Llama-3.1-8B-Instruct-w_fp8_per_channel_sym

EmbeddedLLM/Llama-3.1-8B-Instruct-w_fp8_per_channel_sym

• Introduction

  This model was created by applying Quark with calibration samples from Pile dataset.

• Quantization Stragegy

  • Quantized Layers: All linear layers excluding "lm_head"
  • Weight: FP8 symmetric per-channel
  • Activation: FP8 symmetric per-tensor
  • KV Cache: FP8 symmetric per-tensor

• Quick Start

1. Download and install Quark
2. Run the quantization script in the example folder using the following command line:

export MODEL_DIR = [local model checkpoint folder] or meta-llama/Meta-Llama-3.1-8B-Instruct 
# single GPU
HIP_VISIBLE_DEVICES=0 python quantize_quark.py --model_dir $MODEL_DIR \
                          --output_dir /app/model/quark/Llama-3.1-8B-Instruct-w_fp8_per_channel_sym/ \
                          --quant_scheme w_fp8_per_channel_sym \
                          --kv_cache_dtype fp8 \
                          --num_calib_data 128 \
                          --model_export quark_safetensors

# If model size is too large for single GPU, please use multi GPU instead.
python quantize_quark.py --model_dir $MODEL_DIR \
      --output_dir /app/model/quark/Llama-3.1-8B-Instruct-w_fp8_per_channel_sym/ \
      --quant_scheme w_fp8_per_channel_sym \
      --kv_cache_dtype fp8 \
      --num_calib_data 128 \
      --multi_gpu \
      --model_export quark_safetensors

Deployment

Quark has its own export format and allows FP8 quantized models to be efficiently deployed using the vLLM backend(vLLM-compatible).

Evaluation

Quark currently uses perplexity(PPL) as the evaluation metric for accuracy loss before and after quantization.The specific PPL algorithm can be referenced in the quantize_quark.py. The quantization evaluation results are conducted in pseudo-quantization mode, which may slightly differ from the actual quantized inference accuracy. These results are provided for reference only.

Evaluation scores

<table> <tr> <td><strong>Benchmark</strong> </td> <td><strong>Meta-Llama-3.1-8B-Instruct </strong> </td> <td><strong>EmbeddedLLM/Llama-3.1-8B-Instruct-w_fp8_per_channel_sym(this model)</strong> </td> </tr> <tr> <td>Perplexity-wikitext2 </td> <td>7.2169 </td> <td>7.34375 </td> </tr> </table>