ppiGPLM / ROC-Checkpoints /training_log.md
kouroshSA's picture
Add ppiGPLM ROC/LES trajectory checkpoints (ckpt_250..8000 + final)
5497956 verified
|
Raw
History Blame Contribute Delete
3.24 kB

Vanilla GPT Training Log

Started: 2026-01-01 17:39:37

Output Directory: out_ppiGPLM_Med4-solo_hope.v3-8ki_4k_768_vanilla

Mode: Vanilla (standard GPT)


Hyperparameters

Model Architecture

Parameter Value
n_layer 12
n_head 12
n_embd 768
block_size 4096
vocab_size 29
dropout 0.2
bias True

Training Configuration

Parameter Value
dataset shakespeare_char
batch_size 12
gradient_accumulation_steps 1
tokens_per_iter 98,304
learning_rate 0.0005
min_lr 1e-05
max_iters 8001
warmup_iters 200
lr_decay_iters 8000
weight_decay 0.1
beta1 0.9
beta2 0.99
grad_clip 1.0
eval_interval 250
eval_iters 100

System Configuration

Parameter Value
device cuda:0
dtype bfloat16
compile False
Total Parameters 85,078,272

Training Progress

Iteration Train Loss Val Loss Checkpoint Timestamp
250 2.7533 2.7560 ckpt_250.pt 2026-01-01 17:45:05
500 2.6923 2.6987 ckpt_500.pt 2026-01-01 17:49:45
750 2.5995 2.6146 ckpt_750.pt 2026-01-01 17:54:25
1000 2.3717 2.3937 ckpt_1000.pt 2026-01-01 17:59:05
1250 2.0253 2.0677 ckpt_1250.pt 2026-01-01 18:03:45
1500 1.2687 1.3372 ckpt_1500.pt 2026-01-01 18:08:25
1750 0.6444 0.7193 ckpt_1750.pt 2026-01-01 18:13:06
2000 0.3304 0.4024 ckpt_2000.pt 2026-01-01 18:17:47
2250 0.1911 0.2712 ckpt_2250.pt 2026-01-01 18:22:28
2500 0.1344 0.1991 ckpt_2500.pt 2026-01-01 18:27:08
2750 0.0992 0.1633 ckpt_2750.pt 2026-01-01 18:31:49
3000 0.0761 0.1379 ckpt_3000.pt 2026-01-01 18:36:29
3250 0.0658 0.1258 ckpt_3250.pt 2026-01-01 18:41:10
3500 0.0553 0.1021 ckpt_3500.pt 2026-01-01 18:45:50
3750 0.0519 0.0874 ckpt_3750.pt 2026-01-01 18:50:31
4000 0.0462 0.0907 ckpt_4000.pt 2026-01-01 18:55:11
4250 0.0417 0.0802 ckpt_4250.pt 2026-01-01 18:59:51
4500 0.0389 0.0802 ckpt_4500.pt 2026-01-01 19:04:32
4750 0.0368 0.0802 ckpt_4750.pt 2026-01-01 19:09:12
5000 0.0334 0.0859 ckpt_5000.pt 2026-01-01 19:13:52
5250 0.0318 0.0872 ckpt_5250.pt 2026-01-01 19:18:32
5500 0.0287 0.0824 ckpt_5500.pt 2026-01-01 19:23:13
5750 0.0271 0.0768 ckpt_5750.pt 2026-01-01 19:27:53
6000 0.0256 0.0927 ckpt_6000.pt 2026-01-01 19:32:33
6250 0.0249 0.0833 ckpt_6250.pt 2026-01-01 19:37:13
6500 0.0232 0.0919 ckpt_6500.pt 2026-01-01 19:41:53
6750 0.0224 0.0897 ckpt_6750.pt 2026-01-01 19:46:33
7000 0.0219 0.0891 ckpt_7000.pt 2026-01-01 19:51:17
7250 0.0217 0.0880 ckpt_7250.pt 2026-01-01 19:56:00
7500 0.0211 0.0914 ckpt_7500.pt 2026-01-01 20:00:44
7750 0.0209 0.0872 ckpt_7750.pt 2026-01-01 20:05:26
8000 0.0202 0.0821 ckpt_8000.pt 2026-01-01 20:10:10