Use gepa async evaluator support in GEPA example

harbor_cookbook/gepa/optimize.py

@@ -1,5 +1,5 @@
 # /// script
-# dependencies = ["harbor", "gepa"]
+# dependencies = ["harbor", "gepa @ git+https://github.com/gepa-ai/gepa.git@feat/async-evaluator-support"]
 # requires-python = ">=3.12"
 # ///
 """GEPA prompt optimization for MedAgentBench.
@@ -65,12 +65,12 @@
 _environment: str = DEFAULT_ENVIRONMENT
 
 
-def evaluate(candidate, example):
+async def evaluate(candidate, example):
     """Run one Harbor trial and return (score, side_info)."""
     task_id = example.id.name
     log.info("Evaluating %s ...", task_id)
 
-    result = run_trial(
+    result = await run_trial(
         candidate,
         example.downloaded_path,
         agent_name=_agent_name,
@@ -205,7 +205,6 @@ def main():
             engine=EngineConfig(
                 max_metric_calls=args.max_evals,
                 max_candidate_proposals=args.max_iterations,
-                max_workers=args.max_workers,
                 run_dir=args.output_dir,
             ),
             reflection=ReflectionConfig(

harbor_cookbook/gepa/utils.py

@@ -1,11 +1,9 @@
 """Utilities for downloading MedAgentBench tasks and running Harbor trials."""
 
-import asyncio
 import logging
 import os
 import shutil
 import tempfile
-import threading
 from collections import defaultdict
 from pathlib import Path
 
@@ -26,8 +24,6 @@
 DEFAULT_MODEL = "openai/gpt-5-nano"
 DEFAULT_ENVIRONMENT = EnvironmentType.DOCKER
 
-_loop = asyncio.new_event_loop()
-threading.Thread(target=_loop.run_forever, daemon=True).start()
 _queue: TrialQueue | None = None
 
 
@@ -76,7 +72,7 @@ def _read_trial_file(trials_dir: Path, relative: str, limit: int = 3000) -> str:
     return ""
 
 
-def run_trial(
+async def run_trial(
     prompt: str,
     task_path: Path,
     agent_name: str = DEFAULT_AGENT,
@@ -111,7 +107,7 @@ def run_trial(
         )
 
         log.debug("Starting trial for %s", task_dir.name)
-        result = asyncio.run_coroutine_threadsafe(_queue.submit(config), _loop).result()
+        result = await _queue.submit(config)
 
         rewards = result.verifier_result.rewards if result.verifier_result else {}
         exc = result.exception_info