amd/chatglm3-6b-awq-g128-int4-asym-fp16-onnx-hybrid

chatglm3-6b-awq-w-int4-asym-gs128-a-fp16-onnx-ryzen-strix-hybrid

• Introduction

  • Quantization Tool: Quark 0.6.0
  • OGA Model Builder: v0.5.1
  • Postprocess

• Quantization Strategy

  • AWQ / Group 128 / Asymmetric / UINT4 Weights / FP16 activations
  • Excluded Layers: None

  python3 quantize_quark.py \
        --model_dir "$model" \
        --output_dir "$output_dir" \
        --quant_scheme w_uint4_per_group_asym \
        --num_calib_data 128 \
        --quant_algo awq \
        --dataset pileval_for_awq_benchmark \
        --seq_len 512 \
        --model_export quark_safetensors \
        --data_type float16 \
        --exclude_layers [] \
        --custom_mode awq
  

• OGA Model Builder

  python builder.py \
    -i <quantized safetensor model dir> \
    -o <oga model output dir> \
    -p int4 \
    -e dml
  

• PostProcessed to generate Hybrid Model

• Quick Start

For quickstart, refer to hybrid-llm-artifacts_1.3.0.zip available in RyzenAI-SW-EA

Evaluation scores

The perplexity measurement is run on the wikitext-2-raw-v1 (raw data) dataset provided by Hugging Face. Perplexity score measured for prompt length 2k is 29.7801.

License

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

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.