docs/HF_PUSH_GUIDE.md

HF PUSH GUIDE — ViuAI

Is repo ko do HF repos me push karenge. Bina approval ke push mat karna — token tum doge.

1. Repos banane hain

• Model repo: ViuAI/ViuMini-Dense-360M (type: model) — ye poora ViuMini-Dense-360M/ folder
• Dataset repo: ViuAI/viu-mini-raw-pretrain (type: dataset)

HF website pe New -> Model/Dataset -> naam daalo -> Create (private start kar sakte ho).

2. Local setup (ek baar)

pip install -r requirements.txt
# HF login — token paste karo (write access). Token https://huggingface.co/settings/tokens se lo
hf auth login
# ya: huggingface-cli login
hf auth whoami  # check

3. Model push (root se)

cd "C:\Users\IQC-SURBHI\Documents\S-Work\ViuAI\Mini ViuAI\ViuMini-Dense-360M"

# dry run (kya jayega dekho)
python model/scripts/push_to_hf.py --repo ViuAI/ViuMini-Dense-360M --dry-run

# real push
python model/scripts/push_to_hf.py --repo ViuAI/ViuMini-Dense-360M --message "update: 32L Dense 366.6M ready"

4. Dataset push

# data ko HF dataset format me push
python data/scripts/push_dataset_to_hf.py --repo ViuAI/viu-mini-pretrain-40-30-30 --dry-run
python data/scripts/push_dataset_to_hf.py --repo ViuAI/viu-mini-pretrain-40-30-30

# verify HF pe jaake dataset Viewer me lines dikhe

5. Train.py HF dataset se

python model/scripts/train.py --hf_dataset ViuAI/viu-mini-pretrain-40-30-30 --config model/configs/train_config.yaml

Notes

• Large files (*.pt, *.safetensors, tokenizer/outputs/*.json) LFS se jayenge — .gitattributes ready hai.
• Token kabhi repo me commit mat karna. .gitignore me .hf_token block hai.
• Har push ke baad docs/PROGRESS.md me entry karo.