Upload folder using huggingface_hub

inference.py

@@ -99,12 +99,12 @@ def fetch_reset(task_name: str) -> dict:
     return response.json()
 
 
-def submit_action(action: PharmaAction) -> dict:
-    response = requests.post(
-        f"{ENV_URL}/step",
-        json=action.model_dump(),
-        timeout=30,
-    )
+def submit_action(action: PharmaAction) -> dict:
+    response = requests.post(
+        f"{ENV_URL}/step",
+        json={"action": action.model_dump()},
+        timeout=30,
+    )
     response.raise_for_status()
     return response.json()
 

server/app.py

@@ -19,9 +19,12 @@ class OpenEnvPharmaAdapter:
     expected by OpenEnv's HTTP server and web playground helpers.
     """
 
+    _shared_env = PharmaVigilanceEnv()
+    _shared_state = State(episode_id=None, step_count=0)
+
     def __init__(self) -> None:
-        self._env = PharmaVigilanceEnv()
-        self._last_state = State(episode_id=None, step_count=0)
+        self._env = self.__class__._shared_env
+        self._last_state = self.__class__._shared_state
 
     @staticmethod
     def _normalize_reports(reports):
@@ -36,6 +39,7 @@ class OpenEnvPharmaAdapter:
     def reset(self, task_id: str = "known_signal_easy") -> PharmaObservation:
         observation = self._env.reset(task_id=task_id)
         self._last_state = State(episode_id=task_id, step_count=0)
+        self.__class__._shared_state = self._last_state
         return PharmaObservation(
             task_id=observation.task_id,
             reports=self._normalize_reports(observation.reports),
@@ -57,6 +61,7 @@ class OpenEnvPharmaAdapter:
             episode_id=observation.task_id,
             step_count=observation.step_number,
         )
+        self.__class__._shared_state = self._last_state
         return PharmaObservation(
             task_id=observation.task_id,
             reports=self._normalize_reports(observation.reports),

tests/test_env.py

@@ -1,10 +1,11 @@
-import sys
-from pathlib import Path
-
-sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
-
-from env import Action, PharmaVigilanceEnv
-from tasks import (
+import sys
+from pathlib import Path
+
+import pytest
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from env import Action, PharmaVigilanceEnv
+from tasks import (
     cluster_signal_medium_action_grader,
     cluster_signal_medium_grader,
     confounded_hard_action_grader,
@@ -126,7 +127,35 @@ def test_get_task_returns_hard_truth():
     assert task.ground_truth.suspect_drug == "Tacrolimus+Voriconazole"
 
 
-def test_public_graders_are_strictly_bounded():
-    assert known_signal_easy_grader({"rewards": [1.0]}) == 0.99
-    assert cluster_signal_medium_grader({"rewards": [0.0]}) == 0.01
-    assert confounded_hard_grader({"score": 1.5}) == 0.99
+def test_public_graders_are_strictly_bounded():
+    assert known_signal_easy_grader({"rewards": [1.0]}) == 0.99
+    assert cluster_signal_medium_grader({"rewards": [0.0]}) == 0.01
+    assert confounded_hard_grader({"score": 1.5}) == 0.99
+
+
+def test_http_reset_then_step_roundtrip():
+    pytest.importorskip("openenv")
+    from fastapi.testclient import TestClient
+    from server.app import app
+
+    client = TestClient(app)
+
+    reset_response = client.post("/reset", json={})
+    assert reset_response.status_code == 200
+
+    step_response = client.post(
+        "/step",
+        json={
+            "action": {
+                "classification": "known_side_effect",
+                "suspect_drug": "Lisinopril",
+                "severity_assessment": "mild",
+                "recommended_action": "log_and_monitor",
+                "reasoning": "Known ACE inhibitor cough.",
+            }
+        },
+    )
+    assert step_response.status_code == 200
+    payload = step_response.json()
+    assert payload["done"] is True
+    assert payload["reward"] == 1.0