unparallelled commited on
Commit
b0ceb10
·
verified ·
1 Parent(s): 20b325c

Upload logs/2026-09-22_t128c_evalcollapse_master.log with huggingface_hub

Browse files
logs/2026-09-22_t128c_evalcollapse_master.log ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [08:51:09] === eval across the loss decline: 300/400/450/500 x {4096, 32744} @x32 ===
2
+ [08:51:09] === RESULTS niah_single, T=131072 x32 ORIG=4096, N=100 ===
3
+ step 4096 32744 loss(mean50)
4
+ 300 . .
5
+ 400 . .
6
+ 450 . .
7
+ 500 . .
8
+ [08:51:09] reference: run2 s300 = 100 / 89 ; t64 = 100 / 93 ; old t128 = 98 / 69
9
+ [08:51:09] EVAL DONE
10
+ [08:53:45] === eval across the loss decline: 300/400/450/500 x {4096,32744} @x32 ===
11
+ [08:56:58] s500 @32744 ->
12
+ [08:56:59] s400 @32744 ->
13
+ [08:56:59] s300 @32744 ->
14
+ [08:56:59] s500 @4096 ->
15
+ [08:56:59] s450 @32744 ->
16
+ [08:56:59] s300 @4096 ->
17
+ [08:57:00] s450 @4096 ->
18
+ [08:57:00] s400 @4096 ->
19
+ [08:57:00] === RESULTS niah_single, T=131072 x32 ORIG=4096, N=100 ===
20
+ step 4096 32744
21
+ 300 . .
22
+ 400 . .
23
+ 450 . .
24
+ 500 . .
25
+ [08:57:00] ref: run2 s300 100/89 s500 95/77 | t64 100/93 | old-t128 98/69
26
+ [08:57:00] EVAL DONE
27
+ [09:12:16] === eval across the loss decline: 300/400/450/500 x {4096,32744} @x32 ===
28
+ [09:18:46] s500 @4096 -> 94.0
29
+ [09:18:57] s400 @4096 -> 97.0
30
+ [09:18:58] s300 @4096 -> 99.0
31
+ [09:19:20] s450 @4096 -> 93.0
32
+ [09:25:55] s500 @32744 -> 76.0
33
+ [09:26:12] s300 @32744 -> 92.0
34
+ [09:26:40] s400 @32744 -> 75.0
35
+ [09:26:54] s450 @32744 -> 69.0
36
+ [09:26:54] === RESULTS niah_single, T=131072 x32 ORIG=4096, N=100 ===
37
+ step 4096 32744
38
+ 300 99.0 92.0
39
+ 400 97.0 75.0
40
+ 450 93.0 69.0
41
+ 500 94.0 76.0
42
+ [09:26:54] ref: run2 s300 100/89 s500 95/77 | t64 100/93 | old-t128 98/69
43
+ [09:26:54] EVAL DONE