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 +1,518 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
import contextlib
|
|
4
|
+
import os
|
|
5
|
+
import signal
|
|
6
|
+
import subprocess
|
|
7
|
+
import time
|
|
8
|
+
from collections.abc import Sequence
|
|
9
|
+
from pathlib import Path
|
|
10
|
+
|
|
11
|
+
import click
|
|
12
|
+
from click.exceptions import Abort
|
|
13
|
+
from synth_ai.config.base_url import PROD_BASE_URL_DEFAULT
|
|
14
|
+
from synth_ai.task.apps import TaskAppEntry
|
|
15
|
+
|
|
16
|
+
REPO_ROOT = Path(__file__).resolve().parents[3]
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
def load_env_files_into_process(paths: Sequence[str]) -> None:
|
|
20
|
+
"""Load key/value pairs from .env-style files into the current process."""
|
|
21
|
+
|
|
22
|
+
for path_str in paths:
|
|
23
|
+
try:
|
|
24
|
+
content = Path(path_str).expanduser().read_text()
|
|
25
|
+
except Exception:
|
|
26
|
+
continue
|
|
27
|
+
for line in content.splitlines():
|
|
28
|
+
if not line or line.lstrip().startswith("#") or "=" not in line:
|
|
29
|
+
continue
|
|
30
|
+
key, value = line.split("=", 1)
|
|
31
|
+
key = key.strip()
|
|
32
|
+
val = value.strip().strip('"').strip("'")
|
|
33
|
+
if not key:
|
|
34
|
+
continue
|
|
35
|
+
current = os.environ.get(key, "")
|
|
36
|
+
if not current.strip():
|
|
37
|
+
os.environ[key] = val
|
|
38
|
+
|
|
39
|
+
|
|
40
|
+
def _collect_env_candidates(base_dir: Path) -> list[Path]:
|
|
41
|
+
cwd = Path.cwd()
|
|
42
|
+
candidates: list[Path] = []
|
|
43
|
+
|
|
44
|
+
candidates.extend(sorted(cwd.glob("**/*.env")))
|
|
45
|
+
|
|
46
|
+
repo_candidates = sorted(REPO_ROOT.glob("**/*.env"))
|
|
47
|
+
for candidate in repo_candidates:
|
|
48
|
+
if candidate not in candidates:
|
|
49
|
+
candidates.append(candidate)
|
|
50
|
+
|
|
51
|
+
if base_dir not in (cwd, REPO_ROOT):
|
|
52
|
+
base_candidates = sorted(base_dir.glob("**/*.env"))
|
|
53
|
+
for candidate in base_candidates:
|
|
54
|
+
if candidate not in candidates:
|
|
55
|
+
candidates.append(candidate)
|
|
56
|
+
|
|
57
|
+
return candidates
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
def determine_env_files(entry: TaskAppEntry, user_env_files: Sequence[str]) -> list[Path]:
|
|
61
|
+
"""Resolve env file paths for a task app invocation."""
|
|
62
|
+
|
|
63
|
+
resolved: list[Path] = []
|
|
64
|
+
for candidate in user_env_files:
|
|
65
|
+
path = Path(candidate).expanduser()
|
|
66
|
+
if not path.exists():
|
|
67
|
+
raise click.ClickException(f"Env file not found: {path}")
|
|
68
|
+
resolved.append(path)
|
|
69
|
+
if resolved:
|
|
70
|
+
return resolved
|
|
71
|
+
|
|
72
|
+
candidates = _collect_env_candidates(Path.cwd())
|
|
73
|
+
if not candidates:
|
|
74
|
+
raise click.ClickException("No env file found. Pass --env-file explicitly.")
|
|
75
|
+
|
|
76
|
+
click.echo("Select env file to load:")
|
|
77
|
+
for idx, path in enumerate(candidates, start=1):
|
|
78
|
+
click.echo(f" {idx}) {path.resolve()}")
|
|
79
|
+
choice = click.prompt("Enter choice", type=click.IntRange(1, len(candidates)), default=1)
|
|
80
|
+
selected = candidates[choice - 1]
|
|
81
|
+
return [selected]
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
def resolve_env_paths_for_script(script_path: Path, explicit: Sequence[str]) -> list[Path]:
|
|
85
|
+
"""Resolve env files for a standalone Modal script."""
|
|
86
|
+
|
|
87
|
+
if explicit:
|
|
88
|
+
resolved = []
|
|
89
|
+
for candidate in explicit:
|
|
90
|
+
path = Path(candidate).expanduser()
|
|
91
|
+
if not path.exists():
|
|
92
|
+
raise click.ClickException(f"Env file not found: {path}")
|
|
93
|
+
resolved.append(path)
|
|
94
|
+
return resolved
|
|
95
|
+
|
|
96
|
+
candidates = _collect_env_candidates(script_path.parent.resolve())
|
|
97
|
+
if not candidates:
|
|
98
|
+
created = interactive_create_env(script_path.parent)
|
|
99
|
+
if created is None:
|
|
100
|
+
raise click.ClickException("Env file required (--env-file) for this task app")
|
|
101
|
+
return [created]
|
|
102
|
+
|
|
103
|
+
click.echo("Select env file to load:")
|
|
104
|
+
for idx, path in enumerate(candidates, start=1):
|
|
105
|
+
click.echo(f" {idx}) {path.resolve()}")
|
|
106
|
+
choice = click.prompt("Enter choice", type=click.IntRange(1, len(candidates)), default=1)
|
|
107
|
+
return [candidates[choice - 1]]
|
|
108
|
+
|
|
109
|
+
|
|
110
|
+
def ensure_port_free(port: int, host: str, *, force: bool) -> None:
|
|
111
|
+
"""Ensure a TCP port is not in use, optionally killing processes if --force."""
|
|
112
|
+
|
|
113
|
+
import socket # local import to avoid unnecessary dependency during CLI import
|
|
114
|
+
|
|
115
|
+
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
|
|
116
|
+
in_use = sock.connect_ex((host, port)) == 0
|
|
117
|
+
if not in_use:
|
|
118
|
+
return
|
|
119
|
+
|
|
120
|
+
try:
|
|
121
|
+
out = subprocess.run(
|
|
122
|
+
["lsof", "-ti", f"TCP:{port}"],
|
|
123
|
+
capture_output=True,
|
|
124
|
+
text=True,
|
|
125
|
+
check=False,
|
|
126
|
+
)
|
|
127
|
+
pids = [pid for pid in out.stdout.strip().splitlines() if pid]
|
|
128
|
+
except FileNotFoundError:
|
|
129
|
+
pids = []
|
|
130
|
+
|
|
131
|
+
if not force:
|
|
132
|
+
message = f"Port {port} appears to be in use"
|
|
133
|
+
if pids:
|
|
134
|
+
message += f" (PIDs: {', '.join(pids)})"
|
|
135
|
+
raise click.ClickException(message)
|
|
136
|
+
|
|
137
|
+
for pid in pids:
|
|
138
|
+
try:
|
|
139
|
+
os.kill(int(pid), signal.SIGTERM)
|
|
140
|
+
except Exception as exc:
|
|
141
|
+
raise click.ClickException(f"Failed to terminate PID {pid}: {exc}") from exc
|
|
142
|
+
|
|
143
|
+
time.sleep(0.5)
|
|
144
|
+
|
|
145
|
+
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
|
|
146
|
+
still_in_use = sock.connect_ex((host, port)) == 0
|
|
147
|
+
|
|
148
|
+
if still_in_use:
|
|
149
|
+
for pid in pids:
|
|
150
|
+
try:
|
|
151
|
+
os.kill(int(pid), signal.SIGKILL)
|
|
152
|
+
except Exception as exc:
|
|
153
|
+
raise click.ClickException(f"Failed to force terminate PID {pid}: {exc}") from exc
|
|
154
|
+
time.sleep(0.5)
|
|
155
|
+
|
|
156
|
+
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
|
|
157
|
+
if sock.connect_ex((host, port)) == 0:
|
|
158
|
+
raise click.ClickException(
|
|
159
|
+
f"Port {port} is still in use after attempting to terminate processes."
|
|
160
|
+
)
|
|
161
|
+
|
|
162
|
+
|
|
163
|
+
def save_to_env_file(env_path: Path, key: str, value: str) -> None:
|
|
164
|
+
"""Save or update a key/value pair in a .env file."""
|
|
165
|
+
|
|
166
|
+
try:
|
|
167
|
+
existing_lines = env_path.read_text().splitlines() if env_path.exists() else []
|
|
168
|
+
except Exception as exc:
|
|
169
|
+
raise click.ClickException(f"Failed to read {env_path}: {exc}") from exc
|
|
170
|
+
|
|
171
|
+
env_path.parent.mkdir(parents=True, exist_ok=True)
|
|
172
|
+
|
|
173
|
+
key_updated = False
|
|
174
|
+
updated_lines: list[str] = []
|
|
175
|
+
for line in existing_lines:
|
|
176
|
+
if line.strip().startswith(f"{key}="):
|
|
177
|
+
updated_lines.append(f"{key}={value}")
|
|
178
|
+
key_updated = True
|
|
179
|
+
else:
|
|
180
|
+
updated_lines.append(line)
|
|
181
|
+
|
|
182
|
+
if key_updated:
|
|
183
|
+
env_path.write_text("\n".join(updated_lines) + "\n")
|
|
184
|
+
click.echo(f"Updated {key} in {env_path}")
|
|
185
|
+
return
|
|
186
|
+
|
|
187
|
+
with env_path.open("a", encoding="utf-8") as handle:
|
|
188
|
+
if existing_lines and existing_lines[-1].strip():
|
|
189
|
+
handle.write("\n")
|
|
190
|
+
handle.write(f"{key}={value}\n")
|
|
191
|
+
click.echo(f"Saved {key} to {env_path}")
|
|
192
|
+
|
|
193
|
+
|
|
194
|
+
def persist_env_api_key(env_api_key: str, env_paths: Sequence[Path] | None) -> None:
|
|
195
|
+
"""Persist ENVIRONMENT_API_KEY to provided .env files (or demo directory .env)."""
|
|
196
|
+
|
|
197
|
+
targets: list[Path] = []
|
|
198
|
+
seen: set[Path] = set()
|
|
199
|
+
for path in env_paths or ():
|
|
200
|
+
try:
|
|
201
|
+
resolved = Path(path).resolve()
|
|
202
|
+
except Exception:
|
|
203
|
+
continue
|
|
204
|
+
if resolved in seen:
|
|
205
|
+
continue
|
|
206
|
+
seen.add(resolved)
|
|
207
|
+
targets.append(resolved)
|
|
208
|
+
|
|
209
|
+
if not targets:
|
|
210
|
+
demo_dir = Path(os.environ.get("SYNTH_DEMO_DIR") or Path.cwd())
|
|
211
|
+
targets.append((demo_dir / ".env").resolve())
|
|
212
|
+
|
|
213
|
+
for target in targets:
|
|
214
|
+
save_to_env_file(target, "ENVIRONMENT_API_KEY", env_api_key)
|
|
215
|
+
|
|
216
|
+
|
|
217
|
+
def _load_dotenv_if_present(env_file: Path) -> None:
|
|
218
|
+
try:
|
|
219
|
+
from dotenv import load_dotenv
|
|
220
|
+
except Exception:
|
|
221
|
+
return
|
|
222
|
+
|
|
223
|
+
with contextlib.suppress(Exception):
|
|
224
|
+
load_dotenv(env_file, override=False)
|
|
225
|
+
|
|
226
|
+
|
|
227
|
+
def validate_required_env_keys() -> None:
|
|
228
|
+
"""Ensure ENVIRONMENT_API_KEY (and optional Groq key) are set, prompting if needed."""
|
|
229
|
+
|
|
230
|
+
demo_base = Path(os.environ.get("SYNTH_DEMO_DIR") or Path.cwd())
|
|
231
|
+
env_file = demo_base / ".env"
|
|
232
|
+
|
|
233
|
+
if env_file.exists():
|
|
234
|
+
_load_dotenv_if_present(env_file)
|
|
235
|
+
|
|
236
|
+
env_api_key = os.environ.get("ENVIRONMENT_API_KEY", "").strip()
|
|
237
|
+
if not env_api_key:
|
|
238
|
+
env_api_key = click.prompt(
|
|
239
|
+
"Please enter your RL Environment API key",
|
|
240
|
+
type=str,
|
|
241
|
+
).strip()
|
|
242
|
+
if not env_api_key:
|
|
243
|
+
raise click.ClickException("RL Environment API key is required to start the server")
|
|
244
|
+
os.environ["ENVIRONMENT_API_KEY"] = env_api_key
|
|
245
|
+
save_to_env_file(env_file, "ENVIRONMENT_API_KEY", env_api_key)
|
|
246
|
+
|
|
247
|
+
groq_api_key = os.environ.get("GROQ_API_KEY", "").strip()
|
|
248
|
+
if not groq_api_key:
|
|
249
|
+
click.echo("\nInference API key configuration:")
|
|
250
|
+
click.echo("This workflow requires a Groq API key.")
|
|
251
|
+
groq_api_key = click.prompt(
|
|
252
|
+
"Groq API key (or press Enter to skip)",
|
|
253
|
+
type=str,
|
|
254
|
+
default="",
|
|
255
|
+
show_default=False,
|
|
256
|
+
).strip()
|
|
257
|
+
if groq_api_key:
|
|
258
|
+
os.environ["GROQ_API_KEY"] = groq_api_key
|
|
259
|
+
save_to_env_file(env_file, "GROQ_API_KEY", groq_api_key)
|
|
260
|
+
|
|
261
|
+
|
|
262
|
+
def print_demo_next_steps_if_applicable() -> None:
|
|
263
|
+
"""Print helpful instructions when operating inside a demo directory."""
|
|
264
|
+
|
|
265
|
+
try:
|
|
266
|
+
from synth_ai.demos.demo_task_apps.core import load_demo_dir
|
|
267
|
+
|
|
268
|
+
cwd = Path.cwd().resolve()
|
|
269
|
+
demo_dir = load_demo_dir()
|
|
270
|
+
|
|
271
|
+
if demo_dir and Path(demo_dir).resolve() == cwd and (cwd / "run_local_rollout_traced.py").exists():
|
|
272
|
+
click.echo("\n" + "=" * 60)
|
|
273
|
+
click.echo("Next step: Collect traced rollouts")
|
|
274
|
+
click.echo("=" * 60)
|
|
275
|
+
click.echo("\nIn another terminal, run:")
|
|
276
|
+
click.echo(f" cd {cwd}")
|
|
277
|
+
click.echo(" uv run python run_local_rollout_traced.py")
|
|
278
|
+
click.echo("\nRun this 5-10 times to collect diverse traces.")
|
|
279
|
+
click.echo("=" * 60 + "\n")
|
|
280
|
+
except Exception:
|
|
281
|
+
pass
|
|
282
|
+
|
|
283
|
+
|
|
284
|
+
def _preview_secret(value: str) -> str:
|
|
285
|
+
if len(value) <= 10:
|
|
286
|
+
return value
|
|
287
|
+
return f"{value[:6]}...{value[-4:]}"
|
|
288
|
+
|
|
289
|
+
|
|
290
|
+
def preflight_env_key(env_paths: Sequence[Path] | None = None, *, crash_on_failure: bool = False) -> None:
|
|
291
|
+
"""Ensure ENVIRONMENT_API_KEY exists and attempt to upload it to the backend."""
|
|
292
|
+
|
|
293
|
+
raw_backend = (
|
|
294
|
+
os.environ.get("BACKEND_BASE_URL")
|
|
295
|
+
or os.environ.get("SYNTH_BASE_URL")
|
|
296
|
+
or f"{PROD_BASE_URL_DEFAULT}/api"
|
|
297
|
+
)
|
|
298
|
+
backend_base = raw_backend.rstrip("/")
|
|
299
|
+
if not backend_base.endswith("/api"):
|
|
300
|
+
backend_base += "/api"
|
|
301
|
+
|
|
302
|
+
synth_key = os.environ.get("SYNTH_API_KEY") or ""
|
|
303
|
+
env_api_key = (
|
|
304
|
+
os.environ.get("ENVIRONMENT_API_KEY")
|
|
305
|
+
or os.environ.get("DEV_ENVIRONMENT_API_KEY")
|
|
306
|
+
or ""
|
|
307
|
+
).strip()
|
|
308
|
+
|
|
309
|
+
def _mint_key() -> str | None:
|
|
310
|
+
try:
|
|
311
|
+
from synth_ai.learning.rl.secrets import mint_environment_api_key
|
|
312
|
+
|
|
313
|
+
key = mint_environment_api_key()
|
|
314
|
+
os.environ["ENVIRONMENT_API_KEY"] = key
|
|
315
|
+
os.environ.setdefault("DEV_ENVIRONMENT_API_KEY", key)
|
|
316
|
+
click.echo(f"[preflight] minted ENVIRONMENT_API_KEY ({_preview_secret(key)})")
|
|
317
|
+
return key
|
|
318
|
+
except Exception as exc: # pragma: no cover - defensive fallback
|
|
319
|
+
if crash_on_failure:
|
|
320
|
+
raise click.ClickException(
|
|
321
|
+
f"[CRITICAL] Failed to mint ENVIRONMENT_API_KEY: {exc}"
|
|
322
|
+
) from exc
|
|
323
|
+
click.echo(
|
|
324
|
+
f"[WARN] Failed to mint ENVIRONMENT_API_KEY automatically ({exc}); proceeding without upload"
|
|
325
|
+
)
|
|
326
|
+
return None
|
|
327
|
+
|
|
328
|
+
minted = False
|
|
329
|
+
if not env_api_key:
|
|
330
|
+
env_api_key = _mint_key() or ""
|
|
331
|
+
minted = bool(env_api_key)
|
|
332
|
+
|
|
333
|
+
if env_api_key and minted and env_paths:
|
|
334
|
+
persist_env_api_key(env_api_key, env_paths)
|
|
335
|
+
|
|
336
|
+
if not synth_key.strip():
|
|
337
|
+
click.echo("[preflight] SYNTH_API_KEY not set; skipping backend preflight.")
|
|
338
|
+
return
|
|
339
|
+
|
|
340
|
+
if not env_api_key:
|
|
341
|
+
click.echo("[preflight] ENVIRONMENT_API_KEY missing; continuing without verification.")
|
|
342
|
+
return
|
|
343
|
+
|
|
344
|
+
try:
|
|
345
|
+
import base64
|
|
346
|
+
|
|
347
|
+
import httpx
|
|
348
|
+
from nacl.public import PublicKey, SealedBox
|
|
349
|
+
except Exception: # pragma: no cover - optional deps
|
|
350
|
+
click.echo("[preflight] Optional crypto dependencies missing; skipping upload.")
|
|
351
|
+
return
|
|
352
|
+
|
|
353
|
+
try:
|
|
354
|
+
with httpx.Client(timeout=15.0, headers={"Authorization": f"Bearer {synth_key}"}) as client:
|
|
355
|
+
click.echo(f"[preflight] backend={backend_base}")
|
|
356
|
+
click.echo("[preflight] fetching public key…")
|
|
357
|
+
rpk = client.get(f"{backend_base.rstrip('/')}/v1/crypto/public-key")
|
|
358
|
+
if rpk.status_code != 200:
|
|
359
|
+
click.echo(f"[preflight] public key fetch failed with {rpk.status_code}; skipping upload")
|
|
360
|
+
return
|
|
361
|
+
pk = (rpk.json() or {}).get("public_key")
|
|
362
|
+
if not pk:
|
|
363
|
+
click.echo("[preflight] no public key returned; skipping upload")
|
|
364
|
+
return
|
|
365
|
+
|
|
366
|
+
pk_bytes = base64.b64decode(pk, validate=True)
|
|
367
|
+
sealed_box = SealedBox(PublicKey(pk_bytes))
|
|
368
|
+
ciphertext = sealed_box.encrypt(env_api_key.encode("utf-8"))
|
|
369
|
+
ct_b64 = base64.b64encode(ciphertext).decode()
|
|
370
|
+
payload = {"name": "ENVIRONMENT_API_KEY", "ciphertext_b64": ct_b64}
|
|
371
|
+
|
|
372
|
+
click.echo(f"[preflight] posting to {backend_base.rstrip('/')}/v1/env-keys")
|
|
373
|
+
response = client.post(f"{backend_base.rstrip('/')}/v1/env-keys", json=payload)
|
|
374
|
+
if 200 <= response.status_code < 300:
|
|
375
|
+
click.echo(
|
|
376
|
+
f"✅ ENVIRONMENT_API_KEY uploaded successfully ({_preview_secret(env_api_key)})"
|
|
377
|
+
)
|
|
378
|
+
try:
|
|
379
|
+
ver = client.get(f"{backend_base.rstrip('/')}/v1/env-keys/verify")
|
|
380
|
+
if ver.status_code == 200 and (ver.json() or {}).get("present"):
|
|
381
|
+
click.echo("✅ Key verified in backend")
|
|
382
|
+
else:
|
|
383
|
+
click.echo(
|
|
384
|
+
f"⚠️ Verification returned {ver.status_code}, but upload succeeded - proceeding"
|
|
385
|
+
)
|
|
386
|
+
except Exception as verify_err: # pragma: no cover - verification optional
|
|
387
|
+
click.echo(
|
|
388
|
+
f"⚠️ Verification check failed ({verify_err}), but upload succeeded - proceeding"
|
|
389
|
+
)
|
|
390
|
+
return
|
|
391
|
+
|
|
392
|
+
snippet = response.text[:400] if response.text else ""
|
|
393
|
+
message = (
|
|
394
|
+
f"ENVIRONMENT_API_KEY upload failed with status {response.status_code}"
|
|
395
|
+
+ (f" body={snippet}" if snippet else "")
|
|
396
|
+
)
|
|
397
|
+
if crash_on_failure:
|
|
398
|
+
raise click.ClickException(f"[CRITICAL] {message}")
|
|
399
|
+
click.echo(f"[WARN] {message}; proceeding anyway")
|
|
400
|
+
except Exception as exc: # pragma: no cover - network failures
|
|
401
|
+
message = f"Backend preflight for ENVIRONMENT_API_KEY failed: {exc}"
|
|
402
|
+
if crash_on_failure:
|
|
403
|
+
raise click.ClickException(f"[CRITICAL] {message}") from exc
|
|
404
|
+
click.echo(f"[WARN] {message}; proceeding anyway")
|
|
405
|
+
|
|
406
|
+
|
|
407
|
+
def load_env_values(paths: Sequence[Path], *, allow_empty: bool = False) -> dict[str, str]:
|
|
408
|
+
"""Load values from a sequence of env files, returning a merged dictionary."""
|
|
409
|
+
|
|
410
|
+
values: dict[str, str] = {}
|
|
411
|
+
for path in paths:
|
|
412
|
+
try:
|
|
413
|
+
content = Path(path).read_text(encoding="utf-8")
|
|
414
|
+
except FileNotFoundError:
|
|
415
|
+
continue
|
|
416
|
+
for line in content.splitlines():
|
|
417
|
+
if not line or line.lstrip().startswith("#") or "=" not in line:
|
|
418
|
+
continue
|
|
419
|
+
key, value = line.split("=", 1)
|
|
420
|
+
key = key.strip()
|
|
421
|
+
value = value.strip()
|
|
422
|
+
if key and key not in values:
|
|
423
|
+
values[key] = value
|
|
424
|
+
if not allow_empty and not values:
|
|
425
|
+
raise click.ClickException("No environment values found")
|
|
426
|
+
os.environ.update({k: v for k, v in values.items() if k and v})
|
|
427
|
+
return values
|
|
428
|
+
|
|
429
|
+
|
|
430
|
+
def _parse_env_file(path: Path) -> dict[str, str]:
|
|
431
|
+
data: dict[str, str] = {}
|
|
432
|
+
try:
|
|
433
|
+
for line in path.read_text(encoding="utf-8").splitlines():
|
|
434
|
+
if not line or line.lstrip().startswith("#") or "=" not in line:
|
|
435
|
+
continue
|
|
436
|
+
key, value = line.split("=", 1)
|
|
437
|
+
data[key.strip()] = value.strip()
|
|
438
|
+
except FileNotFoundError:
|
|
439
|
+
pass
|
|
440
|
+
return data
|
|
441
|
+
|
|
442
|
+
|
|
443
|
+
def interactive_fill_env(env_path: Path) -> Path | None:
|
|
444
|
+
"""Interactively collect credentials and write them to a .env file."""
|
|
445
|
+
|
|
446
|
+
existing = _parse_env_file(env_path) if env_path.exists() else {}
|
|
447
|
+
|
|
448
|
+
def _prompt(label: str, *, default: str = "", required: bool) -> str | None:
|
|
449
|
+
while True:
|
|
450
|
+
try:
|
|
451
|
+
value = click.prompt(
|
|
452
|
+
label,
|
|
453
|
+
default=default,
|
|
454
|
+
show_default=bool(default) or not required,
|
|
455
|
+
).strip()
|
|
456
|
+
except (Abort, EOFError, KeyboardInterrupt):
|
|
457
|
+
click.echo("Aborted env creation.")
|
|
458
|
+
return None
|
|
459
|
+
if value or not required:
|
|
460
|
+
return value
|
|
461
|
+
click.echo("This field is required.")
|
|
462
|
+
|
|
463
|
+
env_default = existing.get("ENVIRONMENT_API_KEY", "").strip()
|
|
464
|
+
env_api_key = _prompt("ENVIRONMENT_API_KEY", default=env_default, required=True)
|
|
465
|
+
if env_api_key is None:
|
|
466
|
+
return None
|
|
467
|
+
|
|
468
|
+
synth_default = existing.get("SYNTH_API_KEY", "").strip()
|
|
469
|
+
openai_default = existing.get("OPENAI_API_KEY", "").strip()
|
|
470
|
+
synth_key = _prompt("SYNTH_API_KEY (optional)", default=synth_default, required=False) or ""
|
|
471
|
+
openai_key = _prompt("OPENAI_API_KEY (optional)", default=openai_default, required=False) or ""
|
|
472
|
+
|
|
473
|
+
env_path.parent.mkdir(parents=True, exist_ok=True)
|
|
474
|
+
env_path.write_text(
|
|
475
|
+
"\n".join(
|
|
476
|
+
[
|
|
477
|
+
f"ENVIRONMENT_API_KEY={env_api_key}",
|
|
478
|
+
f"SYNTH_API_KEY={synth_key}",
|
|
479
|
+
f"OPENAI_API_KEY={openai_key}",
|
|
480
|
+
]
|
|
481
|
+
)
|
|
482
|
+
+ "\n",
|
|
483
|
+
encoding="utf-8",
|
|
484
|
+
)
|
|
485
|
+
click.echo(f"Wrote credentials to {env_path}")
|
|
486
|
+
return env_path
|
|
487
|
+
|
|
488
|
+
|
|
489
|
+
def interactive_create_env(target_dir: Path) -> Path | None:
|
|
490
|
+
"""Create a .env file for the provided directory if one does not exist."""
|
|
491
|
+
|
|
492
|
+
env_path = (target_dir / ".env").resolve()
|
|
493
|
+
if env_path.exists():
|
|
494
|
+
existing = _parse_env_file(env_path)
|
|
495
|
+
env_api = (existing.get("ENVIRONMENT_API_KEY") or "").strip()
|
|
496
|
+
if env_api:
|
|
497
|
+
return env_path
|
|
498
|
+
click.echo(f"Existing {env_path} is missing ENVIRONMENT_API_KEY. Let's update it.")
|
|
499
|
+
return interactive_fill_env(env_path)
|
|
500
|
+
|
|
501
|
+
click.echo("No .env found for this task app. Let's create one.")
|
|
502
|
+
return interactive_fill_env(env_path)
|
|
503
|
+
|
|
504
|
+
|
|
505
|
+
def ensure_env_values(env_paths: list[Path], fallback_dir: Path) -> None:
|
|
506
|
+
"""Ensure required env values are present, prompting to create .env if needed."""
|
|
507
|
+
|
|
508
|
+
if (os.environ.get("ENVIRONMENT_API_KEY") or "").strip():
|
|
509
|
+
return
|
|
510
|
+
|
|
511
|
+
target = env_paths[0] if env_paths else (fallback_dir / ".env").resolve()
|
|
512
|
+
result = interactive_fill_env(target)
|
|
513
|
+
if result is None:
|
|
514
|
+
raise click.ClickException("ENVIRONMENT_API_KEY required to continue")
|
|
515
|
+
|
|
516
|
+
load_env_values([result])
|
|
517
|
+
if not (os.environ.get("ENVIRONMENT_API_KEY") or "").strip():
|
|
518
|
+
raise click.ClickException("Failed to load ENVIRONMENT_API_KEY from generated .env")
|
synth_ai/cli/mcp.py
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
|
|
2
|
+
|
|
3
|
+
import typing
|
|
4
|
+
from pathlib import Path
|
|
5
|
+
|
|
6
|
+
import click
|
|
7
|
+
from synth_ai.utils import ClaudeConfig
|
|
8
|
+
|
|
9
|
+
Agent = typing.Literal[
|
|
10
|
+
"claude",
|
|
11
|
+
]
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
@click.command()
|
|
15
|
+
@click.argument(
|
|
16
|
+
"agent",
|
|
17
|
+
type=click.Choice(
|
|
18
|
+
list(typing.get_args(Agent)),
|
|
19
|
+
case_sensitive=False
|
|
20
|
+
),
|
|
21
|
+
default = "claude"
|
|
22
|
+
)
|
|
23
|
+
@click.option(
|
|
24
|
+
"--config-path",
|
|
25
|
+
"config_path",
|
|
26
|
+
type=Path,
|
|
27
|
+
default=None
|
|
28
|
+
)
|
|
29
|
+
def mcp_cmd(agent: Agent, config_path: Path | None) -> None:
|
|
30
|
+
match agent:
|
|
31
|
+
case "claude":
|
|
32
|
+
target_path = config_path or ClaudeConfig.get_default_config_path()
|
|
33
|
+
config = ClaudeConfig(target_path)
|
|
34
|
+
config.update_mcp_config()
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
from .core import command, get_command
|
|
4
|
+
from .errors import ModalServeCliError
|
|
5
|
+
from .validation import validate_modal_serve_options
|
|
6
|
+
|
|
7
|
+
__all__ = [
|
|
8
|
+
"command",
|
|
9
|
+
"get_command",
|
|
10
|
+
"ModalServeCliError",
|
|
11
|
+
"validate_modal_serve_options",
|
|
12
|
+
]
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
import click
|
|
4
|
+
from synth_ai.cli.task_apps import task_app_group
|
|
5
|
+
|
|
6
|
+
__all__ = ["command", "get_command"]
|
|
7
|
+
|
|
8
|
+
command = task_app_group.commands.get("modal-serve")
|
|
9
|
+
|
|
10
|
+
|
|
11
|
+
def get_command() -> click.Command:
|
|
12
|
+
if command is None:
|
|
13
|
+
raise RuntimeError("modal-serve command is not registered on task_app_group")
|
|
14
|
+
return command
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
from collections.abc import MutableMapping
|
|
4
|
+
from typing import Any
|
|
5
|
+
|
|
6
|
+
__all__ = ["validate_modal_serve_options"]
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
def validate_modal_serve_options(options: MutableMapping[str, Any]) -> MutableMapping[str, Any]:
|
|
10
|
+
"""Validate parameters passed to the modal-serve CLI command."""
|
|
11
|
+
return options
|