Delete code/upload_watch.py with huggingface_hub

code/upload_watch.py

@@ -1,73 +0,0 @@
-"""Run on pod-1: watch checkpoint dirs and push the current-best head ckpt
-to HF so the dev pod can pull + eval it.
-Pushes:
-  - newest spec_v8/c*/best.pt  -> spec_v8/best.pt (curriculum champion)
-  - spec_v7b/best.pt           -> spec_v7b/best.pt (legacy run)
-Usage: nohup python3 upload_watch.py > /root/upload_watch.log 2>&1 &
-"""
-import os, time, subprocess, glob
-
-HF = "/root/spec100m/hf_io.py"
-LOG = "/root/upload_watch.log"
-sigs = {}
-
-
-def log(m):
-    line = f"[{time.strftime('%H:%M:%S')}] {m}"
-    print(line, flush=True)
-    open(LOG, "a").write(line + "\n")
-
-
-def newest(pattern):
-    fs = glob.glob(pattern)
-    return max(fs, key=os.path.getmtime) if fs else None
-
-
-def push_if_changed(src, repo_path):
-    if not src or not os.path.exists(src):
-        return
-    sig = str(os.path.getsize(src)) + str(int(os.path.getmtime(src)))
-    if sigs.get(repo_path) == sig:
-        return
-    time.sleep(10)   # let any write finish
-    if "base_" in src:
-        # base checkpoints carry the full model — upload whole (the base
-        # weights ARE the update; spec-stripping would drop them)
-        pass
-    else:
-        # upload spec_* params only (~250MB vs 924MB) — the base weights are
-        # already on HF as base_500m/final.pt; the watcher merges them.
-        import torch
-        try:
-            sd = torch.load(src, map_location="cpu")
-            spec_sd = {k: v for k, v in sd["model"].items()
-                       if k.startswith("spec_")}
-            tmp = f"/tmp/_spec_only_{os.path.basename(repo_path)}"
-            torch.save({"model": spec_sd, "cfg": sd.get("cfg"),
-                        "step": sd.get("step"), "spec_only": True}, tmp)
-            src = tmp
-        except Exception as e:
-            log(f"spec-strip failed {src}: {e}")
-            return
-    r = subprocess.run(
-        ["python3", HF, "push", src, repo_path],
-        capture_output=True, text=True, timeout=1200)
-    ok = "uploaded" in r.stdout
-    log(f"{repo_path}: {'uploaded' if ok else 'fail ' + (r.stdout+r.stderr)[-200:]}")
-    if ok:
-        sigs[repo_path] = sig
-
-
-while True:
-    try:
-        push_if_changed(newest("/root/checkpoints/spec_v8/c*/best.pt"),
-                        "spec_v8/best.pt")
-        push_if_changed(newest("/root/checkpoints/spec_seq*/best.pt"),
-                        "spec_seq/best.pt")
-        push_if_changed(newest("/root/checkpoints/base_1b/best.pt"),
-                        "base_1b/best.pt")
-        push_if_changed("/root/checkpoints/spec_v7b/best.pt",
-                        "spec_v7b/best.pt")
-    except Exception as e:
-        log(f"err: {e}")
-    time.sleep(120)