Upload folder using huggingface_hub

README.md

@@ -14,6 +14,15 @@ base_path: /web
 
 CityFlow-based traffic-control project with intersection-level multi-agent DQN training and district-aware policy variants.
 
+## OpenEnv UI
+
+For the deployed OpenEnv web interface:
+
+- Click `Reset` before using `Step`.
+- Leave `Use Llm` unchecked for the fast, stable DQN-only path.
+- Use `District Actions` = `{}` for a valid no-op step payload.
+- Only enable `Use Llm` when you explicitly want district-level LLM guidance on top of the DQN executor.
+
 ## Training
 
 The default local-policy trainer now uses parameter-shared dueling Double DQN with prioritized replay and n-step returns:

models.py

@@ -4,10 +4,17 @@ import json
 from typing import Any
 
 from openenv.core.env_server import Action, Observation, State
-from pydantic import Field, field_validator
+from pydantic import ConfigDict, Field, field_validator
 
 
 class AgenticTrafficAction(Action):
+    model_config = ConfigDict(
+        extra="forbid",
+        validate_assignment=True,
+        arbitrary_types_allowed=True,
+        validate_default=True,
+    )
+
     use_llm: bool = Field(
         default=False,
         description=(
@@ -16,11 +23,16 @@ class AgenticTrafficAction(Action):
         ),
     )
     district_actions: dict[str, Any] = Field(
-        default_factory=dict,
+        default="{}",
         description=(
             "JSON object keyed by district_id. Use {} for a no-op step, or provide "
             'entries like {"d_00":{"strategy":"hold","phase_bias":"NS","duration_steps":10}}.'
         ),
+        json_schema_extra={
+            "type": "string",
+            "maxLength": 4000,
+            "default": "{}",
+        },
     )
     llm_max_new_tokens: int = Field(
         default=128,