Upload logs/p1_master.log with huggingface_hub
Browse files- logs/p1_master.log +11 -0
logs/p1_master.log
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[14:00:25] === 680M phase 1: 4k -> 16k (2x H100, cp=2) ===
|
| 2 |
+
[14:00:25] parent: varybatch step 3400, 10.5 tok/param, scores 3.05 (1.06B scores 3.17)
|
| 3 |
+
[14:00:25] data: Data_Loader/Data/p1mix b3=50% b2=30% b1=20% usable 1220 steps, running 1200
|
| 4 |
+
[14:00:25] YaRN: SCALE=4 ORIG=4096 -> 16384 == CP_T=16384 OK
|
| 5 |
+
[14:00:25] WSD: warmup 100 stable 1098 tail 2
|
| 6 |
+
[14:00:25] PHASE 1a: 3-step probe (8192 tok/rank)
|
| 7 |
+
[14:00:55] PROBE OK: 3 steps, peak mem: 50117MB of 81559 MiB/card
|
| 8 |
+
[14:00:55] PHASE 1b: training 1200 steps
|
| 9 |
+
[16:39:09] training done, run 12, loss: 2.686266
|
| 10 |
+
[16:39:09] peak mem mem: 50117MB
|
| 11 |
+
[16:39:09] === PHASE 1 TRAINING COMPLETE ===
|