JulianKrgd/julian-600m-40b-instruct

Upload README.md with huggingface_hub

Upload app.py with huggingface_hub

feat: switch to base model (53.5% HellaSwag) for demo

fix: use raw SentencePiece tokenizer instead of HF LlamaTokenizer

fix: clean SentencePiece artifacts in output

fix: remove pre-installed deps from requirements.txt

fix: ZeroGPU v2 compatible rewrite

fix: rewrite app for ZeroGPU v2 compatibility

fix: add sft100k benchmark score (41.6% HellaSwag)

fix: use sft100k model (100K steps on 2.47M instructions, best ChatML)

Merge: use full README

Initial Space: Julian-600M-40B-Instruct demo with ChatML streaming

initial commit