Stage 1 step 0: train/.venv (py3.11, mlx-lm 0.32), MLX 4-bit Qwen3.8-27B, serve script, baseline runner, README

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aUaQeLnwbb1zTpN7kHeat
2026-10-03 18:23:39 +02:00
parent cf8d6d5d07
commit 7351849a76
3 changed files with 125 additions and 0 deletions
--- a/train/baseline.py
+++ b/train/baseline.py
@@ -0,0 +1,60 @@
+"""Stage 1 eval runs on the fixed subset (train/subset.json) with a local OpenAI-compatible server.
+
+  train/.venv not needed: python3 train/baseline.py --label baseline [--only T01] [--run-base 20000]
+
+One task at a time. Results: runs/stage1/<label>.json (written after each task).
+"""
+import argparse
+import json
+import os
+import sys
+
+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, ROOT)
+from harness.adt_client import load_env  # noqa: E402
+from harness.agents import LlmAgent  # noqa: E402
+from harness.runner import Runner  # noqa: E402
+
+MODEL = os.path.expanduser("~/models/Qwen3.8-27B-4bit")
+BASE_URL = "http://127.0.0.1:8080/v1"
+
+
+def main():
+    load_env(os.path.join(ROOT, ".env"))
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--label", default="baseline")
+    ap.add_argument("--only", nargs="*")
+    ap.add_argument("--run-base", type=int, default=20000)
+    a = ap.parse_args()
+    subset = json.load(open(os.path.join(ROOT, "train", "subset.json")))["tasks"]
+    out_path = os.path.join(ROOT, "runs", "stage1", f"{a.label}.json")
+    os.makedirs(os.path.dirname(out_path), exist_ok=True)
+    res = json.load(open(out_path)) if os.path.exists(out_path) else {"model": MODEL, "tasks": {}}
+    runs_root = os.path.join(ROOT, "runs", "stage1", a.label)
+    for i, t in enumerate(subset):
+        tid = t["id"]
+        if (a.only and tid not in a.only) or tid in res["tasks"]:
+            continue
+        pool = os.path.join(ROOT, "tasks") if tid.startswith("T") else os.path.join(ROOT, "tasks_gen", "eval")
+        agent = LlmAgent(MODEL, BASE_URL)
+        try:
+            rep, run_dir = Runner(pool, runs_root).run(tid, agent, a.run_base + i)
+        except Exception as e:  # noqa: BLE001  one broken run must not stop the series
+            res["tasks"][tid] = {"error": str(e)[:300]}
+            json.dump(res, open(out_path, "w"), indent=1)
+            print(json.dumps({"task": tid, "error": str(e)[:300]}), flush=True)
+            continue
+        h = rep.get("hidden_tests") or {}
+        res["tasks"][tid] = {"category": t["category"], "object_type": t["object_type"],
+                             "score": (rep.get("score") or {}).get("total"), "parts": rep.get("score"),
+                             "gates": rep.get("gates"), "hidden": f"{h.get('passed')}/{h.get('total')}",
+                             "tool_calls": rep.get("tool_calls"), "seconds": rep.get("seconds"),
+                             "agent_seconds": rep.get("agent_seconds"), "final": (rep.get("final_report") or "")[:300],
+                             "run_dir": os.path.relpath(run_dir, ROOT)}
+        json.dump(res, open(out_path, "w"), indent=1)
+        print(json.dumps({"task": tid, "score": res["tasks"][tid]["score"], "hidden": res["tasks"][tid]["hidden"],
+                          "tool_calls": rep.get("tool_calls"), "seconds": rep.get("seconds")}), flush=True)
+
+
+if __name__ == "__main__":
+    main()