data: add fine-tuning run results (lr=1e-5, 50 epochs)

val loss 1.24 → 0.80, val perplexity 3.47 → 2.22.
Best epoch 50 (no early stop); convergence epoch 30.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-05-21 20:17:25 +03:00
parent c98a12c4e9
commit 2e6e934564
3 changed files with 121 additions and 0 deletions
+67
View File
@@ -0,0 +1,67 @@
====================================================
FINE-TUNING REPORT
====================================================
Total epochs run : 50
Best epoch (val loss) : 50
Convergence epoch : 30 (val ≤ best+1 %)
Best val loss : 0.7987
Best val perplexity : 2.22
Final train loss : 0.6842
Unique parameters : 1,396,416
Checkpoint : checkpoints/finetuned.pt
Log CSV : checkpoints/finetuned.log.csv
====================================================
epoch train val ppl lr
----- -------- -------- ------- ----------
1 1.1851 1.2447 3.47 8.00e-06
2 1.1298 1.1684 3.22 9.99e-06
3 1.0748 1.1021 3.01 9.97e-06
4 1.0044 1.0452 2.84 9.92e-06
5 0.9631 0.9951 2.71 9.85e-06
6 0.9039 0.9533 2.59 9.77e-06
7 0.8676 0.9271 2.53 9.66e-06
8 0.8483 0.9112 2.49 9.53e-06
9 0.8178 0.8971 2.45 9.39e-06
10 0.8153 0.8859 2.43 9.23e-06
11 0.8041 0.8768 2.40 9.05e-06
12 0.7914 0.8687 2.38 8.85e-06
13 0.7781 0.8604 2.36 8.63e-06
14 0.7714 0.8537 2.35 8.41e-06
15 0.7687 0.8484 2.34 8.16e-06
16 0.7626 0.8429 2.32 7.91e-06
17 0.7494 0.8379 2.31 7.64e-06
18 0.7452 0.8339 2.30 7.36e-06
19 0.7460 0.8307 2.29 7.07e-06
20 0.7266 0.8273 2.29 6.77e-06
21 0.7221 0.8245 2.28 6.47e-06
22 0.7240 0.8222 2.28 6.16e-06
23 0.7214 0.8200 2.27 5.84e-06
24 0.7176 0.8171 2.26 5.52e-06
25 0.7116 0.8146 2.26 5.20e-06
26 0.7123 0.8126 2.25 4.88e-06
27 0.7051 0.8107 2.25 4.56e-06
28 0.6971 0.8088 2.25 4.24e-06
29 0.7002 0.8074 2.24 3.92e-06
30 0.6953 0.8063 2.24 3.61e-06
31 0.6879 0.8050 2.24 3.30e-06
32 0.6918 0.8039 2.23 3.00e-06
33 0.6907 0.8030 2.23 2.71e-06
34 0.6921 0.8018 2.23 2.43e-06
35 0.6917 0.8011 2.23 2.16e-06
36 0.6789 0.8006 2.23 1.90e-06
37 0.6821 0.8004 2.23 1.65e-06
38 0.6891 0.8002 2.23 1.42e-06
39 0.6865 0.7999 2.23 1.20e-06
40 0.6906 0.7996 2.22 1.00e-06
41 0.6841 0.7993 2.22 8.18e-07
42 0.6851 0.7991 2.22 6.50e-07
43 0.6892 0.7990 2.22 5.00e-07
44 0.6899 0.7989 2.22 3.69e-07
45 0.6792 0.7988 2.22 2.57e-07
46 0.6809 0.7987 2.22 1.65e-07
47 0.6885 0.7987 2.22 9.31e-08
48 0.6840 0.7987 2.22 4.15e-08
49 0.6854 0.7987 2.22 1.04e-08
50 0.6842 0.7987 2.22 0.00e+00 ←