switch to prompt

Author: xzrderek

.github/workflows/rollout.yml

@@ -11,14 +11,10 @@ on:
         description: 'Rollout ID for tracking'
         required: true
         type: string
-      messages_b64:
-        description: 'Base64 encoded JSON messages array'
+      prompt:
+        description: 'User prompt for the rollout'
         required: true
         type: string
-      tools_b64:
-        description: 'Base64 encoded JSON tools array (optional)'
-        required: false
-        type: string
 
 jobs:
   rollout:
@@ -46,8 +42,7 @@ jobs:
           python tests/github_actions/rollout_worker.py \
             --model "${{ inputs.model }}" \
             --rollout-id "${{ inputs.rollout_id }}" \
-            --messages-b64 "${{ inputs.messages_b64 }}" \
-            ${{ inputs.tools_b64 && format('--tools-b64 "{0}"', inputs.tools_b64) || '' }}
+            --prompt "${{ inputs.prompt }}"
 
       - name: Upload rollout trace
         uses: actions/upload-artifact@v4

eval_protocol/pytest/github_action_rollout_processor.py

@@ -1,5 +1,4 @@
 import asyncio
-import base64
 import json
 import os
 import tempfile
@@ -63,32 +62,27 @@ async def _process_row(row: EvaluationRow) -> EvaluationRow:
             if model is None:
                 raise ValueError("Model must be provided")
 
-            # Clean and encode messages
-            allowed_fields = {"role", "content", "tool_calls", "tool_call_id", "name"}
-            clean_messages = []
-            for m in row.messages:
-                if hasattr(m, "model_dump"):
-                    md = m.model_dump()
-                elif isinstance(m, dict):
-                    md = m
-                else:
-                    md = {
-                        "role": getattr(m, "role", None),
-                        "content": getattr(m, "content", None),
-                        "tool_calls": getattr(m, "tool_calls", None),
-                        "tool_call_id": getattr(m, "tool_call_id", None),
-                        "name": getattr(m, "name", None),
-                    }
-                clean_messages.append({k: v for k, v in md.items() if k in allowed_fields and v is not None})
+            # Extract user prompt (first user message)
+            user_prompt = None
+            for msg in row.messages:
+                if hasattr(msg, "role"):
+                    if msg.role == "user":
+                        user_prompt = msg.content
+                        break
+                elif isinstance(msg, dict):
+                    if msg.get("role") == "user":
+                        user_prompt = msg.get("content")
+                        break
+
+            if not user_prompt:
+                raise ValueError("At least one user message is required")
 
             # Prepare workflow inputs
             inputs = {
                 "model": model,
                 "rollout_id": row.execution_metadata.rollout_id,
-                "messages_b64": base64.b64encode(json.dumps(clean_messages).encode()).decode(),
+                "prompt": user_prompt,
             }
-            if row.tools:
-                inputs["tools_b64"] = base64.b64encode(json.dumps(row.tools).encode()).decode()
 
             # Dispatch workflow
             def _dispatch():

tests/github_actions/rollout_worker.py

@@ -7,7 +7,6 @@
 """
 
 import argparse
-import base64
 import json
 import os
 
@@ -20,44 +19,24 @@ def main():
     # Required arguments from workflow inputs
     parser.add_argument("--model", required=True, help="Model to use")
     parser.add_argument("--rollout-id", required=True, help="Rollout ID for tracking")
-    parser.add_argument("--messages-b64", required=True, help="Base64 encoded JSON messages")
-    parser.add_argument("--tools-b64", required=False, help="Base64 encoded JSON tools (optional)")
+    parser.add_argument("--prompt", required=True, help="User prompt for the rollout")
 
     args = parser.parse_args()
 
     print(f"🚀 Starting rollout {args.rollout_id}")
     print(f"   Model: {args.model}")
+    print(f"   Prompt: {args.prompt}")
 
-    # Decode messages and tools
-    try:
-        messages = json.loads(base64.b64decode(args.messages_b64).decode("utf-8"))
-        tools = None
-        if args.tools_b64:
-            tools = json.loads(base64.b64decode(args.tools_b64).decode("utf-8"))
-    except Exception as e:
-        print(f"❌ Failed to decode inputs: {e}")
-        # Save error trace
-        error_data = {
-            "status": "error",
-            "rollout_id": args.rollout_id,
-            "model": args.model,
-            "messages": [],
-            "error": f"Failed to decode inputs: {e}",
-        }
-        with open(f"rollout_trace_{args.rollout_id}.json", "w") as f:
-            json.dump(error_data, f, indent=2)
-        exit(1)
+    # Build messages array
+    messages = [{"role": "user", "content": args.prompt}]
 
     print(f"   Messages: {len(messages)} messages")
-    print(f"   Tools: {len(tools) if tools else 0} tools")
 
     # Perform the rollout
     conversation = messages.copy()
 
     try:
         completion_kwargs = {"model": args.model, "messages": messages}
-        if tools:
-            completion_kwargs["tools"] = tools
 
         client = OpenAI(api_key=os.environ.get("FIREWORKS_API_KEY"))
 
@@ -76,7 +55,6 @@ def main():
             "rollout_id": args.rollout_id,
             "model": args.model,
             "messages": conversation,
-            "tools": tools,
             "usage": completion.usage.model_dump() if completion.usage else None,
         }
 
@@ -91,7 +69,6 @@ def main():
             "rollout_id": args.rollout_id,
             "model": args.model,
             "messages": conversation,
-            "tools": tools,
             "error": str(e),
         }