unparallelled commited on
Commit
4eb1cbb
·
verified ·
1 Parent(s): c306eb7

Upload logs/128k_master.log with huggingface_hub

Browse files
Files changed (1) hide show
  1. logs/128k_master.log +20 -0
logs/128k_master.log ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [05:38:41] FATAL: ./build/bluscriptCP_exec not built
2
+ [09:16:06] === 680M 131k extension, 8x H200, cp=8 dp=1 ===
3
+ [09:16:06] parent : t64 (98 @65,512 -- the 64k model, staged ladder)
4
+ [09:16:06] YaRN : SCALE=32 x ORIG=4096 = 131072 == CP_T INVARIANT HOLDS
5
+ [09:16:06] virtual range 0..4095 (full), 32 tokens/virtual unit
6
+ [09:16:06] LR : peak 1.8e-4 (6e-4 x 0.3), warmup 100 / stable 900 / DECAY 200 -> 6e-5
7
+ [09:16:06] data : Data_Loader/Data/mix128 b6=8.9 b5=39 b4=25 b3=15 b1=10, usable 1219 steps, running 1200
8
+ [09:16:06] PROBE cp=8 (16384 tok/rank, grad_accum 4)
9
+ [09:16:58] PROBE FAILED rc=1 steps=0
10
+ [09:17:28] === 680M 131k extension, 8x H200, cp=8 dp=1 ===
11
+ [09:17:28] parent : t64 (98 @65,512 -- the 64k model, staged ladder)
12
+ [09:17:28] YaRN : SCALE=32 x ORIG=4096 = 131072 == CP_T INVARIANT HOLDS
13
+ [09:17:28] virtual range 0..4095 (full), 32 tokens/virtual unit
14
+ [09:17:28] LR : peak 1.8e-4 (6e-4 x 0.3), warmup 100 / stable 900 / DECAY 200 -> 6e-5
15
+ [09:17:28] data : Data_Loader/Data/mix128 b6=8.9 b5=39 b4=25 b3=15 b1=10, usable 1219 steps, running 1200
16
+ [09:17:28] PROBE cp=8 (16384 tok/rank, grad_accum 4)
17
+ [09:19:08] PROBE OK cp=8: peak mem: 86705MB of 143771 MiB (H200), dt dt: 20144.88ms
18
+ [09:19:08] TRAINING 1200 steps
19
+ [11:17:55] done, run 3, loss: 1.024427
20
+ [11:17:55] === 131K TRAINING COMPLETE ===