulerState.STOPPED,
219
+ SchedulerState.CANCELLED,
220
+ ]:
221
+ return False
222
+ return True
223
+
224
+ @property
225
+ def at_runcap(self) -> bool:
226
+ """False if under user-specified cap on # of runs."""
227
+ run_cap = self._sweep_config.get("run_cap")
228
+ if not run_cap:
229
+ return False
230
+ at_runcap: bool = self._num_runs_launched >= run_cap
231
+ return at_runcap
232
+
233
+ @property
234
+ def num_active_runs(self) -> int:
235
+ return len(self._runs)
236
+
237
+ @property
238
+ def busy_workers(self) -> Dict[int, _Worker]:
239
+ """Returns dict of id:worker already assigned to a launch run.
240
+
241
+ runs should always have a worker_id, but are created before
242
+ workers are assigned to the run
243
+ """
244
+ busy_workers = {}
245
+ for _, r in self._yield_runs():
246
+ busy_workers[r.worker_id] = self._workers[r.worker_id]
247
+ return busy_workers
248
+
249
+ @property
250
+ def available_workers(self) -> Dict[int, _Worker]:
251
+ """Returns dict of id:worker ready to launch another run."""
252
+ if len(self._workers) == 0:
253
+ return {}
254
+ return {
255
+ _id: w for _id, w in self._workers.items() if _id not in self.busy_workers
256
+ }
257
+
258
+ def _init_wandb_run(self) -> "SdkRun":
259
+ """Controls resume or init logic for a scheduler wandb run."""
260
+ settings = wandb.Settings(disable_job_creation=True)
261
+ run: SdkRun = wandb.init( # type: ignore
262
+ name=f"Scheduler.{self._sweep_id}",
263
+ resume="allow",
264
+ config=self._kwargs, # when run as a job, this sets config
265
+ settings=settings,
266
+ )
267
+ return run
268
+
269
+ def stop_sweep(self) -> None:
270
+ """Stop the sweep."""
271
+ self._state = SchedulerState.STOPPED
272
+
273
+ def fail_sweep(self, err: Optional[str]) -> None:
274
+ """Fail the sweep w/ optional exception."""
275
+ self._state = SchedulerState.FAILED
276
+ if err:
277
+ raise SchedulerError(err)
278
+
279
+ def start(self) -> None:
280
+ """Start a scheduler, confirms prerequisites, begins execution loop."""
281
+ wandb.termlog(f"{LOG_PREFIX}Scheduler starting.")
282
+ if not self.is_alive:
283
+ wandb.termerror(
284
+ f"{LOG_PREFIX}Sweep already in end state ({self.state.name.lower()}). Exiting..."
285
+ )
286
+ self.exit()
287
+ return
288
+
289
+ self._state = SchedulerState.STARTING
290
+ if not self._try_load_executable():
291
+ wandb.termerror(
292
+ f"{LOG_PREFIX}No 'job' or 'image_uri' loaded from sweep config."
293
+ )
294
+ self.exit()
295
+ return
296
+
297
+ # For resuming sweeps
298
+ self._load_state()
299
+ asyncio.run(self._register_agents())
300
+ self.run()
301
+
302
+ def run(self) -> None:
303
+ """Main run function."""
304
+ wandb.termlog(f"{LOG_PREFIX}Scheduler running")
305
+ self.state = SchedulerState.RUNNING
306
+ try:
307
+ while True:
308
+ self._update_scheduler_run_state()
309
+ if not self.is_alive:
310
+ break
311
+
312
+ wandb.termlog(f"{LOG_PREFIX}Polling for new runs to launch")
313
+
314
+ self._update_run_states()
315
+ self._poll()
316
+ if self.state == SchedulerState.FLUSH_RUNS:
317
+ if self.num_active_runs == 0:
318
+ wandb.termlog(f"{LOG_PREFIX}Done polling on runs, exiting")
319
+ break
320
+ time.sleep(self._polling_sleep)
321
+ continue
322
+
323
+ for worker_id in self.available_workers:
324
+ if self.at_runcap:
325
+ wandb.termlog(
326
+ f"{LOG_PREFIX}Sweep at run_cap ({self._num_runs_launched})"
327
+ )
328
+ self.state = SchedulerState.FLUSH_RUNS
329
+ break
330
+
331
+ try:
332
+ run: Optional[SweepRun] = self._get_next_sweep_run(worker_id)
333
+ if not run:
334
+ break
335
+ except SchedulerError as e:
336
+ raise SchedulerError(e)
337
+ except Exception as e:
338
+ wandb.termerror(
339
+ f"{LOG_PREFIX}Failed to get next sweep run: {e}"
340
+ )
341
+ self.state = SchedulerState.FAILED
342
+ break
343
+
344
+ if self._add_to_launch_queue(run):
345
+ self._num_runs_launched += 1
346
+
347
+ time.sleep(self._polling_sleep)
348
+ except KeyboardInterrupt:
349
+ wandb.termwarn(f"{LOG_PREFIX}Scheduler received KeyboardInterrupt. Exiting")
350
+ self.state = SchedulerState.STOPPED
351
+ self.exit()
352
+ return
353
+ except Exception as e:
354
+ wandb.termlog(f"{LOG_PREFIX}Scheduler failed with exception {e}")
355
+ self.state = SchedulerState.FAILED
356
+ self.exit()
357
+ raise e
358
+ else:
359
+ # scheduler succeeds if at runcap
360
+ if self.state == SchedulerState.FLUSH_RUNS and self.at_runcap:
361
+ self.state = SchedulerState.COMPLETED
362
+ self.exit()
363
+
364
+ def exit(self) -> None:
365
+ self._exit()
366
+ # _save_state isn't controlled, possibly fails
367
+ try:
368
+ self._save_state()
369
+ except Exception:
370
+ wandb.termerror(
371
+ f"{LOG_PREFIX}Failed to save state: {traceback.format_exc()}"
372
+ )
373
+
374
+ status = ""
375
+ if self.state == SchedulerState.FLUSH_RUNS:
376
+ self._set_sweep_state("PAUSED")
377
+ status = "paused"
378
+ elif self.state == SchedulerState.COMPLETED:
379
+ self._set_sweep_state("FINISHED")
380
+ status = "completed"
381
+ elif self.state in [SchedulerState.CANCELLED, SchedulerState.STOPPED]:
382
+ self._set_sweep_state("CANCELED") # one L
383
+ status = "cancelled"
384
+ self._stop_runs()
385
+ else:
386
+ self.state = SchedulerState.FAILED
387
+ self._set_sweep_state("CRASHED")
388
+ status = "crashed"
389
+ self._stop_runs()
390
+
391
+ wandb.termlog(f"{LOG_PREFIX}Scheduler {status}")
392
+ self._wandb_run.finish()
393
+
394
+ def _get_num_runs_launched(self, runs: List[Dict[str, Any]]) -> int:
395
+ """Returns the number of valid runs in the sweep."""
396
+ count = 0
397
+ for run in runs:
398
+ # if bad run, shouldn't be counted against run cap
399
+ if run.get("state", "") in ["killed", "crashed"] and not run.get(
400
+ "summaryMetrics"
401
+ ):
402
+ _logger.debug(
403
+ f"excluding run: {run['name']} with state: {run['state']} from run cap \n{run}"
404
+ )
405
+ continue
406
+ count += 1
407
+
408
+ return count
409
+
410
+ def _try_load_executable(self) -> bool:
411
+ """Check existence of valid executable for a run.
412
+
413
+ logs and returns False when job is unreachable
414
+ """
415
+ if self._kwargs.get("job"):
416
+ try:
417
+ _job_artifact = self._public_api.job(self._kwargs["job"])
418
+ wandb.termlog(
419
+ f"{LOG_PREFIX}Successfully loaded job ({_job_artifact.name}) in scheduler"
420
+ )
421
+ except Exception:
422
+ wandb.termerror(f"{LOG_PREFIX}{traceback.format_exc()}")
423
+ return False
424
+ return True
425
+ elif self._kwargs.get("image_uri"):
426
+ # TODO(gst): check docker existence? Use registry in launch config?
427
+ return True
428
+ else:
429
+ return False
430
+
431
+ async def _register_agents(self) -> None:
432
+ tasks = []
433
+ register_agent = event_loop_thread_exec(self._api.register_agent)
434
+ for worker_id in range(self._num_workers):
435
+ _logger.debug(f"{LOG_PREFIX}Starting AgentHeartbeat worker ({worker_id})")
436
+ try:
437
+ worker = register_agent(
438
+ f"{socket.gethostname()}-{worker_id}", # host
439
+ sweep_id=self._sweep_id,
440
+ project_name=self._project,
441
+ entity=self._entity,
442
+ )
443
+ tasks.append(worker)
444
+ except Exception as e:
445
+ _logger.debug(f"failed to register agent: {e}")
446
+ self.fail_sweep(f"failed to register agent: {e}")
447
+
448
+ finished_tasks = await asyncio.gather(*tasks)
449
+ for idx, agent_config in enumerate(finished_tasks):
450
+ self._workers[idx] = _Worker(
451
+ agent_config=agent_config,
452
+ agent_id=agent_config["id"],
453
+ )
454
+
455
+ def _yield_runs(self) -> Iterator[Tuple[str, SweepRun]]:
456
+ """Thread-safe way to iterate over the runs."""
457
+ with self._threading_lock:
458
+ yield from self._runs.items()
459
+
460
+ def _cleanup_runs(self, runs_to_remove: List[str]) -> None:
461
+ """Helper for removing runs from memory.
462
+
463
+ Can be overloaded to prevent deletion of runs, which is useful
464
+ for debugging or when polling on completed runs.
465
+ """
466
+ with self._threading_lock:
467
+ for run_id in runs_to_remove:
468
+ wandb.termlog(f"{LOG_PREFIX}Cleaning up finished run ({run_id})")
469
+ del self._runs[run_id]
470
+
471
+ def _stop_runs(self) -> None:
472
+ to_delete = []
473
+ for run_id, _ in self._yield_runs():
474
+ to_delete += [run_id]
475
+
476
+ for run_id in to_delete:
477
+ wandb.termlog(f"{LOG_PREFIX}Stopping run ({run_id})")
478
+ if not self._stop_run(run_id):
479
+ wandb.termwarn(f"{LOG_PREFIX}Failed to stop run ({run_id})")
480
+
481
+ def _stop_run(self, run_id: str) -> bool:
482
+ """Stops a run and removes it from the scheduler."""
483
+ if run_id not in self._runs:
484
+ _logger.debug(f"run: {run_id} not in _runs: {self._runs}")
485
+ return False
486
+
487
+ run = self._runs[run_id]
488
+ del self._runs[run_id]
489
+
490
+ if not run.queued_run:
491
+ _logger.debug(
492
+ f"tried to _stop_run but run not queued yet (run_id:{run.id})"
493
+ )
494
+ return False
495
+
496
+ if not run.state.is_alive:
497
+ # run already dead, just delete reference
498
+ return True
499
+
500
+ # run still alive, send stop signal
501
+ encoded_run_id = base64.standard_b64encode(
502
+ f"Run:v1:{run_id}:{self._project}:{self._entity}".encode()
503
+ ).decode("utf-8")
504
+
505
+ try:
506
+ success: bool = self._api.stop_run(run_id=encoded_run_id)
507
+ if success:
508
+ wandb.termlog(f"{LOG_PREFIX}Stopped run {run_id}.")
509
+ return True
510
+ except Exception as e:
511
+ _logger.debug(f"error stopping run ({run_id}): {e}")
512
+
513
+ return False
514
+
515
+ def _update_scheduler_run_state(self) -> None:
516
+ """Update the scheduler state from state of scheduler run and sweep state."""
517
+ state: RunState = self._get_run_state(self._wandb_run.id)
518
+
519
+ # map scheduler run-state to scheduler-state
520
+ if state == RunState.KILLED:
521
+ self.state = SchedulerState.STOPPED
522
+ elif state in [RunState.FAILED, RunState.CRASHED]:
523
+ self.state = SchedulerState.FAILED
524
+ elif state == RunState.FINISHED:
525
+ self.state = SchedulerState.COMPLETED
526
+
527
+ # check sweep state for completed states, overwrite scheduler state
528
+ try:
529
+ sweep_state = self._api.get_sweep_state(
530
+ self._sweep_id, self._entity, self._project
531
+ )
532
+ except Exception as e:
533
+ _logger.debug(f"sweep state error: {e}")
534
+ return
535
+
536
+ if sweep_state == "FINISHED":
537
+ self.state = SchedulerState.COMPLETED
538
+ elif sweep_state in ["CANCELLED", "STOPPED"]:
539
+ self.state = SchedulerState.CANCELLED
540
+ elif sweep_state == "PAUSED":
541
+ self.state = SchedulerState.FLUSH_RUNS
542
+
543
+ def _update_run_states(self) -> None:
544
+ """Iterate through runs.
545
+
546
+ Get state from backend and deletes runs if not in running state. Threadsafe.
547
+ """
548
+ runs_to_remove: List[str] = []
549
+ for run_id, run in self._yield_runs():
550
+ run.state = self._get_run_state(run_id, run.state)
551
+
552
+ try:
553
+ rqi_state = run.queued_run.state if run.queued_run else None
554
+ except (CommError, LaunchError) as e:
555
+ _logger.debug(f"Failed to get queued_run.state: {e}")
556
+ rqi_state = None
557
+
558
+ if not run.state.is_alive or rqi_state == "failed":
559
+ _logger.debug(f"({run_id}) states: ({run.state}, {rqi_state})")
560
+ runs_to_remove.append(run_id)
561
+ self._cleanup_runs(runs_to_remove)
562
+
563
+ def _get_metrics_from_run(self, run_id: str) -> List[Any]:
564
+ """Use the public api to get metrics from a run.
565
+
566
+ Uses the metric name found in the sweep config, any
567
+ misspellings will result in an empty list.
568
+ """
569
+ try:
570
+ queued_run: Optional[QueuedRun] = self._runs[run_id].queued_run
571
+ if not queued_run:
572
+ return []
573
+
574
+ api_run: Run = self._public_api.run(
575
+ f"{queued_run.entity}/{queued_run.project}/{run_id}"
576
+ )
577
+ metric_name = self._sweep_config["metric"]["name"]
578
+ history = api_run.scan_history(keys=["_step", metric_name])
579
+ metrics = [x[metric_name] for x in history]
580
+
581
+ return metrics
582
+ except Exception as e:
583
+ _logger.debug(f"[_get_metrics_from_run] {e}")
584
+ return []
585
+
586
+ def _get_run_info(self, run_id: str) -> Dict[str, Any]:
587
+ """Use the public api to get info about a run."""
588
+ try:
589
+ info: Dict[str, Any] = self._api.get_run_info(
590
+ self._entity, self._project, run_id
591
+ )
592
+ if info:
593
+ return info
594
+ except Exception as e:
595
+ _logger.debug(f"[_get_run_info] {e}")
596
+ return {}
597
+
598
+ def _get_run_state(
599
+ self, run_id: str, prev_run_state: RunState = RunState.UNKNOWN
600
+ ) -> RunState:
601
+ """Use the public api to get state of a run."""
602
+ run_state = None
603
+ try:
604
+ state = self._api.get_run_state(self._entity, self._project, run_id)
605
+ run_state = RunState(state)
606
+ except CommError as e:
607
+ _logger.debug(f"error getting state for run ({run_id}): {e}")
608
+ if prev_run_state == RunState.UNKNOWN:
609
+ # triggers when we get an unknown state for the second time
610
+ wandb.termwarn(
611
+ f"Failed to get runstate for run ({run_id}). Error: {traceback.format_exc()}"
612
+ )
613
+ run_state = RunState.FAILED
614
+ else: # first time we get unknown state
615
+ run_state = RunState.UNKNOWN
616
+ except (AttributeError, ValueError):
617
+ wandb.termwarn(
618
+ f"Bad state ({run_state}) for run ({run_id}). Error: {traceback.format_exc()}"
619
+ )
620
+ run_state = RunState.UNKNOWN
621
+ return run_state
622
+
623
+ def _create_run(self) -> Dict[str, Any]:
624
+ """Use the public api to create a blank run."""
625
+ try:
626
+ run: List[Dict[str, Any]] = self._api.upsert_run(
627
+ project=self._project,
628
+ entity=self._entity,
629
+ sweep_name=self._sweep_id,
630
+ )
631
+ if run:
632
+ return run[0]
633
+ except Exception as e:
634
+ _logger.debug(f"[_create_run] {e}")
635
+ raise SchedulerError(
636
+ "Error creating run from scheduler, check API connection and CLI version."
637
+ )
638
+ return {}
639
+
640
+ def _set_sweep_state(self, state: str) -> None:
641
+ wandb.termlog(f"{LOG_PREFIX}Updating sweep state to: {state.lower()}")
642
+ try:
643
+ self._api.set_sweep_state(sweep=self._sweep_id, state=state)
644
+ except Exception as e:
645
+ _logger.debug(f"[set_sweep_state] {e}")
646
+
647
+ def _encode(self, _id: str) -> str:
648
+ return (
649
+ base64.b64decode(bytes(_id.encode("utf-8"))).decode("utf-8").split(":")[2]
650
+ )
651
+
652
+ def _make_entry_and_launch_config(
653
+ self, run: SweepRun
654
+ ) -> Tuple[Optional[List[str]], Dict[str, Dict[str, Any]]]:
655
+ args = create_sweep_command_args({"args": run.args})
656
+ entry_point, macro_args = make_launch_sweep_entrypoint(
657
+ args, self._sweep_config.get("command")
658
+ )
659
+ # handle program macro
660
+ if entry_point and "${program}" in entry_point:
661
+ if not self._sweep_config.get("program"):
662
+ raise SchedulerError(
663
+ f"{LOG_PREFIX}Program macro in command has no corresponding 'program' in sweep config."
664
+ )
665
+ pidx = entry_point.index("${program}")
666
+ entry_point[pidx] = self._sweep_config["program"]
667
+
668
+ launch_config = copy.deepcopy(self._wandb_run.config.get("launch", {}))
669
+ if "overrides" not in launch_config:
670
+ launch_config["overrides"] = {"run_config": {}}
671
+ if "run_config" not in launch_config["overrides"]:
672
+ launch_config["overrides"]["run_config"] = {}
673
+ launch_config["overrides"]["run_config"].update(args["args_dict"])
674
+
675
+ if macro_args: # pipe in hyperparam args as params to launch
676
+ launch_config["overrides"]["args"] = macro_args
677
+
678
+ if entry_point:
679
+ unresolved = [x for x in entry_point if str(x).startswith("${")]
680
+ if unresolved:
681
+ wandb.termwarn(
682
+ f"{LOG_PREFIX}Sweep command contains unresolved macros: "
683
+ f"{unresolved}, see launch docs for supported macros."
684
+ )
685
+ return entry_point, launch_config
686
+
687
+ def _add_to_launch_queue(self, run: SweepRun) -> bool:
688
+ """Convert a sweeprun into a launch job then push to runqueue."""
689
+ # job and image first from CLI args, then from sweep config
690
+ _job = self._kwargs.get("job") or self._sweep_config.get("job")
691
+ _sweep_config_uri = self._sweep_config.get("image_uri")
692
+ _image_uri = self._kwargs.get("image_uri") or _sweep_config_uri
693
+ if _job is None and _image_uri is None:
694
+ raise SchedulerError(f"{LOG_PREFIX}No 'job' nor 'image_uri' ({run.id})")
695
+ elif _job is not None and _image_uri is not None:
696
+ raise SchedulerError(f"{LOG_PREFIX}Sweep has both 'job' and 'image_uri'")
697
+
698
+ entry_point, launch_config = self._make_entry_and_launch_config(run)
699
+ if entry_point:
700
+ wandb.termwarn(
701
+ f"{LOG_PREFIX}Sweep command {entry_point} will override"
702
+ f' {"job" if _job else "image_uri"} entrypoint'
703
+ )
704
+
705
+ # override resource and args of job
706
+ _job_launch_config = copy.deepcopy(self._wandb_run.config.get("launch")) or {}
707
+
708
+ # default priority is "medium"
709
+ _priority = int(launch_config.get("priority", 2)) # type: ignore
710
+
711
+ # strip resource_args and template_variables from launch_config
712
+ strip_resource_args_and_template_vars(_job_launch_config)
713
+
714
+ run_id = run.id or generate_id()
715
+ queued_run = launch_add(
716
+ run_id=run_id,
717
+ entry_point=entry_point,
718
+ config=launch_config,
719
+ docker_image=_image_uri, # TODO(gst): make agnostic (github? run uri?)
720
+ job=_job,
721
+ project=self._project,
722
+ entity=self._entity,
723
+ queue_name=self._kwargs.get("queue"),
724
+ project_queue=self._project_queue,
725
+ resource=_job_launch_config.get("resource"),
726
+ resource_args=_job_launch_config.get("resource_args"),
727
+ template_variables=_job_launch_config.get("template_variables"),
728
+ author=self._kwargs.get("author"),
729
+ sweep_id=self._sweep_id,
730
+ priority=_priority,
731
+ )
732
+ run.queued_run = queued_run
733
+ # TODO(gst): unify run and queued_run state
734
+ run.state = RunState.RUNNING # assume it will get picked up
735
+ self._runs[run_id] = run
736
+
737
+ wandb.termlog(
738
+ f"{LOG_PREFIX}Added run ({run_id}) to queue ({self._kwargs.get('queue')})"
739
+ )
740
+ return True