synth-ai 0.2.9.dev0__py3-none-any.whl → 0.2.23.dev3__py3-none-any.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.
- examples/README.md +1 -0
- examples/__init__.py +16 -0
- examples/analyze_semantic_words.sh +17 -0
- examples/baseline/banking77_baseline.py +243 -0
- examples/baseline/banking77_pipeline_baseline.py +294 -0
- examples/baseline/crafter_baseline.py +407 -0
- examples/baseline/pokemon_red_baseline.py +326 -0
- examples/baseline/simple_baseline.py +56 -0
- examples/baseline/warming_up_to_rl_baseline.py +239 -0
- examples/blog_posts/gepa/README.md +355 -0
- examples/blog_posts/gepa/configs/banking77_gepa_local.toml +95 -0
- examples/blog_posts/gepa/configs/banking77_gepa_test.toml +80 -0
- examples/blog_posts/gepa/configs/banking77_mipro_local.toml +50 -0
- examples/blog_posts/gepa/configs/banking77_pipeline_gepa_local.toml +101 -0
- examples/blog_posts/gepa/configs/banking77_pipeline_gepa_test.toml +96 -0
- examples/blog_posts/gepa/configs/hotpotqa_gepa_local.toml +57 -0
- examples/blog_posts/gepa/configs/hotpotqa_gepa_qwen.toml +35 -0
- examples/blog_posts/gepa/configs/hotpotqa_mipro_local.toml +51 -0
- examples/blog_posts/gepa/configs/hover_gepa_local.toml +57 -0
- examples/blog_posts/gepa/configs/hover_gepa_qwen.toml +35 -0
- examples/blog_posts/gepa/configs/hover_mipro_local.toml +51 -0
- examples/blog_posts/gepa/configs/ifbench_gepa_local.toml +57 -0
- examples/blog_posts/gepa/configs/ifbench_gepa_qwen.toml +35 -0
- examples/blog_posts/gepa/configs/ifbench_mipro_local.toml +51 -0
- examples/blog_posts/gepa/configs/pupa_gepa_local.toml +58 -0
- examples/blog_posts/gepa/configs/pupa_mipro_local.toml +52 -0
- examples/blog_posts/gepa/deploy_banking77_task_app.sh +54 -0
- examples/blog_posts/gepa/gepa_baseline.py +204 -0
- examples/blog_posts/gepa/query_prompts_example.py +97 -0
- examples/blog_posts/gepa/run_gepa_banking77.sh +112 -0
- examples/blog_posts/gepa/run_gepa_banking77_pipeline.sh +163 -0
- examples/blog_posts/gepa/task_apps.py +105 -0
- examples/blog_posts/gepa/test_gepa_local.sh +67 -0
- examples/blog_posts/gepa/verify_banking77_setup.sh +123 -0
- examples/blog_posts/mipro/README.md +415 -0
- examples/blog_posts/mipro/configs/banking77_mipro_local.toml +91 -0
- examples/blog_posts/mipro/configs/banking77_mipro_test.toml +87 -0
- examples/blog_posts/mipro/configs/banking77_pipeline_mipro_gemini_flash_lite_local.toml +98 -0
- examples/blog_posts/mipro/configs/banking77_pipeline_mipro_gpt41mini_local.toml +96 -0
- examples/blog_posts/mipro/configs/banking77_pipeline_mipro_local.toml +94 -0
- examples/blog_posts/mipro/configs/banking77_pipeline_mipro_test.toml +170 -0
- examples/blog_posts/mipro/deploy_banking77_pipeline_task_app.sh +59 -0
- examples/blog_posts/mipro/deploy_banking77_task_app.sh +41 -0
- examples/blog_posts/mipro/multi_step.md +79 -0
- examples/blog_posts/mipro/run_mipro_banking77.sh +191 -0
- examples/blog_posts/mipro/run_mipro_banking77_pipeline.sh +171 -0
- examples/blog_posts/mipro/run_mipro_banking77_pipeline_gemini_flash_lite.sh +177 -0
- examples/blog_posts/mipro/run_mipro_banking77_pipeline_gpt41mini.sh +173 -0
- examples/blog_posts/mipro/verify_banking77_setup.sh +117 -0
- examples/blog_posts/pokemon_vl/README.md +98 -0
- examples/blog_posts/pokemon_vl/configs/eval_gpt5nano.toml +26 -0
- examples/blog_posts/pokemon_vl/configs/eval_qwen3_vl.toml +27 -0
- examples/blog_posts/pokemon_vl/configs/eval_rl_final.toml +24 -0
- examples/blog_posts/pokemon_vl/configs/filter_high_reward.toml +10 -0
- examples/blog_posts/pokemon_vl/configs/train_rl_from_sft.toml +43 -0
- examples/blog_posts/pokemon_vl/configs/train_sft_qwen4b_vl.toml +40 -0
- examples/blog_posts/pokemon_vl/extract_images.py +239 -0
- examples/blog_posts/pokemon_vl/pokemon_vl_baseline.py +326 -0
- examples/blog_posts/pokemon_vl/run_eval_extract_images.py +209 -0
- examples/blog_posts/pokemon_vl/run_qwen_eval_extract_images.py +212 -0
- examples/blog_posts/pokemon_vl/text_box_analysis.md +106 -0
- examples/blog_posts/warming_up_to_rl/ARCHITECTURE.md +195 -0
- examples/blog_posts/warming_up_to_rl/FINAL_TEST_RESULTS.md +127 -0
- examples/blog_posts/warming_up_to_rl/INFERENCE_SUCCESS.md +132 -0
- examples/blog_posts/warming_up_to_rl/README.md +158 -0
- examples/blog_posts/warming_up_to_rl/SMOKE_TESTING.md +164 -0
- examples/blog_posts/warming_up_to_rl/SMOKE_TEST_COMPLETE.md +253 -0
- examples/blog_posts/warming_up_to_rl/configs/eval_baseline_qwen32b_10x20.toml +25 -0
- examples/blog_posts/warming_up_to_rl/configs/eval_ft_qwen4b.toml +25 -0
- examples/blog_posts/warming_up_to_rl/configs/eval_ft_qwen4b_10x20.toml +26 -0
- examples/blog_posts/warming_up_to_rl/configs/eval_groq_qwen32b.toml +25 -0
- examples/blog_posts/warming_up_to_rl/configs/eval_openai_gpt_oss_120b.toml +29 -0
- examples/blog_posts/warming_up_to_rl/configs/filter_high_reward_dataset.toml +10 -0
- examples/blog_posts/warming_up_to_rl/configs/smoke_test.toml +75 -0
- examples/blog_posts/warming_up_to_rl/configs/train_rl_from_sft.toml +91 -0
- examples/blog_posts/warming_up_to_rl/configs/train_sft_qwen4b.toml +40 -0
- examples/blog_posts/warming_up_to_rl/warming_up_to_rl_baseline.py +187 -0
- examples/crafter_debug_render.py +186 -0
- examples/dev/qwen3_32b_qlora_4xh100.toml +45 -0
- examples/gepa/banking77_pipeline_gepa.toml +96 -0
- examples/gepa/multi_stage_gepa_example.toml +84 -0
- examples/gepa/run_gepa_banking77_pipeline.sh +157 -0
- examples/multi_step/SFT_README.md +147 -0
- examples/multi_step/configs/README_verilog_rl.md +77 -0
- examples/multi_step/configs/VERILOG_REWARDS.md +103 -0
- examples/multi_step/configs/VERILOG_RL_CHECKLIST.md +196 -0
- examples/multi_step/configs/crafter_eval_synth_qwen4b.toml +35 -0
- examples/multi_step/configs/crafter_eval_text_only_groq_qwen32b.toml +36 -0
- examples/multi_step/configs/crafter_rl_outcome.toml +75 -0
- examples/multi_step/configs/crafter_rl_stepwise_hosted_judge.toml +145 -0
- examples/multi_step/configs/crafter_rl_stepwise_shaped.toml +84 -0
- examples/multi_step/configs/crafter_rl_stepwise_simple.toml +79 -0
- examples/multi_step/configs/crafter_rl_stepwise_simple_NEW_FORMAT.toml +105 -0
- examples/multi_step/configs/crafter_sft_qwen30b_lora.toml +62 -0
- examples/multi_step/configs/crafter_synth_backend.md +40 -0
- examples/multi_step/configs/verilog_eval_groq_qwen32b.toml +31 -0
- examples/multi_step/configs/verilog_eval_synth_qwen8b.toml +33 -0
- examples/multi_step/configs/verilog_rl_lora.toml +147 -0
- examples/multi_step/convert_traces_to_sft.py +84 -0
- examples/multi_step/crafter_rl_lora.md +70 -0
- examples/multi_step/judges/crafter_backend_judge.py +220 -0
- examples/multi_step/judges/verilog_backend_judge.py +234 -0
- examples/multi_step/readme.md +48 -0
- examples/multi_step/run_sft_qwen30b.sh +45 -0
- examples/multi_step/sse_metrics_streaming_notes.md +357 -0
- examples/multi_step/task_app_config_notes.md +494 -0
- examples/multi_step/verilog_rl_lora.md +218 -0
- examples/qwen_coder/README.md +102 -0
- examples/qwen_coder/_shared.py +113 -0
- examples/qwen_coder/configs/coder_lora_30b.toml +60 -0
- examples/qwen_coder/configs/coder_lora_4b.toml +61 -0
- examples/qwen_coder/configs/coder_lora_small.toml +57 -0
- examples/qwen_coder/generate_dataset.py +98 -0
- examples/qwen_coder/infer_ft_smoke.py +65 -0
- examples/qwen_coder/infer_prod_proxy.py +73 -0
- examples/qwen_coder/infer_via_synth.py +87 -0
- examples/qwen_coder/scripts/infer_coder.sh +19 -0
- examples/qwen_coder/scripts/train_coder_30b.sh +22 -0
- examples/qwen_coder/sft_full_17b.py +103 -0
- examples/qwen_coder/sft_lora_30b.py +110 -0
- examples/qwen_coder/subset_jsonl.py +39 -0
- examples/qwen_coder/todos.md +38 -0
- examples/qwen_coder/validate_jsonl.py +60 -0
- examples/qwen_vl/BUGS_AND_FIXES.md +232 -0
- examples/qwen_vl/IMAGE_VALIDATION_COMPLETE.md +271 -0
- examples/qwen_vl/IMAGE_VALIDATION_SUMMARY.md +260 -0
- examples/qwen_vl/INFERENCE_SFT_TESTS.md +412 -0
- examples/qwen_vl/NEXT_STEPS_2B.md +325 -0
- examples/qwen_vl/QUICKSTART.md +327 -0
- examples/qwen_vl/QUICKSTART_RL_VISION.md +110 -0
- examples/qwen_vl/README.md +152 -0
- examples/qwen_vl/RL_VISION_COMPLETE.md +475 -0
- examples/qwen_vl/RL_VISION_TESTING.md +333 -0
- examples/qwen_vl/SDK_VISION_INTEGRATION.md +328 -0
- examples/qwen_vl/SETUP_COMPLETE.md +274 -0
- examples/qwen_vl/VISION_TESTS_COMPLETE.md +489 -0
- examples/qwen_vl/VLM_PIPELINE_COMPLETE.md +242 -0
- examples/qwen_vl/__init__.py +2 -0
- examples/qwen_vl/collect_data_via_cli.md +415 -0
- examples/qwen_vl/collect_vision_traces.py +368 -0
- examples/qwen_vl/configs/crafter_rl_vision_qwen3vl4b.toml +110 -0
- examples/qwen_vl/configs/crafter_vlm_sft_example.toml +59 -0
- examples/qwen_vl/configs/eval_gpt4o_mini_vision.toml +26 -0
- examples/qwen_vl/configs/eval_gpt4o_vision_proper.toml +29 -0
- examples/qwen_vl/configs/eval_gpt5nano_vision.toml +26 -0
- examples/qwen_vl/configs/eval_qwen3vl_vision.toml +26 -0
- examples/qwen_vl/configs/filter_qwen3vl_sft.toml +49 -0
- examples/qwen_vl/configs/filter_vision_sft.toml +52 -0
- examples/qwen_vl/configs/filter_vision_test.toml +8 -0
- examples/qwen_vl/configs/sft_qwen3_vl_2b_test.toml +54 -0
- examples/qwen_vl/crafter_gpt5nano_agent.py +308 -0
- examples/qwen_vl/crafter_qwen_vl_agent.py +300 -0
- examples/qwen_vl/run_vision_comparison.sh +61 -0
- examples/qwen_vl/run_vision_sft_pipeline.sh +175 -0
- examples/qwen_vl/test_image_validation.py +201 -0
- examples/qwen_vl/test_sft_vision_data.py +110 -0
- examples/rl/README.md +169 -0
- examples/rl/configs/eval_base_qwen.toml +17 -0
- examples/rl/configs/eval_rl_qwen.toml +13 -0
- examples/rl/configs/rl_from_base_qwen.toml +62 -0
- examples/rl/configs/rl_from_base_qwen17.toml +80 -0
- examples/rl/configs/rl_from_ft_qwen.toml +37 -0
- examples/rl/download_dataset.py +80 -0
- examples/rl/run_eval.py +436 -0
- examples/rl/run_rl_and_save.py +111 -0
- examples/rl/task_app/README.md +21 -0
- {synth_ai/task/apps → examples/rl/task_app}/math_single_step.py +188 -50
- examples/rl/task_app/math_task_app.py +111 -0
- examples/run_crafter_demo.sh +10 -0
- examples/sdk_prompt_learning_example.py +55 -0
- examples/sft/README.md +139 -0
- examples/sft/configs/crafter_fft_qwen0p6b.toml +49 -0
- examples/sft/configs/crafter_lora_qwen0p6b.toml +49 -0
- examples/sft/evaluate.py +117 -0
- examples/sft/export_dataset.py +120 -0
- examples/sft/generate_traces.py +164 -0
- examples/swe/__init__.py +12 -0
- examples/swe/task_app/README.md +135 -0
- examples/swe/task_app/__init__.py +2 -0
- examples/swe/task_app/grpo_swe_mini.py +604 -0
- examples/swe/task_app/grpo_swe_mini_task_app.py +124 -0
- examples/swe/task_app/hosted/README.md +173 -0
- examples/swe/task_app/hosted/__init__.py +5 -0
- examples/swe/task_app/hosted/branching.py +143 -0
- examples/swe/task_app/hosted/environment_routes.py +1289 -0
- examples/swe/task_app/hosted/envs/__init__.py +1 -0
- examples/swe/task_app/hosted/envs/crafter/__init__.py +6 -0
- examples/swe/task_app/hosted/envs/crafter/app.py +1 -0
- examples/swe/task_app/hosted/envs/crafter/environment.py +522 -0
- examples/swe/task_app/hosted/envs/crafter/policy.py +478 -0
- examples/swe/task_app/hosted/envs/crafter/react_agent.py +108 -0
- examples/swe/task_app/hosted/envs/crafter/shared.py +305 -0
- examples/swe/task_app/hosted/envs/crafter/tools.py +47 -0
- examples/swe/task_app/hosted/envs/mini_swe/__init__.py +8 -0
- examples/swe/task_app/hosted/envs/mini_swe/environment.py +1191 -0
- examples/swe/task_app/hosted/envs/mini_swe/policy.py +355 -0
- examples/swe/task_app/hosted/envs/mini_swe/shared.py +83 -0
- examples/swe/task_app/hosted/envs/mini_swe/tools.py +96 -0
- examples/swe/task_app/hosted/hosted_app.py +204 -0
- examples/swe/task_app/hosted/inference/__init__.py +5 -0
- examples/swe/task_app/hosted/inference/openai_client.py +584 -0
- examples/swe/task_app/hosted/main.py +100 -0
- examples/swe/task_app/hosted/policy_routes.py +1094 -0
- examples/swe/task_app/hosted/registry.py +195 -0
- examples/swe/task_app/hosted/rollout.py +1905 -0
- examples/swe/task_app/hosted/storage/__init__.py +5 -0
- examples/swe/task_app/hosted/storage/volume.py +211 -0
- examples/swe/task_app/hosted/test_agents.py +161 -0
- examples/swe/task_app/hosted/test_service.py +136 -0
- examples/swe/task_app/hosted/utils.py +62 -0
- examples/swe/task_app/morph_backend.py +178 -0
- examples/task_apps/IMAGE_ONLY_EVAL_QUICKSTART.md +258 -0
- examples/task_apps/TESTING.md +275 -0
- examples/task_apps/banking77/__init__.py +6 -0
- examples/task_apps/banking77/banking77_task_app.py +912 -0
- examples/task_apps/banking77/deploy_wrapper.py +46 -0
- examples/task_apps/banking77_pipeline/__init__.py +6 -0
- examples/task_apps/banking77_pipeline/banking77_pipeline_task_app.py +489 -0
- examples/task_apps/banking77_pipeline/deploy_wrapper.py +50 -0
- examples/task_apps/crafter/CREATE_SFT_DATASET.md +286 -0
- examples/task_apps/crafter/EVAL_IMAGE_ONLY_RESULTS.md +152 -0
- examples/task_apps/crafter/FILTER_COMMAND_STATUS.md +187 -0
- examples/task_apps/crafter/FILTER_COMMAND_SUCCESS.md +281 -0
- examples/task_apps/crafter/QUERY_EXAMPLES.md +203 -0
- examples/task_apps/crafter/README_IMAGE_ONLY_EVAL.md +316 -0
- examples/task_apps/crafter/eval_image_only_gpt4o.toml +28 -0
- examples/task_apps/crafter/eval_text_only_groq_llama.toml +36 -0
- examples/task_apps/crafter/filter_sft_dataset.toml +16 -0
- examples/task_apps/crafter/task_app/README.md +42 -0
- examples/task_apps/crafter/task_app/__init__.py +5 -0
- examples/task_apps/crafter/task_app/grpo_crafter.py +1055 -0
- examples/task_apps/crafter/task_app/grpo_crafter_task_app.py +146 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/README.md +173 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/__init__.py +5 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/branching.py +143 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/environment_routes.py +1226 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/__init__.py +1 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/__init__.py +6 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/app.py +1 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/environment.py +532 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/policy.py +583 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/react_agent.py +122 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/shared.py +305 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/envs/crafter/tools.py +47 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/hosted_app.py +253 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/inference/__init__.py +5 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/inference/openai_client.py +999 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/main.py +100 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/policy_routes.py +1252 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/registry.py +195 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/rollout.py +2233 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/storage/__init__.py +5 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/storage/volume.py +211 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/test_agents.py +161 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/test_service.py +136 -0
- examples/task_apps/crafter/task_app/synth_envs_hosted/utils.py +411 -0
- examples/task_apps/dev/pokemon_emerald/__init__.py +2 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/README.md +811 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/__init__.py +120 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/action.py +160 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/memory.py +155 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/perception.py +69 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/planning.py +96 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/simple.py +1502 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/agent/system_prompt.py +4 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/grab_map.py +68 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/manual.py +216 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/__init__.py +35 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/emerald_utils.py +631 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/emulator.py +1544 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/enums.py +1428 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/memory_reader.py +4848 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/types.py +41 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pokemon_env/utils.py +298 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/pyproject.toml +95 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/run.py +204 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/server/app.py +2152 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/server/client.py +429 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/server/frame_server.py +155 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/README.md +78 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/run_tests.py +122 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_agent_direct.py +76 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_agent_prompts.py +413 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_battle_state_formatting.py +204 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_dialogue_detection.py +133 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_dialogue_detection_comprehensive.py +229 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_direct_agent_emulator.py +300 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_fps_adjustment_pytest.py +205 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_house_to_outside_direct.py +200 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_house_to_outside_transition.py +284 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_map_ground_truth_comparison.py +468 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_memory_map.py +575 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_server_map_validation.py +311 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests/test_torchic_state.py +259 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/anticheat.py +372 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/checkpoint.py +296 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/error_handler.py +275 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/get_local_ip.py +22 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/helpers.py +44 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/llm_logger.py +514 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/map_formatter.py +415 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/map_stitcher.py +1763 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/map_stitcher_singleton.py +33 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/map_trimmer.py +106 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/map_visualizer.py +334 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/ocr_dialogue.py +1020 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/recording.py +188 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/state_formatter.py +1481 -0
- examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils/vlm.py +862 -0
- examples/task_apps/dev/pokemon_emerald/modal_app.py +114 -0
- examples/task_apps/dev/pokemon_emerald/task_app/README.md +81 -0
- examples/task_apps/dev/pokemon_emerald/task_app/__init__.py +6 -0
- examples/task_apps/dev/pokemon_emerald/task_app/pokemon_emerald.py +685 -0
- examples/task_apps/enron/__init__.py +2 -0
- examples/task_apps/enron/eval_groq_qwen32.toml +16 -0
- examples/task_apps/enron/filter_sft.toml +5 -0
- examples/task_apps/enron/task_app/README.md +14 -0
- examples/task_apps/enron/task_app/__init__.py +1 -0
- examples/task_apps/enron/task_app/grpo_enron.py +906 -0
- examples/task_apps/enron/task_app/grpo_enron_task_app.py +146 -0
- examples/task_apps/enron/tests/__init__.py +4 -0
- examples/task_apps/enron/tests/conftest.py +115 -0
- examples/task_apps/enron/tests/integration/__init__.py +4 -0
- examples/task_apps/enron/tests/integration/test_enron_eval.py +179 -0
- examples/task_apps/enron/tests/integration/test_enron_rollout.py +135 -0
- examples/task_apps/enron/tests/unit/__init__.py +4 -0
- examples/task_apps/enron/tests/unit/test_enron_environment.py +126 -0
- examples/task_apps/gepa_benchmarks/__init__.py +7 -0
- examples/task_apps/gepa_benchmarks/common.py +260 -0
- examples/task_apps/gepa_benchmarks/hotpotqa_task_app.py +507 -0
- examples/task_apps/gepa_benchmarks/hover_task_app.py +436 -0
- examples/task_apps/gepa_benchmarks/ifbench_task_app.py +563 -0
- examples/task_apps/gepa_benchmarks/pupa_task_app.py +460 -0
- examples/task_apps/math/README.md +21 -0
- examples/task_apps/math/math_single_step.py +1000 -0
- examples/task_apps/math/math_task_app.py +115 -0
- examples/task_apps/pokemon_battle/__init__.py +2 -0
- examples/task_apps/pokemon_battle/modal_app.py +104 -0
- examples/task_apps/pokemon_battle/task_app/README.md +68 -0
- examples/task_apps/pokemon_battle/task_app/__init__.py +6 -0
- examples/task_apps/pokemon_battle/task_app/pokemon_showdown.py +932 -0
- examples/task_apps/pokemon_red/EVAL_IMAGE_ONLY_COMPLETE.md +283 -0
- examples/task_apps/pokemon_red/EVAL_IMAGE_ONLY_STATUS.md +155 -0
- examples/task_apps/pokemon_red/README.md +356 -0
- examples/task_apps/pokemon_red/README_IMAGE_ONLY_EVAL.md +428 -0
- examples/task_apps/pokemon_red/__init__.py +3 -0
- examples/task_apps/pokemon_red/eval_image_only_gpt4o.toml +30 -0
- examples/task_apps/pokemon_red/eval_pokemon_red_policy.py +224 -0
- examples/task_apps/pokemon_red/pallet_town_rl_config.toml +75 -0
- examples/task_apps/pokemon_red/task_app.py +1048 -0
- examples/task_apps/pokemon_red/test_pallet_town_rewards.py +193 -0
- examples/task_apps/sokoban/README.md +306 -0
- examples/task_apps/sokoban/__init__.py +3 -0
- examples/task_apps/sokoban/eval_groq_qwen32.toml +16 -0
- examples/task_apps/sokoban/eval_openai_gpt5.toml +16 -0
- examples/task_apps/sokoban/filter_sft.toml +5 -0
- examples/task_apps/sokoban/task_app.py +1058 -0
- examples/task_apps/sokoban/tests/__init__.py +4 -0
- examples/task_apps/sokoban/tests/conftest.py +113 -0
- examples/task_apps/sokoban/tests/integration/__init__.py +4 -0
- examples/task_apps/sokoban/tests/integration/test_sokoban_eval.py +57 -0
- examples/task_apps/sokoban/tests/integration/test_sokoban_rollout.py +198 -0
- examples/task_apps/sokoban/tests/unit/__init__.py +4 -0
- examples/task_apps/sokoban/tests/unit/test_sokoban_environment.py +114 -0
- examples/task_apps/verilog/__init__.py +1 -0
- examples/task_apps/verilog/eval_groq_qwen32b.toml +22 -0
- examples/task_apps/verilog/filter_sft.toml +5 -0
- examples/task_apps/verilog/task_app/README.md +12 -0
- examples/task_apps/verilog/task_app/__init__.py +1 -0
- examples/task_apps/verilog/task_app/grpo_verilog.py +1166 -0
- examples/task_apps/verilog/task_app/grpo_verilog_task_app.py +145 -0
- examples/task_apps/verilog/tests/__init__.py +4 -0
- examples/task_apps/verilog/tests/conftest.py +115 -0
- examples/task_apps/verilog/tests/integration/__init__.py +4 -0
- examples/task_apps/verilog/tests/integration/test_verilog_eval.py +181 -0
- examples/task_apps/verilog/tests/integration/test_verilog_rollout.py +55 -0
- examples/task_apps/verilog/tests/unit/__init__.py +4 -0
- examples/task_apps/verilog/tests/unit/test_verilog_scoring.py +118 -0
- examples/tunnel_gepa_banking77/README.md +106 -0
- examples/tunnel_gepa_banking77/banking77_gepa_tunnel.toml +95 -0
- examples/tunnel_gepa_banking77/keep_tunnel_running.py +60 -0
- examples/tunnel_gepa_banking77/run_gepa_with_tunnel.sh +226 -0
- examples/vlm/PROPOSAL.md +53 -0
- examples/vlm/README.md +68 -0
- examples/vlm/configs/crafter_vlm_gpt4o.toml +49 -0
- examples/vlm/crafter_image_only_agent.py +207 -0
- examples/vlm/crafter_openai_vlm_agent.py +275 -0
- examples/vlm/filter_image_rows.py +63 -0
- examples/vlm/run_crafter_vlm_benchmark.py +316 -0
- examples/warming_up_to_rl/_utils.py +92 -0
- examples/warming_up_to_rl/analyze_trace_db.py +422 -0
- examples/warming_up_to_rl/configs/crafter_fft.toml +53 -0
- examples/warming_up_to_rl/configs/crafter_fft_4b.toml +54 -0
- examples/warming_up_to_rl/configs/eval_fft_qwen4b.toml +22 -0
- examples/warming_up_to_rl/configs/eval_groq_qwen32b.toml +15 -0
- examples/warming_up_to_rl/configs/eval_modal_qwen4b.toml +24 -0
- examples/warming_up_to_rl/configs/eval_stepwise_complex.toml +35 -0
- examples/warming_up_to_rl/configs/eval_stepwise_consistent.toml +26 -0
- examples/warming_up_to_rl/configs/eval_stepwise_per_achievement.toml +36 -0
- examples/warming_up_to_rl/configs/eval_stepwise_simple.toml +32 -0
- examples/warming_up_to_rl/configs/rl_from_base_qwen4b.toml +85 -0
- examples/warming_up_to_rl/configs/rl_from_ft.toml +58 -0
- examples/warming_up_to_rl/export_trace_sft.py +837 -0
- examples/warming_up_to_rl/groq_test.py +97 -0
- examples/warming_up_to_rl/manage_secrets.py +131 -0
- examples/warming_up_to_rl/old/event_rewards.md +234 -0
- examples/warming_up_to_rl/old/notes.md +73 -0
- examples/warming_up_to_rl/readme.md +110 -0
- examples/warming_up_to_rl/run_eval.py +736 -0
- examples/warming_up_to_rl/run_fft_and_save.py +380 -0
- examples/warming_up_to_rl/run_local_rollout.py +239 -0
- examples/warming_up_to_rl/run_local_rollout_modal.py +248 -0
- examples/warming_up_to_rl/run_local_rollout_parallel.py +405 -0
- examples/warming_up_to_rl/run_local_rollout_traced.py +477 -0
- examples/warming_up_to_rl/run_rl_and_save.py +124 -0
- examples/warming_up_to_rl/run_rollout_remote.py +156 -0
- examples/warming_up_to_rl/task_app/README.md +42 -0
- examples/warming_up_to_rl/task_app/grpo_crafter.py +876 -0
- examples/warming_up_to_rl/task_app/grpo_crafter_task_app.py +135 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/README.md +173 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/__init__.py +5 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/branching.py +143 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/environment_routes.py +1226 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/__init__.py +1 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/__init__.py +6 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/app.py +1 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/environment.py +522 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/policy.py +454 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/react_agent.py +108 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/shared.py +305 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/tools.py +47 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/hosted_app.py +253 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/inference/__init__.py +5 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/inference/openai_client.py +729 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/main.py +100 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/policy_routes.py +1114 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/registry.py +195 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/rollout.py +1891 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/storage/__init__.py +5 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/storage/volume.py +211 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/test_agents.py +161 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/test_service.py +137 -0
- examples/warming_up_to_rl/task_app/synth_envs_hosted/utils.py +129 -0
- examples/workflows/math_rl/configs/eval_base_qwen.toml +15 -0
- examples/workflows/math_rl/configs/eval_rl_qwen.toml +11 -0
- examples/workflows/math_rl/configs/rl_from_base_qwen.toml +62 -0
- examples/workflows/math_rl/configs/rl_from_base_qwen17.toml +80 -0
- examples/workflows/math_rl/configs/rl_from_ft_qwen.toml +35 -0
- examples/workflows/math_rl/download_dataset.py +80 -0
- examples/workflows/math_rl/run_eval.py +436 -0
- examples/workflows/math_rl/run_rl_and_save.py +111 -0
- synth_ai/__init__.py +47 -23
- synth_ai/_utils/__init__.py +47 -0
- synth_ai/_utils/base_url.py +10 -0
- synth_ai/_utils/http.py +10 -0
- synth_ai/_utils/prompts.py +10 -0
- synth_ai/_utils/task_app_state.py +12 -0
- synth_ai/_utils/user_config.py +10 -0
- synth_ai/api/models/supported.py +514 -0
- synth_ai/api/train/__init__.py +60 -2
- synth_ai/api/train/builders.py +347 -39
- synth_ai/api/train/cli.py +895 -160
- synth_ai/api/train/config_finder.py +103 -25
- synth_ai/api/train/configs/__init__.py +65 -0
- synth_ai/api/train/configs/prompt_learning.py +496 -0
- synth_ai/api/train/configs/rl.py +188 -0
- synth_ai/api/train/configs/sft.py +99 -0
- synth_ai/api/train/configs/shared.py +81 -0
- synth_ai/api/train/env_resolver.py +70 -20
- synth_ai/api/train/pollers.py +29 -4
- synth_ai/api/train/prompt_learning.py +425 -0
- synth_ai/api/train/sft.py +390 -0
- synth_ai/api/train/supported_algos.py +147 -0
- synth_ai/api/train/task_app.py +6 -4
- synth_ai/api/train/utils.py +64 -52
- synth_ai/api/train/validators.py +1117 -0
- synth_ai/api/tunnel.py +49 -0
- synth_ai/auth/credentials.py +94 -0
- synth_ai/baseline/__init__.py +25 -0
- synth_ai/baseline/config.py +209 -0
- synth_ai/baseline/discovery.py +214 -0
- synth_ai/baseline/execution.py +146 -0
- synth_ai/cfgs.py +227 -0
- synth_ai/cli/__init__.py +85 -63
- synth_ai/cli/_modal_wrapper.py +31 -0
- synth_ai/cli/_storage.py +20 -0
- synth_ai/cli/_typer_patch.py +47 -0
- synth_ai/cli/_validate_task_app.py +29 -0
- synth_ai/cli/balance.py +16 -4
- synth_ai/cli/calc.py +36 -21
- synth_ai/cli/claude.py +70 -0
- synth_ai/cli/codex.py +267 -0
- synth_ai/cli/commands/__init__.py +18 -0
- synth_ai/cli/commands/baseline/__init__.py +12 -0
- synth_ai/cli/commands/baseline/core.py +637 -0
- synth_ai/cli/commands/baseline/list.py +93 -0
- synth_ai/cli/commands/demo/__init__.py +6 -0
- synth_ai/cli/commands/demo/core.py +163 -0
- synth_ai/cli/commands/eval/__init__.py +19 -0
- synth_ai/cli/commands/eval/core.py +1112 -0
- synth_ai/cli/commands/eval/errors.py +81 -0
- synth_ai/cli/commands/eval/validation.py +133 -0
- synth_ai/cli/commands/filter/__init__.py +12 -0
- synth_ai/cli/commands/filter/core.py +424 -0
- synth_ai/cli/commands/filter/errors.py +55 -0
- synth_ai/cli/commands/filter/validation.py +77 -0
- synth_ai/cli/commands/help/__init__.py +185 -0
- synth_ai/cli/commands/help/core.py +72 -0
- synth_ai/cli/commands/smoke/__init__.py +7 -0
- synth_ai/cli/commands/smoke/core.py +1437 -0
- synth_ai/cli/commands/status/__init__.py +66 -0
- synth_ai/cli/commands/status/client.py +192 -0
- synth_ai/cli/commands/status/config.py +92 -0
- synth_ai/cli/commands/status/errors.py +20 -0
- synth_ai/cli/commands/status/formatters.py +164 -0
- synth_ai/cli/commands/status/subcommands/__init__.py +9 -0
- synth_ai/cli/commands/status/subcommands/files.py +79 -0
- synth_ai/cli/commands/status/subcommands/jobs.py +334 -0
- synth_ai/cli/commands/status/subcommands/models.py +79 -0
- synth_ai/cli/commands/status/subcommands/pricing.py +22 -0
- synth_ai/cli/commands/status/subcommands/runs.py +81 -0
- synth_ai/cli/commands/status/subcommands/session.py +183 -0
- synth_ai/cli/commands/status/subcommands/summary.py +47 -0
- synth_ai/cli/commands/status/subcommands/usage.py +203 -0
- synth_ai/cli/commands/status/utils.py +114 -0
- synth_ai/cli/commands/train/__init__.py +53 -0
- synth_ai/cli/commands/train/core.py +21 -0
- synth_ai/cli/commands/train/errors.py +117 -0
- synth_ai/cli/commands/train/judge_schemas.py +200 -0
- synth_ai/cli/commands/train/judge_validation.py +305 -0
- synth_ai/cli/commands/train/validation.py +386 -0
- synth_ai/cli/demo.py +32 -140
- synth_ai/cli/deploy.py +233 -0
- synth_ai/cli/eval/__init__.py +36 -0
- synth_ai/cli/eval/core.py +5 -0
- synth_ai/cli/eval/errors.py +31 -0
- synth_ai/cli/eval/validation.py +5 -0
- synth_ai/cli/filter/__init__.py +28 -0
- synth_ai/cli/filter/core.py +5 -0
- synth_ai/cli/filter/errors.py +23 -0
- synth_ai/cli/filter/validation.py +5 -0
- synth_ai/cli/legacy_root_backup.py +28 -22
- synth_ai/cli/lib/__init__.py +10 -0
- synth_ai/cli/lib/task_app_discovery.py +7 -0
- synth_ai/cli/lib/task_app_env.py +518 -0
- synth_ai/cli/mcp.py +34 -0
- synth_ai/cli/modal_serve/__init__.py +12 -0
- synth_ai/cli/modal_serve/core.py +14 -0
- synth_ai/cli/modal_serve/errors.py +8 -0
- synth_ai/cli/modal_serve/validation.py +11 -0
- synth_ai/cli/opencode.py +256 -0
- synth_ai/cli/recent.py +13 -7
- synth_ai/cli/rl_demo.py +156 -116
- synth_ai/cli/root.py +131 -132
- synth_ai/cli/serve/__init__.py +12 -0
- synth_ai/cli/serve/core.py +14 -0
- synth_ai/cli/serve/errors.py +8 -0
- synth_ai/cli/serve/validation.py +11 -0
- synth_ai/cli/setup.py +49 -0
- synth_ai/cli/status.py +7 -125
- synth_ai/cli/task_app_deploy.py +7 -0
- synth_ai/cli/task_app_list.py +25 -0
- synth_ai/cli/task_app_modal_serve.py +11 -0
- synth_ai/cli/task_app_serve.py +11 -0
- synth_ai/cli/task_apps.py +2284 -257
- synth_ai/cli/traces.py +9 -5
- synth_ai/cli/train/__init__.py +12 -0
- synth_ai/cli/train/core.py +21 -0
- synth_ai/cli/train/errors.py +8 -0
- synth_ai/cli/train/validation.py +24 -0
- synth_ai/cli/train.py +5 -0
- synth_ai/cli/turso.py +73 -0
- synth_ai/cli/watch.py +13 -18
- synth_ai/demos/__init__.py +10 -0
- synth_ai/demos/core/__init__.py +28 -1
- synth_ai/demos/core/cli.py +579 -291
- synth_ai/demos/crafter/__init__.py +1 -0
- synth_ai/demos/crafter/crafter_fft_4b.toml +55 -0
- synth_ai/demos/crafter/grpo_crafter_task_app.py +185 -0
- synth_ai/demos/crafter/rl_from_base_qwen4b.toml +74 -0
- synth_ai/demos/demo_registry.py +176 -0
- synth_ai/demos/demo_task_apps/__init__.py +3 -3
- synth_ai/demos/demo_task_apps/core.py +64 -28
- synth_ai/demos/demo_task_apps/crafter/__init__.py +1 -0
- synth_ai/demos/demo_task_apps/crafter/configs/crafter_fft_4b.toml +53 -0
- synth_ai/demos/demo_task_apps/crafter/configs/rl_from_base_qwen4b.toml +73 -0
- synth_ai/demos/demo_task_apps/crafter/grpo_crafter_task_app.py +184 -0
- synth_ai/demos/demo_task_apps/math/_common.py +1 -2
- synth_ai/demos/demo_task_apps/math/app.py +2 -1
- synth_ai/demos/demo_task_apps/math/deploy_modal.py +3 -6
- synth_ai/demos/demo_task_apps/math/modal_task_app.py +185 -83
- synth_ai/demos/demo_task_apps/math/task_app_entry.py +0 -2
- synth_ai/demos/math/__init__.py +1 -0
- synth_ai/demos/math/_common.py +16 -0
- synth_ai/demos/math/app.py +38 -0
- synth_ai/demos/math/config.toml +76 -0
- synth_ai/demos/math/deploy_modal.py +54 -0
- synth_ai/demos/math/modal_task_app.py +703 -0
- synth_ai/demos/math/task_app_entry.py +51 -0
- synth_ai/environments/environment/core.py +7 -1
- synth_ai/environments/examples/bandit/engine.py +12 -5
- synth_ai/environments/examples/bandit/environment.py +0 -1
- synth_ai/environments/examples/bandit/taskset.py +4 -4
- synth_ai/environments/examples/crafter_classic/engine_deterministic_patch.py +7 -4
- synth_ai/environments/examples/crafter_classic/engine_serialization_patch_v3.py +9 -5
- synth_ai/environments/examples/crafter_classic/environment.py +93 -2
- synth_ai/environments/examples/crafter_classic/world_config_patch_simple.py +4 -3
- synth_ai/environments/examples/enron/engine.py +7 -2
- synth_ai/environments/examples/enron/environment.py +68 -0
- synth_ai/environments/examples/red/engine.py +60 -12
- synth_ai/environments/examples/red/engine_helpers/memory_map.py +7 -0
- synth_ai/environments/examples/red/engine_helpers/reward_components.py +151 -179
- synth_ai/environments/examples/red/engine_helpers/reward_library/pallet_town_progression.py +477 -0
- synth_ai/environments/examples/red/engine_helpers/state_extraction.py +32 -0
- synth_ai/environments/examples/red/environment.py +86 -0
- synth_ai/environments/examples/red/trace_hooks_v3.py +168 -0
- synth_ai/environments/examples/sokoban/taskset.py +116 -0
- synth_ai/environments/examples/verilog/engine.py +104 -12
- synth_ai/environments/examples/wordle/environment.py +0 -1
- synth_ai/environments/reproducibility/tree.py +5 -6
- synth_ai/environments/service/app.py +11 -12
- synth_ai/environments/service/core_routes.py +10 -9
- synth_ai/environments/stateful/engine.py +1 -1
- synth_ai/environments/tasks/core.py +1 -0
- synth_ai/environments/tasks/filters.py +5 -6
- synth_ai/environments/tasks/utils.py +4 -5
- synth_ai/evals/__init__.py +15 -0
- synth_ai/evals/base.py +14 -5
- synth_ai/evals/client.py +82 -0
- synth_ai/evals/types.py +42 -0
- synth_ai/http.py +8 -22
- synth_ai/http_client.py +45 -12
- synth_ai/inference/__init__.py +0 -2
- synth_ai/inference/client.py +21 -7
- synth_ai/jobs/client.py +129 -80
- synth_ai/judge_schemas.py +127 -0
- synth_ai/learning/__init__.py +51 -6
- synth_ai/learning/algorithms.py +14 -0
- synth_ai/learning/client.py +122 -30
- synth_ai/learning/config.py +2 -40
- synth_ai/learning/constants.py +0 -2
- synth_ai/learning/ft_client.py +4 -56
- synth_ai/learning/health.py +14 -8
- synth_ai/learning/jobs.py +43 -47
- synth_ai/learning/prompt_learning_client.py +276 -0
- synth_ai/learning/prompt_learning_types.py +185 -0
- synth_ai/{rl → learning/rl}/__init__.py +14 -5
- synth_ai/learning/rl/client.py +269 -0
- synth_ai/learning/rl/config.py +31 -0
- synth_ai/{rl → learning/rl}/contracts.py +5 -10
- synth_ai/{rl → learning/rl}/env_keys.py +45 -16
- synth_ai/learning/rl/secrets.py +13 -0
- synth_ai/learning/rl_client.py +2 -253
- synth_ai/learning/sft/__init__.py +29 -0
- synth_ai/learning/sft/client.py +68 -0
- synth_ai/learning/sft/config.py +270 -0
- synth_ai/learning/sft/data.py +698 -0
- synth_ai/learning/sse.py +25 -26
- synth_ai/learning/validators.py +29 -25
- synth_ai/mcp/__init__.py +5 -0
- synth_ai/mcp/__main__.py +8 -0
- synth_ai/mcp/main.py +254 -0
- synth_ai/mcp/setup.py +100 -0
- synth_ai/modal.py +257 -0
- synth_ai/pricing/__init__.py +3 -0
- synth_ai/pricing/model_pricing.py +64 -0
- synth_ai/session/__init__.py +75 -0
- synth_ai/session/client.py +383 -0
- synth_ai/session/constants.py +63 -0
- synth_ai/session/exceptions.py +105 -0
- synth_ai/session/manager.py +139 -0
- synth_ai/session/models.py +89 -0
- synth_ai/session/query.py +110 -0
- synth_ai/spec/__init__.py +46 -0
- synth_ai/spec/dataclasses.py +149 -0
- synth_ai/spec/loader.py +144 -0
- synth_ai/spec/serializer.py +199 -0
- synth_ai/spec/validation.py +250 -0
- synth_ai/streaming/__init__.py +29 -0
- synth_ai/streaming/config.py +94 -0
- synth_ai/streaming/handlers.py +589 -0
- synth_ai/streaming/streamer.py +320 -0
- synth_ai/streaming/types.py +95 -0
- synth_ai/task/__init__.py +50 -30
- synth_ai/task/apps/__init__.py +63 -19
- synth_ai/task/auth.py +35 -23
- synth_ai/task/client.py +15 -13
- synth_ai/task/config.py +261 -0
- synth_ai/task/contracts.py +165 -64
- synth_ai/task/datasets.py +9 -6
- synth_ai/task/errors.py +11 -10
- synth_ai/task/health.py +17 -11
- synth_ai/task/inference_api.py +101 -0
- synth_ai/task/json.py +58 -24
- synth_ai/task/proxy.py +59 -66
- synth_ai/task/rubrics/__init__.py +55 -0
- synth_ai/task/rubrics/loaders.py +156 -0
- synth_ai/task/rubrics/models.py +57 -0
- synth_ai/task/rubrics/scoring.py +116 -0
- synth_ai/task/rubrics/strict.py +149 -0
- synth_ai/task/rubrics.py +22 -15
- synth_ai/task/server.py +65 -31
- synth_ai/task/trace_correlation_helpers.py +328 -0
- synth_ai/task/tracing_utils.py +44 -28
- synth_ai/task/validators.py +449 -6
- synth_ai/task/vendors.py +5 -7
- synth_ai/tracing_v3/__init__.py +4 -0
- synth_ai/tracing_v3/abstractions.py +21 -4
- synth_ai/tracing_v3/config.py +167 -22
- synth_ai/tracing_v3/constants.py +21 -0
- synth_ai/tracing_v3/db_config.py +42 -29
- synth_ai/tracing_v3/decorators.py +80 -45
- synth_ai/tracing_v3/examples/basic_usage.py +15 -9
- synth_ai/tracing_v3/hooks.py +6 -4
- synth_ai/tracing_v3/llm_call_record_helpers.py +161 -61
- synth_ai/tracing_v3/migration_helper.py +1 -2
- synth_ai/tracing_v3/replica_sync.py +12 -7
- synth_ai/tracing_v3/serialization.py +130 -0
- synth_ai/tracing_v3/session_tracer.py +73 -16
- synth_ai/tracing_v3/storage/base.py +89 -1
- synth_ai/tracing_v3/storage/config.py +63 -16
- synth_ai/tracing_v3/storage/factory.py +11 -9
- synth_ai/tracing_v3/storage/utils.py +15 -11
- synth_ai/tracing_v3/trace_utils.py +317 -0
- synth_ai/tracing_v3/turso/__init__.py +8 -21
- synth_ai/tracing_v3/turso/daemon.py +123 -15
- synth_ai/tracing_v3/turso/models.py +5 -2
- synth_ai/tracing_v3/turso/native_manager.py +1293 -0
- synth_ai/tracing_v3/utils.py +5 -4
- synth_ai/tunnel.py +143 -0
- synth_ai/tunnel_deploy.py +278 -0
- synth_ai/types.py +8 -0
- synth_ai/urls.py +11 -0
- synth_ai/utils/__init__.py +166 -0
- synth_ai/utils/agents.py +74 -0
- synth_ai/utils/apps.py +152 -0
- synth_ai/utils/base_url.py +94 -0
- synth_ai/utils/bin.py +39 -0
- synth_ai/utils/claude.py +36 -0
- synth_ai/utils/cli.py +284 -0
- synth_ai/utils/config.py +81 -0
- synth_ai/utils/env.py +346 -0
- synth_ai/utils/errors.py +85 -0
- synth_ai/utils/http.py +172 -0
- synth_ai/utils/json.py +72 -0
- synth_ai/utils/log_filter.py +99 -0
- synth_ai/utils/logging.py +198 -0
- synth_ai/utils/modal.py +299 -0
- synth_ai/utils/paths.py +95 -0
- synth_ai/utils/process.py +233 -0
- synth_ai/utils/prompts.py +39 -0
- synth_ai/utils/sqld.py +122 -0
- synth_ai/utils/ssl.py +25 -0
- synth_ai/utils/task_app_discovery.py +882 -0
- synth_ai/utils/task_app_env.py +186 -0
- synth_ai/utils/task_app_state.py +318 -0
- synth_ai/utils/tunnel/__init__.py +12 -0
- synth_ai/utils/tunnel/config.py +55 -0
- synth_ai/utils/user_config.py +137 -0
- synth_ai/uvicorn.py +77 -0
- synth_ai-0.2.23.dev3.dist-info/METADATA +357 -0
- synth_ai-0.2.23.dev3.dist-info/RECORD +983 -0
- {synth_ai-0.2.9.dev0.dist-info → synth_ai-0.2.23.dev3.dist-info}/entry_points.txt +0 -1
- {synth_ai-0.2.9.dev0.dist-info → synth_ai-0.2.23.dev3.dist-info}/top_level.txt +1 -0
- synth_ai/cli/man.py +0 -106
- synth_ai/core/experiment.py +0 -15
- synth_ai/core/system.py +0 -15
- synth_ai/demo_registry.py +0 -258
- synth_ai/environments/examples/sokoban/units/astar_common.py +0 -95
- synth_ai/experimental/synth_oss.py +0 -446
- synth_ai/handshake.py +0 -107
- synth_ai/install_sqld.sh +0 -40
- synth_ai/learning/offline/dpo.py +0 -0
- synth_ai/learning/offline/providers.py +0 -7
- synth_ai/learning/offline/sft.py +0 -0
- synth_ai/learning/offline/shared.py +0 -0
- synth_ai/learning/online/grpo.py +0 -0
- synth_ai/learning/online/irft.py +0 -0
- synth_ai/learning/prompts/banking77_injection_eval.py +0 -168
- synth_ai/learning/prompts/gepa.py +0 -0
- synth_ai/learning/prompts/hello_world_in_context_injection_ex.py +0 -213
- synth_ai/learning/prompts/mipro.py +0 -289
- synth_ai/learning/prompts/random_search.py +0 -246
- synth_ai/learning/prompts/run_mipro_banking77.py +0 -172
- synth_ai/learning/prompts/run_random_search_banking77.py +0 -324
- synth_ai/lm/__init__.py +0 -51
- synth_ai/lm/caching/constants.py +0 -6
- synth_ai/lm/caching/dbs.py +0 -0
- synth_ai/lm/caching/ephemeral.py +0 -102
- synth_ai/lm/caching/handler.py +0 -137
- synth_ai/lm/caching/initialize.py +0 -11
- synth_ai/lm/caching/persistent.py +0 -114
- synth_ai/lm/config.py +0 -110
- synth_ai/lm/constants.py +0 -32
- synth_ai/lm/core/__init__.py +0 -8
- synth_ai/lm/core/all.py +0 -73
- synth_ai/lm/core/exceptions.py +0 -7
- synth_ai/lm/core/main.py +0 -319
- synth_ai/lm/core/main_v3.py +0 -594
- synth_ai/lm/core/synth_models.py +0 -48
- synth_ai/lm/core/vendor_clients.py +0 -188
- synth_ai/lm/cost/monitor.py +0 -1
- synth_ai/lm/cost/statefulness.py +0 -1
- synth_ai/lm/injection.py +0 -80
- synth_ai/lm/overrides.py +0 -206
- synth_ai/lm/provider_support/__init__.py +0 -8
- synth_ai/lm/provider_support/anthropic.py +0 -972
- synth_ai/lm/provider_support/openai.py +0 -1139
- synth_ai/lm/provider_support/suppress_logging.py +0 -31
- synth_ai/lm/structured_outputs/handler.py +0 -440
- synth_ai/lm/structured_outputs/inject.py +0 -297
- synth_ai/lm/structured_outputs/rehabilitate.py +0 -185
- synth_ai/lm/tools/__init__.py +0 -3
- synth_ai/lm/tools/base.py +0 -172
- synth_ai/lm/unified_interface.py +0 -202
- synth_ai/lm/vendors/base.py +0 -81
- synth_ai/lm/vendors/core/anthropic_api.py +0 -387
- synth_ai/lm/vendors/core/gemini_api.py +0 -292
- synth_ai/lm/vendors/core/mistral_api.py +0 -322
- synth_ai/lm/vendors/core/openai_api.py +0 -225
- synth_ai/lm/vendors/core/synth_dev_api.py +0 -0
- synth_ai/lm/vendors/local/ollama.py +0 -0
- synth_ai/lm/vendors/openai_standard.py +0 -780
- synth_ai/lm/vendors/openai_standard_responses.py +0 -256
- synth_ai/lm/vendors/retries.py +0 -22
- synth_ai/lm/vendors/supported/custom_endpoint.py +0 -417
- synth_ai/lm/vendors/supported/deepseek.py +0 -69
- synth_ai/lm/vendors/supported/grok.py +0 -75
- synth_ai/lm/vendors/supported/groq.py +0 -16
- synth_ai/lm/vendors/supported/ollama.py +0 -15
- synth_ai/lm/vendors/supported/openrouter.py +0 -74
- synth_ai/lm/vendors/supported/together.py +0 -11
- synth_ai/lm/vendors/synth_client.py +0 -808
- synth_ai/lm/warmup.py +0 -186
- synth_ai/rl/secrets.py +0 -19
- synth_ai/scripts/verify_rewards.py +0 -100
- synth_ai/task/apps/grpo_crafter.py +0 -438
- synth_ai/tracing/__init__.py +0 -30
- synth_ai/tracing_v1/__init__.py +0 -33
- synth_ai/tracing_v3/turso/manager.py +0 -774
- synth_ai/v0/tracing/abstractions.py +0 -224
- synth_ai/v0/tracing/base_client.py +0 -91
- synth_ai/v0/tracing/client_manager.py +0 -131
- synth_ai/v0/tracing/config.py +0 -142
- synth_ai/v0/tracing/context.py +0 -146
- synth_ai/v0/tracing/decorators.py +0 -682
- synth_ai/v0/tracing/events/__init__.py +0 -0
- synth_ai/v0/tracing/events/manage.py +0 -147
- synth_ai/v0/tracing/events/scope.py +0 -86
- synth_ai/v0/tracing/events/store.py +0 -228
- synth_ai/v0/tracing/immediate_client.py +0 -151
- synth_ai/v0/tracing/local.py +0 -18
- synth_ai/v0/tracing/log_client_base.py +0 -73
- synth_ai/v0/tracing/retry_queue.py +0 -186
- synth_ai/v0/tracing/trackers.py +0 -515
- synth_ai/v0/tracing/upload.py +0 -512
- synth_ai/v0/tracing/utils.py +0 -9
- synth_ai/v0/tracing_v1/__init__.py +0 -16
- synth_ai/v0/tracing_v1/abstractions.py +0 -224
- synth_ai/v0/tracing_v1/base_client.py +0 -91
- synth_ai/v0/tracing_v1/client_manager.py +0 -131
- synth_ai/v0/tracing_v1/config.py +0 -142
- synth_ai/v0/tracing_v1/context.py +0 -146
- synth_ai/v0/tracing_v1/decorators.py +0 -703
- synth_ai/v0/tracing_v1/events/__init__.py +0 -0
- synth_ai/v0/tracing_v1/events/manage.py +0 -147
- synth_ai/v0/tracing_v1/events/scope.py +0 -86
- synth_ai/v0/tracing_v1/events/store.py +0 -228
- synth_ai/v0/tracing_v1/immediate_client.py +0 -151
- synth_ai/v0/tracing_v1/local.py +0 -18
- synth_ai/v0/tracing_v1/log_client_base.py +0 -73
- synth_ai/v0/tracing_v1/retry_queue.py +0 -186
- synth_ai/v0/tracing_v1/trackers.py +0 -515
- synth_ai/v0/tracing_v1/upload.py +0 -527
- synth_ai/v0/tracing_v1/utils.py +0 -9
- synth_ai/zyk/__init__.py +0 -30
- synth_ai-0.2.9.dev0.dist-info/METADATA +0 -131
- synth_ai-0.2.9.dev0.dist-info/RECORD +0 -444
- {synth_ai/lm/caching → examples/task_apps}/__init__.py +0 -0
- {synth_ai/lm/cost → examples/task_apps/crafter}/__init__.py +0 -0
- {synth_ai/lm/structured_outputs → examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/server}/__init__.py +0 -0
- {synth_ai/lm/vendors → examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/tests}/__init__.py +0 -0
- {synth_ai/lm/vendors/core → examples/task_apps/dev/pokemon_emerald/external/pokeagent-speedrun/utils}/__init__.py +0 -0
- {synth_ai/lm/vendors/local → examples/task_apps/math}/__init__.py +0 -0
- {synth_ai/lm/vendors/supported → examples/workflows}/__init__.py +0 -0
- {synth_ai/v0/tracing → examples/workflows/math_rl}/__init__.py +0 -0
- /synth_ai/{compound/cais.py → cli/__main__.py} +0 -0
- /synth_ai/{learning/filtering.py → py.typed} +0 -0
- {synth_ai-0.2.9.dev0.dist-info → synth_ai-0.2.23.dev3.dist-info}/WHEEL +0 -0
- {synth_ai-0.2.9.dev0.dist-info → synth_ai-0.2.23.dev3.dist-info}/licenses/LICENSE +0 -0
|
@@ -0,0 +1,317 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
import json
|
|
4
|
+
import sqlite3
|
|
5
|
+
from collections.abc import Sequence
|
|
6
|
+
from dataclasses import dataclass
|
|
7
|
+
from typing import Any
|
|
8
|
+
|
|
9
|
+
Row = sqlite3.Row
|
|
10
|
+
|
|
11
|
+
|
|
12
|
+
def connect(db_path: str | bytes | int) -> sqlite3.Connection:
|
|
13
|
+
conn = sqlite3.connect(db_path)
|
|
14
|
+
conn.row_factory = sqlite3.Row
|
|
15
|
+
return conn
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
def _json_load(value: Any) -> Any:
|
|
19
|
+
if value is None:
|
|
20
|
+
return None
|
|
21
|
+
if isinstance(value, dict | list):
|
|
22
|
+
return value
|
|
23
|
+
if isinstance(value, bytes | bytearray):
|
|
24
|
+
value = value.decode("utf-8", errors="ignore")
|
|
25
|
+
try:
|
|
26
|
+
return json.loads(value)
|
|
27
|
+
except Exception:
|
|
28
|
+
return value
|
|
29
|
+
|
|
30
|
+
|
|
31
|
+
def fetch_crafter_sessions(
|
|
32
|
+
conn: sqlite3.Connection,
|
|
33
|
+
*,
|
|
34
|
+
limit: int,
|
|
35
|
+
metadata_filter: str | None = None,
|
|
36
|
+
session_ids: Sequence[str] | None = None,
|
|
37
|
+
min_event_count: int = 0,
|
|
38
|
+
) -> list[str]:
|
|
39
|
+
if session_ids:
|
|
40
|
+
placeholders = ",".join("?" for _ in session_ids)
|
|
41
|
+
rows = conn.execute(
|
|
42
|
+
f"""
|
|
43
|
+
SELECT session_id
|
|
44
|
+
FROM session_traces
|
|
45
|
+
WHERE session_id IN ({placeholders})
|
|
46
|
+
ORDER BY created_at DESC
|
|
47
|
+
""",
|
|
48
|
+
tuple(session_ids),
|
|
49
|
+
).fetchall()
|
|
50
|
+
return [row["session_id"] for row in rows]
|
|
51
|
+
|
|
52
|
+
params: list[Any] = []
|
|
53
|
+
where_clauses: list[str] = []
|
|
54
|
+
if metadata_filter:
|
|
55
|
+
where_clauses.append("session_traces.metadata LIKE ?")
|
|
56
|
+
params.append(f"%{metadata_filter}%")
|
|
57
|
+
where_sql = ""
|
|
58
|
+
if where_clauses:
|
|
59
|
+
where_sql = "WHERE " + " AND ".join(where_clauses)
|
|
60
|
+
|
|
61
|
+
having_sql = ""
|
|
62
|
+
if min_event_count > 0:
|
|
63
|
+
having_sql = "HAVING COUNT(events.id) >= ?"
|
|
64
|
+
params.append(min_event_count)
|
|
65
|
+
|
|
66
|
+
query = f"""
|
|
67
|
+
SELECT session_traces.session_id
|
|
68
|
+
FROM session_traces
|
|
69
|
+
LEFT JOIN events ON session_traces.session_id = events.session_id
|
|
70
|
+
{where_sql}
|
|
71
|
+
GROUP BY session_traces.session_id
|
|
72
|
+
{having_sql}
|
|
73
|
+
ORDER BY session_traces.created_at DESC
|
|
74
|
+
LIMIT ?
|
|
75
|
+
"""
|
|
76
|
+
rows = conn.execute(query, (*params, limit)).fetchall()
|
|
77
|
+
return [row["session_id"] for row in rows]
|
|
78
|
+
|
|
79
|
+
|
|
80
|
+
def load_session_trace(conn: sqlite3.Connection, session_id: str) -> dict[str, Any]:
|
|
81
|
+
session_row = conn.execute(
|
|
82
|
+
"""
|
|
83
|
+
SELECT session_id, created_at, metadata
|
|
84
|
+
FROM session_traces
|
|
85
|
+
WHERE session_id = ?
|
|
86
|
+
""",
|
|
87
|
+
(session_id,),
|
|
88
|
+
).fetchone()
|
|
89
|
+
if not session_row:
|
|
90
|
+
raise ValueError(f"Session {session_id} not found")
|
|
91
|
+
|
|
92
|
+
timesteps = conn.execute(
|
|
93
|
+
"""
|
|
94
|
+
SELECT step_id,
|
|
95
|
+
step_index,
|
|
96
|
+
turn_number,
|
|
97
|
+
started_at,
|
|
98
|
+
completed_at,
|
|
99
|
+
step_metadata
|
|
100
|
+
FROM session_timesteps
|
|
101
|
+
WHERE session_id = ?
|
|
102
|
+
ORDER BY step_index ASC
|
|
103
|
+
""",
|
|
104
|
+
(session_id,),
|
|
105
|
+
).fetchall()
|
|
106
|
+
|
|
107
|
+
event_rows = conn.execute(
|
|
108
|
+
"""
|
|
109
|
+
SELECT *
|
|
110
|
+
FROM events
|
|
111
|
+
WHERE session_id = ?
|
|
112
|
+
ORDER BY event_time ASC, id ASC
|
|
113
|
+
""",
|
|
114
|
+
(session_id,),
|
|
115
|
+
).fetchall()
|
|
116
|
+
|
|
117
|
+
message_rows = conn.execute(
|
|
118
|
+
"""
|
|
119
|
+
SELECT *
|
|
120
|
+
FROM messages
|
|
121
|
+
WHERE session_id = ?
|
|
122
|
+
ORDER BY event_time ASC, id ASC
|
|
123
|
+
""",
|
|
124
|
+
(session_id,),
|
|
125
|
+
).fetchall()
|
|
126
|
+
|
|
127
|
+
event_rewards = conn.execute(
|
|
128
|
+
"""
|
|
129
|
+
SELECT *
|
|
130
|
+
FROM event_rewards
|
|
131
|
+
WHERE session_id = ?
|
|
132
|
+
ORDER BY turn_number ASC, id ASC
|
|
133
|
+
""",
|
|
134
|
+
(session_id,),
|
|
135
|
+
).fetchall()
|
|
136
|
+
|
|
137
|
+
outcome_rewards = conn.execute(
|
|
138
|
+
"""
|
|
139
|
+
SELECT *
|
|
140
|
+
FROM outcome_rewards
|
|
141
|
+
WHERE session_id = ?
|
|
142
|
+
ORDER BY created_at ASC
|
|
143
|
+
""",
|
|
144
|
+
(session_id,),
|
|
145
|
+
).fetchall()
|
|
146
|
+
|
|
147
|
+
metadata = _json_load(session_row["metadata"]) or {}
|
|
148
|
+
if isinstance(metadata, dict):
|
|
149
|
+
episode_id = metadata.get("episode_id")
|
|
150
|
+
if episode_id is not None and not isinstance(episode_id, str):
|
|
151
|
+
metadata["episode_id"] = str(episode_id)
|
|
152
|
+
|
|
153
|
+
events_payload = [
|
|
154
|
+
{
|
|
155
|
+
"id": row["id"],
|
|
156
|
+
"event_type": row["event_type"],
|
|
157
|
+
"system_instance_id": row["system_instance_id"],
|
|
158
|
+
"time_record": {
|
|
159
|
+
"event_time": row["event_time"],
|
|
160
|
+
"message_time": row["message_time"],
|
|
161
|
+
"created_at": row["created_at"],
|
|
162
|
+
},
|
|
163
|
+
"model_name": row["model_name"],
|
|
164
|
+
"provider": row["provider"],
|
|
165
|
+
"input_tokens": row["input_tokens"],
|
|
166
|
+
"output_tokens": row["output_tokens"],
|
|
167
|
+
"total_tokens": row["total_tokens"],
|
|
168
|
+
"cost_usd": row["cost_usd"],
|
|
169
|
+
"latency_ms": row["latency_ms"],
|
|
170
|
+
"span_id": row["span_id"],
|
|
171
|
+
"trace_id": row["trace_id"],
|
|
172
|
+
"call_records": _json_load(row["call_records"]) or [],
|
|
173
|
+
"reward": row["reward"],
|
|
174
|
+
"terminated": row["terminated"],
|
|
175
|
+
"truncated": row["truncated"],
|
|
176
|
+
"system_state_before": _json_load(row["system_state_before"]),
|
|
177
|
+
"system_state_after": _json_load(row["system_state_after"]),
|
|
178
|
+
"metadata": _json_load(row["metadata"]) or {},
|
|
179
|
+
"event_metadata": _json_load(row["event_metadata"]),
|
|
180
|
+
}
|
|
181
|
+
for row in event_rows
|
|
182
|
+
]
|
|
183
|
+
|
|
184
|
+
messages_payload = [
|
|
185
|
+
{
|
|
186
|
+
"id": row["id"],
|
|
187
|
+
"message_type": row["message_type"],
|
|
188
|
+
"content": row["content"],
|
|
189
|
+
"time_record": {
|
|
190
|
+
"event_time": row["event_time"],
|
|
191
|
+
"message_time": row["message_time"],
|
|
192
|
+
"timestamp": row["timestamp"],
|
|
193
|
+
},
|
|
194
|
+
"metadata": _json_load(row["metadata"]) or {},
|
|
195
|
+
}
|
|
196
|
+
for row in message_rows
|
|
197
|
+
]
|
|
198
|
+
|
|
199
|
+
trace: dict[str, Any] = {
|
|
200
|
+
"session_id": session_row["session_id"],
|
|
201
|
+
"created_at": session_row["created_at"],
|
|
202
|
+
"metadata": metadata,
|
|
203
|
+
"session_time_steps": [
|
|
204
|
+
{
|
|
205
|
+
"step_id": row["step_id"],
|
|
206
|
+
"step_index": row["step_index"],
|
|
207
|
+
"turn_number": row["turn_number"],
|
|
208
|
+
"started_at": row["started_at"],
|
|
209
|
+
"completed_at": row["completed_at"],
|
|
210
|
+
"metadata": _json_load(row["step_metadata"]) or {},
|
|
211
|
+
}
|
|
212
|
+
for row in timesteps
|
|
213
|
+
],
|
|
214
|
+
"event_history": events_payload,
|
|
215
|
+
"events": events_payload,
|
|
216
|
+
"markov_blanket_message_history": messages_payload,
|
|
217
|
+
"messages": messages_payload,
|
|
218
|
+
"event_rewards": [
|
|
219
|
+
{
|
|
220
|
+
"id": row["id"],
|
|
221
|
+
"event_id": row["event_id"],
|
|
222
|
+
"turn_number": row["turn_number"],
|
|
223
|
+
"reward_value": row["reward_value"],
|
|
224
|
+
"reward_type": row["reward_type"],
|
|
225
|
+
"key": row["key"],
|
|
226
|
+
"annotation": _json_load(row["annotation"]) or {},
|
|
227
|
+
"source": row["source"],
|
|
228
|
+
"created_at": row["created_at"],
|
|
229
|
+
}
|
|
230
|
+
for row in event_rewards
|
|
231
|
+
],
|
|
232
|
+
"outcome_rewards": [
|
|
233
|
+
{
|
|
234
|
+
"id": row["id"],
|
|
235
|
+
"total_reward": row["total_reward"],
|
|
236
|
+
"reward_metadata": _json_load(row["reward_metadata"]) or {},
|
|
237
|
+
"created_at": row["created_at"],
|
|
238
|
+
}
|
|
239
|
+
for row in outcome_rewards
|
|
240
|
+
],
|
|
241
|
+
}
|
|
242
|
+
return trace
|
|
243
|
+
|
|
244
|
+
|
|
245
|
+
@dataclass
|
|
246
|
+
class DeterministicMetrics:
|
|
247
|
+
session_id: str
|
|
248
|
+
unique_achievement_reward: float
|
|
249
|
+
achievement_reward: float
|
|
250
|
+
outcome_total_reward: float
|
|
251
|
+
unique_achievement_count: int
|
|
252
|
+
final_achievement_count: int
|
|
253
|
+
|
|
254
|
+
|
|
255
|
+
def compute_deterministic_metrics(conn: sqlite3.Connection, session_id: str) -> DeterministicMetrics:
|
|
256
|
+
event_rows = conn.execute(
|
|
257
|
+
"""
|
|
258
|
+
SELECT reward_type, reward_value, annotation
|
|
259
|
+
FROM event_rewards
|
|
260
|
+
WHERE session_id = ?
|
|
261
|
+
""",
|
|
262
|
+
(session_id,),
|
|
263
|
+
).fetchall()
|
|
264
|
+
|
|
265
|
+
unique_total = 0.0
|
|
266
|
+
all_total = 0.0
|
|
267
|
+
unique_achievements: set[str] = set()
|
|
268
|
+
|
|
269
|
+
for row in event_rows:
|
|
270
|
+
reward_type = row["reward_type"]
|
|
271
|
+
value = float(row["reward_value"] or 0.0)
|
|
272
|
+
if reward_type == "unique_achievement_delta":
|
|
273
|
+
unique_total += value
|
|
274
|
+
annotation = _json_load(row["annotation"]) or {}
|
|
275
|
+
for name in annotation.get("new_unique") or []:
|
|
276
|
+
if isinstance(name, str):
|
|
277
|
+
unique_achievements.add(name)
|
|
278
|
+
elif reward_type == "achievement_delta":
|
|
279
|
+
all_total += value
|
|
280
|
+
|
|
281
|
+
outcome_rows = conn.execute(
|
|
282
|
+
"""
|
|
283
|
+
SELECT total_reward, reward_metadata
|
|
284
|
+
FROM outcome_rewards
|
|
285
|
+
WHERE session_id = ?
|
|
286
|
+
""",
|
|
287
|
+
(session_id,),
|
|
288
|
+
).fetchall()
|
|
289
|
+
|
|
290
|
+
outcome_total = 0.0
|
|
291
|
+
final_achievements: set[str] = set()
|
|
292
|
+
for row in outcome_rows:
|
|
293
|
+
outcome_total += float(row["total_reward"] or 0.0)
|
|
294
|
+
metadata = _json_load(row["reward_metadata"]) or {}
|
|
295
|
+
for name in metadata.get("achievements") or []:
|
|
296
|
+
if isinstance(name, str):
|
|
297
|
+
final_achievements.add(name)
|
|
298
|
+
|
|
299
|
+
return DeterministicMetrics(
|
|
300
|
+
session_id=session_id,
|
|
301
|
+
unique_achievement_reward=unique_total,
|
|
302
|
+
achievement_reward=all_total,
|
|
303
|
+
outcome_total_reward=outcome_total,
|
|
304
|
+
unique_achievement_count=len(unique_achievements),
|
|
305
|
+
final_achievement_count=len(final_achievements),
|
|
306
|
+
)
|
|
307
|
+
|
|
308
|
+
|
|
309
|
+
__all__ = [
|
|
310
|
+
"DeterministicMetrics",
|
|
311
|
+
"compute_deterministic_metrics",
|
|
312
|
+
"connect",
|
|
313
|
+
"fetch_crafter_sessions",
|
|
314
|
+
"load_session_trace",
|
|
315
|
+
]
|
|
316
|
+
|
|
317
|
+
|
|
@@ -1,25 +1,12 @@
|
|
|
1
|
-
"""Turso
|
|
1
|
+
"""Turso integration package for tracing v3."""
|
|
2
2
|
|
|
3
|
-
from .
|
|
4
|
-
from .
|
|
5
|
-
Base,
|
|
6
|
-
Event,
|
|
7
|
-
Experiment,
|
|
8
|
-
Message,
|
|
9
|
-
SessionTimestep,
|
|
10
|
-
SessionTrace,
|
|
11
|
-
System,
|
|
12
|
-
SystemVersion,
|
|
13
|
-
)
|
|
3
|
+
from .daemon import SqldDaemon, get_daemon, start_sqld, stop_sqld
|
|
4
|
+
from .native_manager import NativeLibsqlTraceManager
|
|
14
5
|
|
|
15
6
|
__all__ = [
|
|
16
|
-
"
|
|
17
|
-
"
|
|
18
|
-
"
|
|
19
|
-
"
|
|
20
|
-
"
|
|
21
|
-
"Message",
|
|
22
|
-
"Experiment",
|
|
23
|
-
"System",
|
|
24
|
-
"SystemVersion",
|
|
7
|
+
"SqldDaemon",
|
|
8
|
+
"NativeLibsqlTraceManager",
|
|
9
|
+
"get_daemon",
|
|
10
|
+
"start_sqld",
|
|
11
|
+
"stop_sqld",
|
|
25
12
|
]
|
|
@@ -1,39 +1,123 @@
|
|
|
1
1
|
"""sqld daemon management utilities."""
|
|
2
2
|
|
|
3
|
+
import logging
|
|
4
|
+
import os
|
|
3
5
|
import pathlib
|
|
4
6
|
import shutil
|
|
5
7
|
import subprocess
|
|
8
|
+
import sys
|
|
6
9
|
import time
|
|
7
10
|
|
|
8
11
|
import requests
|
|
12
|
+
from requests import RequestException
|
|
9
13
|
|
|
10
14
|
from ..config import CONFIG
|
|
11
15
|
|
|
16
|
+
logger = logging.getLogger(__name__)
|
|
17
|
+
|
|
12
18
|
|
|
13
19
|
class SqldDaemon:
|
|
14
20
|
"""Manages local sqld daemon lifecycle."""
|
|
15
21
|
|
|
16
|
-
def __init__(
|
|
22
|
+
def __init__(
|
|
23
|
+
self,
|
|
24
|
+
db_path: str | None = None,
|
|
25
|
+
http_port: int | None = None,
|
|
26
|
+
hrana_port: int | None = None,
|
|
27
|
+
binary_path: str | None = None,
|
|
28
|
+
):
|
|
17
29
|
"""Initialize sqld daemon manager.
|
|
18
30
|
|
|
19
31
|
Args:
|
|
20
32
|
db_path: Path to database file (uses config default if not provided)
|
|
21
|
-
http_port: HTTP port for
|
|
33
|
+
http_port: HTTP port for health/API (uses config default + 1 if not provided)
|
|
34
|
+
hrana_port: Hrana WebSocket port for libsql connections (uses config default if not provided)
|
|
22
35
|
binary_path: Path to sqld binary (auto-detected if not provided)
|
|
23
36
|
"""
|
|
24
37
|
self.db_path = db_path or CONFIG.sqld_db_path
|
|
25
|
-
self.
|
|
38
|
+
self.hrana_port = hrana_port or CONFIG.sqld_http_port # Main port for libsql://
|
|
39
|
+
self.http_port = http_port or (self.hrana_port + 1) # HTTP API on next port
|
|
26
40
|
self.binary_path = binary_path or self._find_binary()
|
|
27
|
-
self.process: subprocess.Popen | None = None
|
|
41
|
+
self.process: subprocess.Popen[str] | None = None
|
|
28
42
|
|
|
29
43
|
def _find_binary(self) -> str:
|
|
30
|
-
"""Find sqld binary in PATH.
|
|
44
|
+
"""Find sqld binary in PATH, auto-installing if needed.
|
|
45
|
+
|
|
46
|
+
Search order:
|
|
47
|
+
1. CONFIG.sqld_binary in PATH
|
|
48
|
+
2. libsql-server in PATH
|
|
49
|
+
3. Common install locations (~/.turso/bin, /usr/local/bin, etc.)
|
|
50
|
+
4. Auto-install via synth_ai.utils.sqld (if interactive terminal)
|
|
51
|
+
|
|
52
|
+
Returns:
|
|
53
|
+
Path to sqld binary
|
|
54
|
+
|
|
55
|
+
Raises:
|
|
56
|
+
RuntimeError: If binary not found and auto-install fails/disabled
|
|
57
|
+
"""
|
|
58
|
+
# Check PATH first
|
|
31
59
|
binary = shutil.which(CONFIG.sqld_binary) or shutil.which("libsql-server")
|
|
32
|
-
if
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
60
|
+
if binary:
|
|
61
|
+
logger.debug(f"Found sqld binary in PATH: {binary}")
|
|
62
|
+
return binary
|
|
63
|
+
|
|
64
|
+
# Check common install locations
|
|
65
|
+
try:
|
|
66
|
+
from synth_ai.utils.sqld import find_sqld_binary
|
|
67
|
+
binary = find_sqld_binary()
|
|
68
|
+
if binary:
|
|
69
|
+
logger.debug(f"Found sqld binary in common location: {binary}")
|
|
70
|
+
return binary
|
|
71
|
+
except ImportError:
|
|
72
|
+
logger.debug("synth_ai.utils.sqld not available, skipping common location check")
|
|
73
|
+
|
|
74
|
+
# Try auto-install if enabled and interactive
|
|
75
|
+
auto_install_enabled = os.getenv("SYNTH_AI_AUTO_INSTALL_SQLD", "true").lower() == "true"
|
|
76
|
+
|
|
77
|
+
if auto_install_enabled and sys.stdin.isatty():
|
|
78
|
+
try:
|
|
79
|
+
from synth_ai.utils.sqld import install_sqld
|
|
80
|
+
logger.info("sqld binary not found. Attempting automatic installation...")
|
|
81
|
+
|
|
82
|
+
# Use click if available for better UX, otherwise proceed automatically
|
|
83
|
+
try:
|
|
84
|
+
import click
|
|
85
|
+
if not click.confirm(
|
|
86
|
+
"sqld not found. Install automatically via Homebrew?",
|
|
87
|
+
default=True
|
|
88
|
+
):
|
|
89
|
+
raise RuntimeError("User declined automatic installation")
|
|
90
|
+
except ImportError:
|
|
91
|
+
# click not available, auto-install without prompt
|
|
92
|
+
logger.info("Installing sqld automatically (non-interactive mode)")
|
|
93
|
+
|
|
94
|
+
binary = install_sqld()
|
|
95
|
+
logger.info(f"Successfully installed sqld to: {binary}")
|
|
96
|
+
return binary
|
|
97
|
+
|
|
98
|
+
except Exception as exc:
|
|
99
|
+
logger.warning(f"Auto-install failed: {exc}")
|
|
100
|
+
# Fall through to error message below
|
|
101
|
+
elif not auto_install_enabled:
|
|
102
|
+
logger.debug("Auto-install disabled via SYNTH_AI_AUTO_INSTALL_SQLD=false")
|
|
103
|
+
elif not sys.stdin.isatty():
|
|
104
|
+
logger.debug("Non-interactive terminal, skipping auto-install prompt")
|
|
105
|
+
|
|
106
|
+
# If we get here, all methods failed
|
|
107
|
+
raise RuntimeError(
|
|
108
|
+
"sqld binary not found. Install using one of these methods:\n"
|
|
109
|
+
"\n"
|
|
110
|
+
"Quick install (recommended):\n"
|
|
111
|
+
" synth-ai turso\n"
|
|
112
|
+
"\n"
|
|
113
|
+
"Manual install:\n"
|
|
114
|
+
" brew install turso-tech/tools/sqld\n"
|
|
115
|
+
" # or\n"
|
|
116
|
+
" curl -sSfL https://get.tur.so/install.sh | bash && turso dev\n"
|
|
117
|
+
"\n"
|
|
118
|
+
"For CI/CD environments:\n"
|
|
119
|
+
" Set SYNTH_AI_AUTO_INSTALL_SQLD=false and pre-install sqld"
|
|
120
|
+
)
|
|
37
121
|
|
|
38
122
|
def start(self, wait_for_ready: bool = True) -> subprocess.Popen:
|
|
39
123
|
"""Start the sqld daemon."""
|
|
@@ -47,6 +131,8 @@ class SqldDaemon:
|
|
|
47
131
|
self.binary_path,
|
|
48
132
|
"--db-path",
|
|
49
133
|
str(db_file),
|
|
134
|
+
"--hrana-listen-addr",
|
|
135
|
+
f"127.0.0.1:{self.hrana_port}",
|
|
50
136
|
"--http-listen-addr",
|
|
51
137
|
f"127.0.0.1:{self.http_port}",
|
|
52
138
|
]
|
|
@@ -74,11 +160,11 @@ class SqldDaemon:
|
|
|
74
160
|
response = requests.get(health_url, timeout=1)
|
|
75
161
|
if response.status_code == 200:
|
|
76
162
|
return
|
|
77
|
-
except
|
|
163
|
+
except RequestException:
|
|
78
164
|
pass
|
|
79
165
|
|
|
80
166
|
# Check if process crashed
|
|
81
|
-
if self.process.poll() is not None:
|
|
167
|
+
if self.process and self.process.poll() is not None:
|
|
82
168
|
stdout, stderr = self.process.communicate()
|
|
83
169
|
raise RuntimeError(
|
|
84
170
|
f"sqld daemon failed to start:\nstdout: {stdout}\nstderr: {stderr}"
|
|
@@ -106,6 +192,14 @@ class SqldDaemon:
|
|
|
106
192
|
"""Check if daemon is running."""
|
|
107
193
|
return self.process is not None and self.process.poll() is None
|
|
108
194
|
|
|
195
|
+
def get_hrana_port(self) -> int:
|
|
196
|
+
"""Get the Hrana WebSocket port for libsql:// connections."""
|
|
197
|
+
return self.hrana_port
|
|
198
|
+
|
|
199
|
+
def get_http_port(self) -> int:
|
|
200
|
+
"""Get the HTTP API port for health checks."""
|
|
201
|
+
return self.http_port
|
|
202
|
+
|
|
109
203
|
def __enter__(self):
|
|
110
204
|
"""Context manager entry."""
|
|
111
205
|
self.start()
|
|
@@ -120,13 +214,27 @@ class SqldDaemon:
|
|
|
120
214
|
_daemon: SqldDaemon | None = None
|
|
121
215
|
|
|
122
216
|
|
|
123
|
-
def start_sqld(
|
|
124
|
-
|
|
217
|
+
def start_sqld(
|
|
218
|
+
db_path: str | None = None,
|
|
219
|
+
port: int | None = None,
|
|
220
|
+
hrana_port: int | None = None,
|
|
221
|
+
http_port: int | None = None,
|
|
222
|
+
) -> SqldDaemon:
|
|
223
|
+
"""Start a global sqld daemon instance.
|
|
224
|
+
|
|
225
|
+
Args:
|
|
226
|
+
db_path: Path to database file
|
|
227
|
+
port: Legacy parameter - used as hrana_port if hrana_port not specified
|
|
228
|
+
hrana_port: Hrana WebSocket port for libsql:// connections
|
|
229
|
+
http_port: HTTP API port for health checks
|
|
230
|
+
"""
|
|
125
231
|
global _daemon
|
|
126
232
|
if _daemon and _daemon.is_running():
|
|
127
233
|
return _daemon
|
|
128
234
|
|
|
129
|
-
|
|
235
|
+
# Support legacy 'port' parameter by using it as hrana_port
|
|
236
|
+
final_hrana_port = hrana_port or port
|
|
237
|
+
_daemon = SqldDaemon(db_path=db_path, hrana_port=final_hrana_port, http_port=http_port)
|
|
130
238
|
_daemon.start()
|
|
131
239
|
return _daemon
|
|
132
240
|
|
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
from __future__ import annotations
|
|
2
1
|
"""SQLAlchemy declarative models for tracing v3."""
|
|
3
2
|
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
4
5
|
import json
|
|
5
6
|
|
|
6
7
|
from sqlalchemy import (
|
|
@@ -452,7 +453,9 @@ class EventReward(Base):
|
|
|
452
453
|
message_id = Column(Integer, ForeignKey("messages.id"), nullable=True)
|
|
453
454
|
turn_number = Column(Integer, nullable=True)
|
|
454
455
|
reward_value = Column(Float, nullable=False, default=0.0)
|
|
455
|
-
reward_type = Column(
|
|
456
|
+
reward_type = Column(
|
|
457
|
+
String, nullable=True
|
|
458
|
+
) # shaped | sparse | achievement | penalty | evaluator | human
|
|
456
459
|
key = Column(String, nullable=True) # e.g., achievement name
|
|
457
460
|
annotation = Column(JSONText) # free-form JSON
|
|
458
461
|
source = Column(String, nullable=True) # environment | runner | evaluator | human
|