32B: use full max-seq-length 2048 (whole samples) on 2x RTX PRO 6000

32B: use full max-seq-length 2048 (whole samples) on 2x RTX PRO 6000

Correct docs: 8B models used default seqlen 2048, not 512

Add model quantization scripts and command reference

Add shell scripts

Add results

Add eval_logs

Upload test_vllm_load.py with huggingface_hub

Upload run_evals.sh with huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

initial commit