Upload folder using huggingface_hub

openenv_CrisisWorldCortex.egg-info/PKG-INFO

@@ -5,6 +5,8 @@ Summary: Crisisworldcortex environment for OpenEnv
 Requires-Python: >=3.10
 Requires-Dist: openenv-core[core]==0.2.3
 Requires-Dist: openai<3.0,>=2.0
+Requires-Dist: datasets>=4.0.0
+Requires-Dist: huggingface-hub>=1.0.0
 Provides-Extra: dev
 Requires-Dist: pre-commit>=4.0.0; extra == "dev"
 Requires-Dist: pytest>=8.0.0; extra == "dev"

openenv_CrisisWorldCortex.egg-info/SOURCES.txt

@@ -7,6 +7,7 @@ pyproject.toml
 ./__init__.py
 ./client.py
 ./inference.py
+./mm.py
 ./models.py
 openenv_CrisisWorldCortex.egg-info/PKG-INFO
 openenv_CrisisWorldCortex.egg-info/SOURCES.txt
@@ -19,9 +20,21 @@ server/__init__.py
 server/app.py
 tests/test_actions_round_trip.py
 tests/test_baseline_b1.py
+tests/test_baseline_b2.py
+tests/test_baseline_b3.py
+tests/test_cortex_brain_executive.py
+tests/test_cortex_brain_smoke.py
+tests/test_cortex_council.py
+tests/test_cortex_council_smoke.py
+tests/test_cortex_lenses.py
+tests/test_cortex_metacognition.py
+tests/test_cortex_perception.py
+tests/test_cortex_routing_policy.py
+tests/test_cortex_subagents.py
 tests/test_env_reset_kwargs.py
 tests/test_env_step_reward_wiring.py
 tests/test_import_graph.py
+tests/test_inference_agent_dispatch.py
 tests/test_legal_constraint_enforcement.py
 tests/test_llm_client.py
 tests/test_observation_no_latent_leak.py
@@ -29,9 +42,18 @@ tests/test_outer_reward_in_range.py
 tests/test_outer_reward_non_constancy.py
 tests/test_outer_reward_terminal_bonus.py
 tests/test_package_exports.py
+tests/test_reward_signal_quality.py
 tests/test_schemas_roundtrip.py
 tests/test_simulator_determinism.py
 tests/test_simulator_random_episode.py
 tests/test_simulator_task_configs.py
 tests/test_smoke_env.py
-tests/test_stdout_format.py
+tests/test_stdout_format.py
+tests/test_synthetic_rejection_payload.py
+tests/test_training_eval_metrics.py
+tests/test_training_multi_model_skeleton.py
+tests/test_training_reward_shaping.py
+tests/test_training_rollout_buffer.py
+tests/test_training_scripts_b1.py
+tests/test_training_scripts_collect_sft.py
+tests/test_training_scripts_sft.py

openenv_CrisisWorldCortex.egg-info/requires.txt

@@ -1,5 +1,7 @@
 openenv-core[core]==0.2.3
 openai<3.0,>=2.0
+datasets>=4.0.0
+huggingface-hub>=1.0.0
 
 [dev]
 pre-commit>=4.0.0

pyproject.toml

@@ -25,6 +25,9 @@ dependencies = [
     # future uv sync from silently pulling 3.x. Bump explicitly when 3.0
     # ships and we've verified compatibility.
     "openai>=2.0,<3.0",
+    # Training / data-collection scripts push trajectory datasets to HF.
+    "datasets>=4.0.0",
+    "huggingface-hub>=1.0.0",
 ]
 
 [project.optional-dependencies]

training/scripts/collect_sft_data.py

@@ -273,59 +273,68 @@ def collect() -> int:
 
     for task in tasks:
         for ep in range(NUM_EPISODES):
+            env = CrisisworldcortexEnv(base_url=ENV_URL).sync()
             try:
-                env = CrisisworldcortexEnv(base_url=ENV_URL)
-                obs = env.reset(task_name=task, seed=ep, max_ticks=EPISODE_TICKS)
+                reset_result = env.reset(task_name=task, seed=ep, max_ticks=EPISODE_TICKS)
+                obs = (
+                    reset_result.observation
+                    if hasattr(reset_result, "observation")
+                    else reset_result
+                )
             except Exception as exc:
                 log(f"WARN env.reset failed task={task} ep={ep}: {exc}")
+                env.close()
                 continue
-            last_reward = 0.0
-            for tick in range(EPISODE_TICKS):
-                user_prompt = serialize_observation(obs, last_reward)
-                try:
-                    completion = call_teacher(client, _SYSTEM_PROMPT_BODY, user_prompt)
-                except Exception as exc:
-                    log(f"WARN teacher call failed task={task} ep={ep} tick={tick}: {exc}")
-                    break
-                action_dict = parse_action_json(completion)
-                if action_dict is None:
-                    parse_fail_count += 1
-                    break
-                # Submit to env (Pydantic validates here).
-                try:
-                    result = env.step(
-                        CrisisworldcortexAction.model_validate({"action": action_dict})
-                    )
-                except Exception as exc:
-                    log(f"WARN env.step rejected task={task} ep={ep} tick={tick}: {exc}")
-                    parse_fail_count += 1
-                    break
-                next_obs = result.observation if hasattr(result, "observation") else result
-                reward = next_obs.reward if next_obs.reward is not None else 0.0
-                accepted = bool(
-                    next_obs.recent_action_log and next_obs.recent_action_log[-1].accepted
-                )
-                if not accepted:
-                    rejected_count += 1
-                elif reward < MIN_REWARD_THRESHOLD:
-                    low_reward_count += 1
-                else:
-                    rows.append(
-                        {
-                            "prompt": user_prompt,
-                            "completion": json.dumps(action_dict, separators=(",", ":")),
-                            "task": task,
-                            "seed": ep,
-                            "tick": tick,
-                            "reward": float(reward),
-                            "accepted": True,
-                        }
+            try:
+                last_reward = 0.0
+                for tick in range(EPISODE_TICKS):
+                    user_prompt = serialize_observation(obs, last_reward)
+                    try:
+                        completion = call_teacher(client, _SYSTEM_PROMPT_BODY, user_prompt)
+                    except Exception as exc:
+                        log(f"WARN teacher call failed task={task} ep={ep} tick={tick}: {exc}")
+                        break
+                    action_dict = parse_action_json(completion)
+                    if action_dict is None:
+                        parse_fail_count += 1
+                        break
+                    # Submit to env (Pydantic validates here).
+                    try:
+                        result = env.step(
+                            CrisisworldcortexAction.model_validate({"action": action_dict})
+                        )
+                    except Exception as exc:
+                        log(f"WARN env.step rejected task={task} ep={ep} tick={tick}: {exc}")
+                        parse_fail_count += 1
+                        break
+                    next_obs = result.observation if hasattr(result, "observation") else result
+                    reward = next_obs.reward if next_obs.reward is not None else 0.0
+                    accepted = bool(
+                        next_obs.recent_action_log and next_obs.recent_action_log[-1].accepted
                     )
-                    kept_count += 1
-                last_reward = float(reward)
-                obs = next_obs
-                if next_obs.done:
-                    break
+                    if not accepted:
+                        rejected_count += 1
+                    elif reward < MIN_REWARD_THRESHOLD:
+                        low_reward_count += 1
+                    else:
+                        rows.append(
+                            {
+                                "prompt": user_prompt,
+                                "completion": json.dumps(action_dict, separators=(",", ":")),
+                                "task": task,
+                                "seed": ep,
+                                "tick": tick,
+                                "reward": float(reward),
+                                "accepted": True,
+                            }
+                        )
+                        kept_count += 1
+                    last_reward = float(reward)
+                    obs = next_obs
+                    if next_obs.done:
+                        break
+            finally:
+                env.close()
             if (ep + 1) % 5 == 0:
                 log(
                     f"[{task}] {ep + 1}/{NUM_EPISODES} kept={kept_count} "

training/scripts/train_b1_grpo.py

@@ -339,8 +339,8 @@ def main() -> int:
     tasks = tuple(t.strip() for t in TASKS_CSV.split(",") if t.strip())
     log(f"tasks={tasks}")
 
-    def make_env() -> CrisisworldcortexEnv:
-        return CrisisworldcortexEnv(base_url=ENV_URL)
+    def make_env() -> Any:
+        return CrisisworldcortexEnv(base_url=ENV_URL).sync()
 
     SYSTEM_PROMPT = build_system_prompt()
 
@@ -363,9 +363,17 @@ def main() -> int:
     meta: list[dict] = []
     for entry in seed_pool:
         env = make_env()
-        obs = env.reset(task_name=entry["task"], seed=entry["seed"], max_ticks=EPISODE_TICKS)
-        prompts.append(make_chat_prompt(obs))
-        meta.append(entry)
+        try:
+            result = env.reset(
+                task_name=entry["task"],
+                seed=entry["seed"],
+                max_ticks=EPISODE_TICKS,
+            )
+            obs = result.observation if hasattr(result, "observation") else result
+            prompts.append(make_chat_prompt(obs))
+            meta.append(entry)
+        finally:
+            env.close()
 
     train_dataset = Dataset.from_dict(
         {
@@ -387,16 +395,19 @@ def main() -> int:
         rewards: list[float] = []
         for completion, t, s in zip(completions, task, seed):
             env = make_env()
-            env.reset(task_name=t, seed=int(s), max_ticks=EPISODE_TICKS)
-            payload = parse_action(completion) or parse_failure_marker()
             try:
-                result = env.step(CrisisworldcortexAction(action=payload))
-                obs = result.observation if hasattr(result, "observation") else result
-                reward = obs.reward if obs.reward is not None else 0.0
-                rewards.append(float(reward))
-            except Exception as exc:
-                log(f"WARN env.step failed task={t} seed={s}: {exc}")
-                rewards.append(-1.0)
+                env.reset(task_name=t, seed=int(s), max_ticks=EPISODE_TICKS)
+                payload = parse_action(completion) or parse_failure_marker()
+                try:
+                    result = env.step(CrisisworldcortexAction(action=payload))
+                    obs = result.observation if hasattr(result, "observation") else result
+                    reward = obs.reward if obs.reward is not None else 0.0
+                    rewards.append(float(reward))
+                except Exception as exc:
+                    log(f"WARN env.step failed task={t} seed={s}: {exc}")
+                    rewards.append(-1.0)
+            finally:
+                env.close()
         return rewards
 
     # ---- GRPO training ----

training/scripts/train_cortex_multi_model.py

@@ -398,8 +398,8 @@ def main() -> int:
     tasks = tuple(t.strip() for t in TASKS_CSV.split(",") if t.strip())
     log(f"tasks={tasks}")
 
-    def make_env() -> CrisisworldcortexEnv:
-        return CrisisworldcortexEnv(base_url=ENV_URL)
+    def make_env() -> Any:
+        return CrisisworldcortexEnv(base_url=ENV_URL).sync()
 
     train_dataset = Dataset.from_dict(
         {
@@ -419,10 +419,16 @@ def main() -> int:
     ) -> list[float]:
         rewards: list[float] = []
         for _completion, t, s in zip(completions, task, seed):
+            env = None
             try:
                 council = Council(brains=brains, routing_policy=router_policy)
                 env = make_env()
-                obs = env.reset(task_name=t, seed=int(s), max_ticks=EPISODE_TICKS)
+                reset_result = env.reset(task_name=t, seed=int(s), max_ticks=EPISODE_TICKS)
+                obs = (
+                    reset_result.observation
+                    if hasattr(reset_result, "observation")
+                    else reset_result
+                )
                 cumulative = 0.0
                 last_reward = 0.0
                 for _ in range(EPISODE_TICKS):
@@ -438,6 +444,9 @@ def main() -> int:
             except Exception as exc:
                 log(f"WARN rollout failed task={t} seed={s}: {exc}")
                 rewards.append(-1.0)
+            finally:
+                if env is not None:
+                    env.close()
         return rewards
 
     # ---- GRPO config + trainer ----