train: preflight -- hubckpt.ensure_repo writes LFS attributes before the first checkpoint push (found by P1)

train/preflight.py

@@ -115,10 +115,7 @@ ounce100m_credentials.install()
 from huggingface_hub import HfApi
 api = HfApi(); tok = os.environ["HF_TOKEN"]
 repo = "Cion-lab/ounce100m-ckptbench-DELETEME"
-try:
-    api.create_repo(repo_id=repo, repo_type="dataset", exist_ok=True, token=tok)
-except Exception as e:
-    print("repo create:", type(e).__name__, str(e)[:120])
+print("ensure_repo:", json.dumps(hubckpt.ensure_repo(repo, api, tok)), flush=True)
 d = "/kaggle/working/ckptbench/checkpoint-1"
 os.makedirs(d, exist_ok=True)
 # 106,194,240 params x 4 B x 4 arrays (fp16-saved weights + fp32 master + Adam m and v) is the real
@@ -146,15 +143,18 @@ try:
     # to mistake it for a checkpoint. Everything measurable is in the log by this point.
     api.delete_repo(repo_id=repo, repo_type="dataset", token=tok)
 except Exception as e:
-    res, ver, got, same = {"error": f"{type(e).__name__}: {str(e)[:300]}"}, {}, {}, False
+    res, ver, got, same = {"error": f"{type(e).__name__}: {str(e)[:600]}"}, {}, {}, False
+    print("   P1 raised:", res["error"], flush=True)
 print("HUB_JSON_BEGIN")
-print(json.dumps({"push_verify_seconds": round(time.time()-up_t0,1),
+print(json.dumps({"error": res.get("error"),
+                  "push_verify_seconds": round(time.time()-up_t0,1),
                   "cycle": {k: res.get(k) for k in ("verify","pruned","free_before_gb","free_after_gb")},
                   "downloaded": got, "readback_ok": bool(ver.get("ok")) and same,
                   "gen_seconds": round(time.time()-t0,1)}, default=str))
 print("HUB_JSON_END")
 '''
-    r = sh([sys.executable, "-c", src], timeout=5400, label="P1 hub cycle (425 MB mock checkpoint)")
+    r = sh([sys.executable, "-c", src], timeout=5400,
+           label="P1 hub cycle (1.7 GB mock checkpoint: weights+master+Adam m,v)")
     R["P1"] = last_json(r["out"], "HUB_JSON_BEGIN", "HUB_JSON_END")
     R["P1_rc"] = r["rc"]
     j = R["P1"] or {}