unparallelled commited on
Commit
fe26edf
·
verified ·
1 Parent(s): 08fbf4f

Upload logs/mix64k/finish.log with huggingface_hub

Browse files
Files changed (1) hide show
  1. logs/mix64k/finish.log +49 -0
logs/mix64k/finish.log ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [11:57:40] === corrected finisher armed ===
2
+ [11:57:40] waiting for the trainer to exit (it runs to step 1300, then frees the GPUs)
3
+ [12:02:42] trainer exited; last step: step 1299
4
+ [12:02:42] preserved checkpoints: 12
5
+ [12:02:42] === evaluating the FINAL checkpoint, step 1200 ===
6
+ [12:02:42] EVAL 1200 from keep_step_1200.ckpt -- 4 lengths, one GPU each
7
+ [12:15:42] 1200 len 8192 -> niah_single,83.0
8
+ [12:15:42] 1200 len 16360 -> niah_single,70.0
9
+ [12:15:42] 1200 len 32744 -> niah_single,50.0
10
+ parent {8192: 67.0, 16360: 65.0, 32744: 52.0, 65512: 29.0}
11
+ pure-b5 {8192: 45.0, 16360: 17.0, 32744: 2.0, 65512: 1.0}
12
+ mix@1200 {8192: 83.0, 16360: 70.0, 32744: 50.0}
13
+ wrote mix64k_results_1200.png (397 KB)
14
+ wrote mix64k_results_1200.html (531 KB)
15
+ [12:15:43] plotted mix64k_results_1200.{png,html}
16
+ VERIFIED plots/mix64k_results_1200.png local=407,387 remote=407387
17
+ VERIFIED plots/mix64k_results_1200.html local=544,619 remote=544619
18
+ [12:15:51] EARLY PLOT for 1200 is on HF (8k/16k/32k)
19
+ [12:31:34] FINAL 1200 len 8192 -> niah_single,83.0
20
+ [12:31:34] FINAL 1200 len 16360 -> niah_single,70.0
21
+ [12:31:34] FINAL 1200 len 32744 -> niah_single,50.0
22
+ [12:31:34] FINAL 1200 len 65512 -> niah_single,31.0
23
+ parent {8192: 67.0, 16360: 65.0, 32744: 52.0, 65512: 29.0}
24
+ pure-b5 {8192: 45.0, 16360: 17.0, 32744: 2.0, 65512: 1.0}
25
+ mix@1200 {8192: 83.0, 16360: 70.0, 32744: 50.0, 65512: 31.0}
26
+ wrote mix64k_results_1200.png (410 KB)
27
+ wrote mix64k_results_1200.html (548 KB)
28
+ [12:31:35] plotted mix64k_results_1200.{png,html}
29
+ VERIFIED plots/mix64k_results_1200.png local=420,109 remote=420109
30
+ VERIFIED plots/mix64k_results_1200.html local=561,590 remote=561590
31
+ [12:31:41] --- verdict for step 1200 against the parent (67/65/52/29) ---
32
+ length parent mix delta bar verdict
33
+ 8192 67.0 83.0 +16.0 11.7 ok
34
+ 16360 65.0 70.0 +5.0 12.8 ok
35
+ 32744 52.0 50.0 -2.0 13.6 ok
36
+ 65512 29.0 31.0 +2.0 12.5 ok
37
+ VERDICT GOOD no resolvable degradation at 16k/32k/64k
38
+ [12:31:41] verdict: GOOD
39
+ [12:31:41] === final is GOOD -- skipping the 1200 eval (saves ~38 min of GPU) ===
40
+ [12:31:41] cliff detector on the MIX run: CLEAN (onset=0 baseline=2.778 current=2.711)
41
+ [12:31:41] for contrast, 100% b5 from the SAME parent: CLIFF at step 1199, ended 45/17/2/1
42
+ [12:31:41] === uploading everything ===
43
+ [12:31:44] sent (dir) logs/mix64k_1200_len_16360/
44
+ [12:31:47] sent (dir) logs/mix64k_1200_len_32744/
45
+ [12:31:51] sent (dir) logs/mix64k_1200_len_65512/
46
+ [12:31:54] sent (dir) logs/mix64k_1200_len_8192/
47
+ VERIFIED logs/mix64k/master.log local=2,673 remote=2673
48
+ VERIFIED logs/mix64k/train.log local=317,801 remote=317801
49
+ VERIFIED logs/mix64k/probe.log local=4,361 remote=4361