Run the quantization script in the example folder using the following command line:
export MODEL_DIR = [local model checkpoint folder] or mistralai/Mixtral-8x22B-Instruct-v0.1
# single GPU
python3 quantize_quark.py \
--model_dir $MODEL_DIR \
--output_dir Mixtral-8x22B-Instruct-v0.1-FP8-KV \
--quant_scheme w_fp8_a_fp8 \
--kv_cache_dtype fp8 \
--num_calib_data 128 \
--model_export quark_safetensors \
--no_weight_matrix_merge
# If model size is too large for single GPU, please use multi GPU instead.
python3 quantize_quark.py \
--model_dir $MODEL_DIR \
--output_dir Mixtral-8x22B-Instruct-v0.1-FP8-KV \
--quant_scheme w_fp8_a_fp8 \
--kv_cache_dtype fp8 \
--num_calib_data 128 \
--model_export quark_safetensors \
--no_weight_matrix_merge \
--multi_gpu
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
Mixtral-8x22B-Instruct-v0.1
Mixtral-8x22B-Instruct-v0.1-FP8-KV(this model)
Perplexity-wikitext2
2.8871
2.9193
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.
Runs of amd Mixtral-8x22B-Instruct-v0.1-FP8-KV on huggingface.co
1.9K
Total runs
-76
24-hour runs
-71
3-day runs
-142
7-day runs
-989
30-day runs
More Information About Mixtral-8x22B-Instruct-v0.1-FP8-KV huggingface.co Model
More Mixtral-8x22B-Instruct-v0.1-FP8-KV license Visit here:
Mixtral-8x22B-Instruct-v0.1-FP8-KV huggingface.co is an AI model on huggingface.co that provides Mixtral-8x22B-Instruct-v0.1-FP8-KV's model effect (), which can be used instantly with this amd Mixtral-8x22B-Instruct-v0.1-FP8-KV model. huggingface.co supports a free trial of the Mixtral-8x22B-Instruct-v0.1-FP8-KV model, and also provides paid use of the Mixtral-8x22B-Instruct-v0.1-FP8-KV. Support call Mixtral-8x22B-Instruct-v0.1-FP8-KV model through api, including Node.js, Python, http.
Mixtral-8x22B-Instruct-v0.1-FP8-KV huggingface.co is an online trial and call api platform, which integrates Mixtral-8x22B-Instruct-v0.1-FP8-KV's modeling effects, including api services, and provides a free online trial of Mixtral-8x22B-Instruct-v0.1-FP8-KV, you can try Mixtral-8x22B-Instruct-v0.1-FP8-KV online for free by clicking the link below.
amd Mixtral-8x22B-Instruct-v0.1-FP8-KV online free url in huggingface.co:
Mixtral-8x22B-Instruct-v0.1-FP8-KV is an open source model from GitHub that offers a free installation service, and any user can find Mixtral-8x22B-Instruct-v0.1-FP8-KV on GitHub to install. At the same time, huggingface.co provides the effect of Mixtral-8x22B-Instruct-v0.1-FP8-KV install, users can directly use Mixtral-8x22B-Instruct-v0.1-FP8-KV installed effect in huggingface.co for debugging and trial. It also supports api for free installation.
Mixtral-8x22B-Instruct-v0.1-FP8-KV install url in huggingface.co: