amd

/

DeepSeek-R1-Distill-Qwen-1.5B-awq-asym-uint4-g128-lmhead-onnx-hybrid

like 1

Follow

AMD 1.43k

python3 quantize_quark.py \
    --model_dir deepseek-ai/DeepSeek-R1-Distill-Qwen-14B \
    --output_dir DeepSeek-R1-Distill-Qwen-14B-AWQ \
    --quant_scheme w_uint4_per_group_asym \
    --quant_algo awq \
    --num_calib_data 128  \
    --dataset pileval_for_awq_benchmark \
    --seq_len 512 \
    --data_type bfloat16 \
    --exclude_layers \