SWAN adaptive mixed-precision quantization (5.51 avg bits, MAD bounds) abf7841 verified
Trevor Kennedy commited on
How to use baa-ai/Llama-3.1-70B-Instruct-SWAN-5bit-MLX with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Llama-3.1-70B-Instruct-SWAN-5bit-MLX baa-ai/Llama-3.1-70B-Instruct-SWAN-5bit-MLX