danielharkin21 commited on
Commit
1fbea4e
·
verified ·
1 Parent(s): f44bb98

Upload training_status.txt with huggingface_hub

Browse files
Files changed (1) hide show
  1. training_status.txt +11 -11
training_status.txt CHANGED
@@ -1,21 +1,21 @@
1
  =======================================================
2
  EDDIE TRAINING STATUS — 2x T4 DDP (SPEED EXP)
3
  =======================================================
4
- Updated : 2026-09-16 04:26:50 UTC
5
- Epoch : 3205
6
- Train CE : 0.816944
7
  Best CE : 0.809309
8
  Cortex base : nan
9
- U-loss (Huber): 0.109222
10
- Total loss : 0.926166
11
  -------------------------------------------------------
12
  u target : excess (signed excess of p; ~0 = trust p)
13
- excess mean : -0.0001 (p calibrated on avg if ~0)
14
- u mean : -0.0416 (should track excess mean)
15
- u corr : 0.0095 (Pearson u vs realized excess)
16
- u R^2 : -0.0307 (>0 = beats constant)
17
- u spread : +0.0026 (excess top-q u minus bottom-q u; want >0)
18
- const Huber : 0.1124 (u useful if U-loss < this: YES)
19
  -------------------------------------------------------
20
  Speed : 341 samples/s | 1.6 min/epoch
21
  Batch/GPU : 32 | GPUs: 2
 
1
  =======================================================
2
  EDDIE TRAINING STATUS — 2x T4 DDP (SPEED EXP)
3
  =======================================================
4
+ Updated : 2026-09-16 04:34:41 UTC
5
+ Epoch : 3210
6
+ Train CE : 0.819417
7
  Best CE : 0.809309
8
  Cortex base : nan
9
+ U-loss (Huber): 0.108780
10
+ Total loss : 0.928197
11
  -------------------------------------------------------
12
  u target : excess (signed excess of p; ~0 = trust p)
13
+ excess mean : -0.0006 (p calibrated on avg if ~0)
14
+ u mean : -0.0417 (should track excess mean)
15
+ u corr : 0.0048 (Pearson u vs realized excess)
16
+ u R^2 : -0.0324 (>0 = beats constant)
17
+ u spread : +0.0001 (excess top-q u minus bottom-q u; want >0)
18
+ const Huber : 0.1121 (u useful if U-loss < this: YES)
19
  -------------------------------------------------------
20
  Speed : 341 samples/s | 1.6 min/epoch
21
  Batch/GPU : 32 | GPUs: 2