Upload folder using huggingface_hub
Browse files- .gitattributes +1 -0
- smollm_v3/ckpt.pt +3 -0
- smollm_v3/hparams.txt +39 -0
- smollm_v3/loss.png +3 -0
- smollm_v3/loss.txt +0 -0
.gitattributes
CHANGED
|
@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
shoggoth_telescope_sticker_2.png filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
shoggoth_telescope_sticker_2.png filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
smollm_v3/loss.png filter=lfs diff=lfs merge=lfs -text
|
smollm_v3/ckpt.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:05a7b75d693ac0f27ce9b0961ebab6cc6976221a9043d29c87a008726c42befc
|
| 3 |
+
size 7706567525
|
smollm_v3/hparams.txt
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
AstroPT-3204.9M
|
| 2 |
+
time: 1757360733
|
| 3 |
+
log_via_wandb: True
|
| 4 |
+
log_emissions: False
|
| 5 |
+
out_dir: logs/smollm
|
| 6 |
+
eval_interval: 100
|
| 7 |
+
log_interval: 50
|
| 8 |
+
checkpoint_interval: 1000
|
| 9 |
+
eval_iters: 100
|
| 10 |
+
eval_only: False
|
| 11 |
+
always_save_checkpoint: False
|
| 12 |
+
init_from: scratch
|
| 13 |
+
use_hf: True
|
| 14 |
+
stream_hf_dataset: True
|
| 15 |
+
leak_check: True
|
| 16 |
+
gradient_accumulation_steps: 5
|
| 17 |
+
batch_size: 32
|
| 18 |
+
spiral: True
|
| 19 |
+
block_size: 1024
|
| 20 |
+
image_size: 256
|
| 21 |
+
num_workers: 64
|
| 22 |
+
n_chan: 3
|
| 23 |
+
llm_model_name: HuggingFaceTB/SmolLM3-3B
|
| 24 |
+
lora_r: 256
|
| 25 |
+
tokeniser: affine
|
| 26 |
+
learning_rate: 0.0006
|
| 27 |
+
max_iters: 12000
|
| 28 |
+
weight_decay: 0.1
|
| 29 |
+
beta1: 0.9
|
| 30 |
+
beta2: 0.95
|
| 31 |
+
grad_clip: 1.0
|
| 32 |
+
decay_lr: True
|
| 33 |
+
warmup_iters: 1000
|
| 34 |
+
lr_decay_iters: 11000.0
|
| 35 |
+
min_lr: 5.9999999999999995e-05
|
| 36 |
+
backend: nccl
|
| 37 |
+
device: cuda
|
| 38 |
+
dtype: bfloat16
|
| 39 |
+
compile: True
|
smollm_v3/loss.png
ADDED
|
Git LFS Details
|
smollm_v3/loss.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|