Llama-3.3-70B-Instruct-FP8-KV

• Introduction

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

• Quantization Stragegy

  • Quantized Layers: All linear layers excluding "lm_head"
  • Weight: FP8 symmetric per-tensor
  • 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/Llama-3.3-70B-Instruct
python3 quantize_quark.py \
        --model_dir $MODEL_DIR \
        --output_dir $QUANT_MODEL_DIR \
        --quant_scheme w_fp8_a_fp8 \
        --kv_cache_dtype fp8 \
        --num_calib_data 128 \
        --model_export quark_safetensors \
        --no_weight_matrix_merge \
        --multi_gpu \
        --custom_mode fp8

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

Benchmark	Llama-3.3-70B-Instruct	Llama-3.3-70B-Instruct-FP8-KV(this model)
Perplexity-wikitext2	3.862387180328369	3.9621500968933105

License

Modifications copyright(c) 2024 Advanced Micro Devices,Inc. All rights reserved.