Datasets:

Amogh1221
/

nanorush_training

Modalities:

Text

Formats:

text

Size:

10K - 100K

Libraries:

Datasets

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 14k rows

text stringlengths 0 133

════════════════════════════════════════════════════════
TRAINING STARTED — 2026-08-13 11:50:07
════════════════════════════════════════════════════════
Model params : 283,171,584
vocab_size : 32768
n_embd : 768
n_head : 12
n_layer : 36
block_size : 4096
dropout : 0.0
bias : False
batch_size : 2
gradient_accumulation_steps : 40
max_iters : 35000
learning_rate : 0.0003
weight_decay : 0.1
beta1 : 0.95
beta2 : 0.95
warmup_iters : 3000
lr_decay_iters : 35000
min_lr : 3e-05
eval_interval : 50
eval_iters : 200
log_interval : 10
save_interval : 50
gen_interval : 5000
max_new_tokens_gen : 256
num_generations : 3
device : cuda
dtype : float16
compile : False
fused_adam : True
tf32 : False
dataset : train.bin
data_dir : data
temperature : 0.8
top_k : 50
top_p : 0.95
ema_decay : 0.999
use_ema : True
grad_clip : 1.0
gradient_checkpointing : 1
preload : False
════════════════════════════════════════════════════════
[2026-08-13 11:59:53] STEP 10/35000 \| loss=11.2862 \| lr=1.10e-06 \| grad_norm=24.502 \| tok/s=5,592 \| VRAM=5.1GB \| ETA=23d 17h 35m
[2026-08-13 12:08:44] STEP 20/35000 \| loss=9.5701 \| lr=2.10e-06 \| grad_norm=8.606 \| tok/s=6,174 \| VRAM=5.1GB \| ETA=22d 14h 34m
[2026-08-13 12:17:32] STEP 30/35000 \| loss=9.3498 \| lr=3.10e-06 \| grad_norm=3.552 \| tok/s=6,202 \| VRAM=5.1GB \| ETA=22d 4h 41m
[2026-08-13 12:26:20] STEP 40/35000 \| loss=9.1746 \| lr=4.10e-06 \| grad_norm=2.523 \| tok/s=6,207 \| VRAM=5.1GB \| ETA=21d 23h 33m
[2026-08-13 12:35:06] STEP 50/35000 \| loss=9.2155 \| lr=5.10e-06 \| grad_norm=1.879 \| tok/s=6,228 \| VRAM=5.1GB \| ETA=21d 20h 5m

════════════════════════════════════════════════════════
EVALUATION @ Step 50 / 35000 (0.1%)
════════════════════════════════════════════════════════
Train Loss : 9.0887
Val Loss : 9.0851 (best: inf NEW BEST)
Perplexity : 8822.82
EMA Val Loss : 10.3798
Learning Rate : 5.10e-06
Avg Grad Norm : 0.000
Tokens/sec : 5,444
VRAM : 5.1 / 15.6 GB
Elapsed : 00:50:09
ETA : 24d 8h 18m
════════════════════════════════════════════════════════
[2026-08-13 12:49:45] STEP 60/35000 \| loss=9.1002 \| lr=6.10e-06 \| grad_norm=1.691 \| tok/s=3,729 \| VRAM=5.1GB \| ETA=24d 2h 44m
[2026-08-13 12:58:34] STEP 70/35000 \| loss=8.9643 \| lr=7.10e-06 \| grad_norm=1.662 \| tok/s=6,190 \| VRAM=5.1GB \| ETA=23d 17h 18m
[2026-08-13 13:07:24] STEP 80/35000 \| loss=8.9374 \| lr=8.10e-06 \| grad_norm=1.786 \| tok/s=6,181 \| VRAM=5.1GB \| ETA=23d 10h 16m
[2026-08-13 13:16:15] STEP 90/35000 \| loss=8.8230 \| lr=9.10e-06 \| grad_norm=2.196 \| tok/s=6,176 \| VRAM=5.1GB \| ETA=23d 4h 49m
[2026-08-13 13:25:06] STEP 100/35000 \| loss=8.6778 \| lr=1.01e-05 \| grad_norm=1.976 \| tok/s=6,173 \| VRAM=5.1GB \| ETA=23d 0h 27m

════════════════════════════════════════════════════════
EVALUATION @ Step 100 / 35000 (0.3%)
════════════════════════════════════════════════════════
Train Loss : 8.6493
Val Loss : 8.6517 (best: 9.0851 Δ: -0.4334)
Perplexity : 5719.84
EMA Val Loss : 9.9234
Learning Rate : 1.01e-05
Avg Grad Norm : 0.000
Tokens/sec : 5,453
VRAM : 5.1 / 15.6 GB
Elapsed : 01:40:09
ETA : 24d 6h 34m
════════════════════════════════════════════════════════
[2026-08-13 13:40:13] STEP 110/35000 \| loss=8.5727 \| lr=1.11e-05 \| grad_norm=1.883 \| tok/s=3,611 \| VRAM=5.1GB \| ETA=24d 6h 2m
[2026-08-13 13:49:05] STEP 120/35000 \| loss=8.5178 \| lr=1.21e-05 \| grad_norm=1.964 \| tok/s=6,164 \| VRAM=5.1GB \| ETA=24d 0h 18m
[2026-08-13 13:57:57] STEP 130/35000 \| loss=8.3891 \| lr=1.31e-05 \| grad_norm=2.150 \| tok/s=6,155 \| VRAM=5.1GB \| ETA=23d 19h 29m
[2026-08-13 14:06:50] STEP 140/35000 \| loss=8.2792 \| lr=1.41e-05 \| grad_norm=2.666 \| tok/s=6,150 \| VRAM=5.1GB \| ETA=23d 15h 22m
[2026-08-13 14:15:43] STEP 150/35000 \| loss=8.1301 \| lr=1.51e-05 \| grad_norm=2.014 \| tok/s=6,145 \| VRAM=5.1GB \| ETA=23d 11h 48m

════════════════════════════════════════════════════════
EVALUATION @ Step 150 / 35000 (0.4%)
════════════════════════════════════════════════════════
Train Loss : 8.0841
Val Loss : 8.0718 (best: 8.6517 Δ: -0.5799)
Perplexity : 3202.93
EMA Val Loss : 9.5549
Learning Rate : 1.51e-05
Avg Grad Norm : 0.000

End of preview. Expand in Data Studio

No dataset card yet

Downloads last month: 4,584