using tracing.fireworks.ai works

Author: Dylan Huang

tests/remote_server/remote_server.py

@@ -1,17 +1,15 @@
 import os
 import threading
-from typing import Any, Dict, List
+from typing import Any, Dict
 
 import uvicorn
 from fastapi import FastAPI, HTTPException
-from langfuse.openai import openai  # pyright: ignore[reportPrivateImportUsage]
+from openai import OpenAI
 
 from eval_protocol.types.remote_rollout_processor import (
     InitRequest,
     StatusResponse,
-    create_langfuse_config_tags,
 )
-from eval_protocol.models import Message
 
 
 app = FastAPI()
@@ -28,21 +26,20 @@ def init(req: InitRequest):
     # Kick off worker thread that does a single-turn chat via Langfuse OpenAI integration
     def _worker():
         try:
-            metadata = {"langfuse_tags": create_langfuse_config_tags(req)}
-
             if not req.messages:
                 raise ValueError("messages is required")
 
             completion_kwargs = {
                 "model": req.model,
                 "messages": req.messages,
-                "metadata": metadata,
             }
 
             if req.tools:
                 completion_kwargs["tools"] = req.tools
 
-            completion = openai.chat.completions.create(**completion_kwargs)
+            client = OpenAI(base_url=req.model_base_url, api_key=os.environ.get("FIREWORKS_API_KEY"))
+
+            completion = client.chat.completions.create(**completion_kwargs)
 
         except Exception as e:
             # Best-effort; mark as done even on error to unblock polling

tests/remote_server/test_remote_langfuse.py

@@ -56,7 +56,7 @@ def rows() -> List[EvaluationRow]:
 
 
 @pytest.mark.skipif(os.environ.get("CI") == "true", reason="Only run this test locally (skipped in CI)")
-@pytest.mark.parametrize("completion_params", [{"model": "gpt-4o"}])
+@pytest.mark.parametrize("completion_params", [{"model": "fireworks_ai/accounts/fireworks/models/gpt-oss-120b"}])
 @evaluation_test(
     data_loaders=DynamicDataLoader(
         generators=[rows],
@@ -65,6 +65,7 @@ def rows() -> List[EvaluationRow]:
         remote_base_url="http://127.0.0.1:3000",
         timeout_seconds=30,
         output_data_loader=langfuse_output_data_loader,
+        model_base_url="https://tracing.fireworks.ai/project_id/cmg5fd57b0006y107kuxkcrhk",
     ),
 )
 async def test_remote_rollout_and_fetch_langfuse(row: EvaluationRow) -> EvaluationRow: