Upload logs/2026-09-19_adamw_chain.log with huggingface_hub
Browse files
logs/2026-09-19_adamw_chain.log
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[04:48:18] waiting for build...
|
| 2 |
+
[04:48:19] binary 28300512 flags=CP_FUSED_ROPE=1|CP_ROPE_DEBUG=|CP_ATTN_SPLIT=1
|
| 3 |
+
[04:48:19] waiting for downloads...
|
| 4 |
+
[04:48:19] downloads done
|
| 5 |
+
[04:48:19] building ablmix16 (must match p1mix token-for-token)
|
| 6 |
+
|
| 7 |
+
USABLE STEPS at window 16,384 x grad_accum 32: 1220
|
| 8 |
+
(naive total//global_batch would say 1220 -- use 1220, it accounts for the
|
| 9 |
+
per-shard tail the loader discards, and for the silent wrap)
|
| 10 |
+
wrote Data_Loader/Data/ablmix16/MANIFEST.txt
|
| 11 |
+
[04:48:19] mix VERIFIED token-identical to p1mix (b3=320011377)
|
| 12 |
+
[04:48:19] === LAUNCHING: 8e-5 constant, AdamW only ===
|
| 13 |
+
[04:48:19] [const] === phase 1: 4k -> 16k, arm=const opt=adamw on GPU 0 ===
|
| 14 |
+
[04:48:19] [const] LR : max 8e-5 min 8e-5 warmup 12 (1% of 1200) stable 1186
|
| 15 |
+
[04:48:19] [const] optim : AdamW only (ZeRO-1 v2)
|
| 16 |
+
[04:48:19] [const] parent : varybatch data: Data_Loader/Data/ablmix16 steps: 1200
|
| 17 |
+
[04:48:19] [const] PROBE 3 steps
|
| 18 |
+
[04:49:25] [const] PROBE OK: 3 steps, peak mem: 78665MB, step-0 lr=6.6667e-06 (expect ~8.0e-05 at warmup 1/12)
|
| 19 |
+
[04:49:25] [const] TRAINING 1200 steps
|