hamori/checkpoints/finetuned.log.csv at master

Files

T

H1K0 d9585ec008 data: add fine-tuning run results (lr=3e-5, 30 epochs)

val loss 1.19 → 0.77, val perplexity 3.29 → 2.15.
Best epoch 20, early stop at epoch 30 (patience=10).
Improvement over previous lr=1e-5 run (best val ppl 2.22).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-05-21 20:52:39 +03:00

1.3 KiB

Raw Permalink Blame History

1	epoch	train_loss	val_loss	val_ppl	lr	elapsed_s
2	1	1.173289	1.190454	3.29	2.400000e-05	11.1
3	2	1.031668	1.017097	2.77	2.998248e-05	9.3
4	3	0.900459	0.914644	2.50	2.990471e-05	8.2
5	4	0.826531	0.877679	2.41	2.976507e-05	9.1
6	5	0.797045	0.851717	2.34	2.956413e-05	9.2
7	6	0.759114	0.836767	2.31	2.930272e-05	8.7
8	7	0.736987	0.819369	2.27	2.898194e-05	8.6
9	8	0.722803	0.806593	2.24	2.860312e-05	9.1
10	9	0.693306	0.797257	2.22	2.816782e-05	9.1
11	10	0.683278	0.792332	2.21	2.767785e-05	8.2
12	11	0.673061	0.787863	2.20	2.713525e-05	9.1
13	12	0.655914	0.782984	2.19	2.654228e-05	9.1
14	13	0.643172	0.777573	2.18	2.590139e-05	8.3
15	14	0.635985	0.774572	2.17	2.521524e-05	9.0
16	15	0.630730	0.773065	2.17	2.448668e-05	9.1
17	16	0.622494	0.771514	2.16	2.371874e-05	9.0
18	17	0.606942	0.769548	2.16	2.291460e-05	8.1
19	18	0.601119	0.768194	2.16	2.207761e-05	9.1
20	19	0.601939	0.768208	2.16	2.121123e-05	9.1
21	20	0.580447	0.766817	2.15	2.031907e-05	8.2
22	21	0.574881	0.767509	2.15	1.940483e-05	9.1
23	22	0.576981	0.769625	2.16	1.847230e-05	9.1
24	23	0.567170	0.770998	2.16	1.752536e-05	8.6
25	24	0.564600	0.771246	2.16	1.656793e-05	8.7
26	25	0.556949	0.772251	2.16	1.560399e-05	9.1
27	26	0.556080	0.770962	2.16	1.463754e-05	9.0
28	27	0.551530	0.769089	2.16	1.367260e-05	8.2
29	28	0.542789	0.768548	2.16	1.271317e-05	9.0
30	29	0.542809	0.770213	2.16	1.176324e-05	9.1
31	30	0.537889	0.771124	2.16	1.082674e-05	8.2

1.3 KiB Raw Permalink Blame History

1.3 KiB

Raw Permalink Blame History