Fix state endpoint and score range clamping

create_app_sig.txt

@@ -0,0 +1 @@
+(env: 'Callable[[], Environment]', action_cls: 'Type[Action]', observation_cls: 'Type[Observation]', env_name: 'Optional[str]' = None, max_concurrent_envs: 'Optional[int]' = None, concurrency_config: 'Optional[ConcurrencyConfig]' = None, gradio_builder: 'Optional[Callable[..., Any]]' = None) -> 'FastAPI'

final_inference.log

@@ -0,0 +1,18 @@
+[START] task=single-label-classify env=meta-content-moderation model=meta-llama/Llama-3.3-70B-Instruct
+[STEP] step=1 action={"content_id":"unknown","labels":["clean"],"action":"approve","confidence":0.0,"reasoning":"No content provided.","policy_citations":[]} reward=0.80 done=false error=null
+[STEP] step=2 action={"content_id":"post_004","labels":["hate_speech"],"action":"remove","confidence":0.95,"reasoning":"Direct call for violence against a group based on protected characteristics.","policy_citations":["Content that directly attacks people based on protected characteristics"]} reward=0.93 done=false error=null
+[STEP] step=3 action={"content_id":"post_011","labels":["clean"],"action":"approve","confidence":0.99,"reasoning":"The post expresses a positive sentiment and does not contain any harmful or violating content.","policy_citations":[]} reward=1.00 done=false error=null
+[STEP] step=4 action={"content_id":"post_027","labels":["misinformation"],"action":"remove","confidence":0.9,"reasoning":"The post spreads debunked conspiracy theories about vaccines, posing a risk to public health.","policy_citations":[]} reward=1.00 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=5 action={"content_id":"post_038","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.13 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=6 action={"content_id":"post_019","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=7 action={"content_id":"post_040","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=8 action={"content_id":"img_020","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=9 action={"content_id":"img_006","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.95 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=10 action={"content_id":"img_015","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=true error=null
+[END] success=true steps=10 score=0.748 rewards=0.80,0.93,1.00,1.00,0.13,0.04,0.04,0.04,0.95,0.04

output.log

@@ -0,0 +1,18 @@
+[START] task=single-label-classify env=meta-content-moderation model=meta-llama/Llama-3.3-70B-Instruct
+[STEP] step=1 action={"content_id":"unknown","labels":["clean"],"action":"approve","confidence":1.0,"reasoning":"No content provided to moderate.","policy_citations":[]} reward=1.00 done=false error=null
+[STEP] step=2 action={"content_id":"post_004","labels":["hate_speech","violence"],"action":"remove","confidence":0.95,"reasoning":"The post directly incites violence against a group, violating policies against hate speech and violence.","policy_citations":["Content that directly attacks people based on protected characteristics"]} reward=1.00 done=false error=null
+[STEP] step=3 action={"content_id":"post_011","labels":["clean"],"action":"approve","confidence":0.99,"reasoning":"The post expresses a positive sentiment and does not contain any harmful or violating content.","policy_citations":[]} reward=1.00 done=false error=null
+[STEP] step=4 action={"content_id":"post_027","labels":["misinformation"],"action":"remove","confidence":0.9,"reasoning":"The post spreads debunked conspiracy theories about vaccines, posing a risk to public health.","policy_citations":[]} reward=1.00 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=5 action={"content_id":"post_038","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.13 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=6 action={"content_id":"post_019","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=7 action={"content_id":"post_040","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=8 action={"content_id":"img_020","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=9 action={"content_id":"img_006","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.95 done=false error=null
+[DEBUG] Model call failed: Error code: 402 - {'error': 'You have depleted your monthly included credits. Purchase pre-paid credits to continue using Inference Providers. Alternatively, subscribe to PRO to get 20x more included usage.'}
+[STEP] step=10 action={"content_id":"img_015","labels":["clean"],"action":"approve","confidence":0.5,"reasoning":"Model fallback \u2014 defaulting to approve","policy_citations":[]} reward=0.04 done=true error=null
+[END] success=false steps=10 score=0.000 rewards=1.00,1.00,1.00,1.00,0.13,0.04,0.04,0.04,0.95,0.04

routes.txt

@@ -0,0 +1 @@
+['/openapi.json', '/docs', '/docs/oauth2-redirect', '/redoc', '/mcp', '/reset', '/step', '/state', '/metadata', '/health', '/schema', '/mcp', '/ws', '/state', '/', '/tasks', '/openenv.yaml']

server/app.py

@@ -29,6 +29,17 @@ app = create_app(
     max_concurrent_envs=1
 )
 
+# Remove the default /state route added by create_app so our custom one below is used
+app.router.routes[:] = [r for r in app.router.routes if getattr(r, "path", None) != "/state"]
+
+# Force the state endpoint to return our full ModerationState (including 'score')
+@app.get("/state")
+def get_state():
+    """Returns the full environment state including custom fields like score."""
+    # We use the singleton instance
+    env = MetaContentModerationEnv()
+    return env.state
+
 # Custom extra routes specific to this environment
 @app.get("/", include_in_schema=False)
 def root():

server/env.py

@@ -228,8 +228,16 @@ class MetaContentModerationEnv(Environment[ModerationDecision, ModerationObserva
 
     def _compute_score(self) -> float:
         if not self._decisions_log:
-            return 0.0
+            return 0.01
         max_possible = self._max_steps * 1.0
-        raw = self._cumulative_reward / max_possible if max_possible > 0 else 0.0
-        return round(min(max(raw, 0.0), 1.0), 4)
+        if max_possible <= 0:
+            return 0.01
+            
+        avg_reward = self._cumulative_reward / max_possible
+        # Map avg_reward from [-1.0, 1.0] to [0.0, 1.0]
+        normalized = (avg_reward + 1.0) / 2.0
+        
+        # Clamp strictly between 0.01 and 0.99 for OpenEnv
+        score = min(max(normalized, 0.01), 0.99)
+        return round(score, 4)