amd / DeepSeek-V3.2-mxfp4

huggingface.co
Total runs: 976
24-hour runs: 3
7-day runs: -219
30-day runs: -1.1K
Model's Last Updated: February 11 2026

Introduction of DeepSeek-V3.2-mxfp4

Model Details of DeepSeek-V3.2-mxfp4

Model Overview

  • Model Architecture: DeepSeek-V3.2
    • Input: Text
    • Output: Text
  • Supported Hardware Microarchitecture: AMD MI350/MI355
  • ROCm : 7.0
  • PyTorch : 2.8.0
  • Transformers : 4.53.0
  • Operating System(s): Linux
  • Inference Engine: SGLang / vLLM
  • Model Optimizer: AMD-Quark (V0.10)
    • Weight quantization: OCP MXFP4, Static
    • Activation quantization: OCP MXFP4, Dynamic
  • Calibration Dataset: Pile

This model was built with deepseek-ai DeepSeek-V3.2 model by applying AMD-Quark for MXFP4 quantization.

Model Quantization

The model was quantized from deepseek-ai/DeepSeek-V3.2 using AMD-Quark . Both weights and activations were quantized to MXFP4 format.

Deployment

This model can be deployed efficiently using the SGLang and vLLM backends.

Evaluation
export VLLM_USE_V1=1  
export SAFETENSORS_FAST_GPU=1
export VLLM_ROCM_USE_AITER=1
export VLLM_ROCM_USE_AITER_MOE=1
export VLLM_ROCM_USE_AITER_FP8BMM=0
export VLLM_ROCM_USE_AITER_FP4BMM=0
model_path="/shareddata/deepseek-ai/DeepSeek-V3.2-mxfp4"
vllm serve $model_path \
  --tensor-parallel-size 4 \
  --data-parallel-size 1 \
  --max-num-batched-tokens 32768 \
  --trust-remote-code \
  --no-enable-prefix-caching \
  --disable-log-requests \
  --kv-cache-dtype bfloat16 \
  --gpu_memory_utilization 0.85 \
  --compilation-config '{"cudagraph_mode": "FULL_AND_PIECEWISE"}' \
  --block-size 1

lm_eval \
  --model local-completions \
  --tasks gsm8k \
  --model_args model=/shareddata/deepseek-ai/DeepSeek-V3.2-mxfp4,base_url=http://127.0.0.1:8000/v1/completions \
  --batch_size auto \
 
Accuracy
Benchmark DeepSeek-V3.2 DeepSeek-V3.2-mxfp4(this model) Recovery
gsm8k (flexible-extract) 95.68 95.38 99.68%
Reproduction

docker: rocm/vllm-private:nightly-dpskv3.2-mxfp4

vllm main: 0900cedb3f89e475bea256c4cf5a13b5f02635bc

License

Modifications Copyright(c) 2025 Advanced Micro Devices, Inc. All rights reserved.

Runs of amd DeepSeek-V3.2-mxfp4 on huggingface.co

976
Total runs
3
24-hour runs
-142
3-day runs
-219
7-day runs
-1.1K
30-day runs

More Information About DeepSeek-V3.2-mxfp4 huggingface.co Model

More DeepSeek-V3.2-mxfp4 license Visit here:

https://choosealicense.com/licenses/mit

DeepSeek-V3.2-mxfp4 huggingface.co

DeepSeek-V3.2-mxfp4 huggingface.co is an AI model on huggingface.co that provides DeepSeek-V3.2-mxfp4's model effect (), which can be used instantly with this amd DeepSeek-V3.2-mxfp4 model. huggingface.co supports a free trial of the DeepSeek-V3.2-mxfp4 model, and also provides paid use of the DeepSeek-V3.2-mxfp4. Support call DeepSeek-V3.2-mxfp4 model through api, including Node.js, Python, http.

DeepSeek-V3.2-mxfp4 huggingface.co Url

https://huggingface.co/amd/DeepSeek-V3.2-mxfp4

amd DeepSeek-V3.2-mxfp4 online free

DeepSeek-V3.2-mxfp4 huggingface.co is an online trial and call api platform, which integrates DeepSeek-V3.2-mxfp4's modeling effects, including api services, and provides a free online trial of DeepSeek-V3.2-mxfp4, you can try DeepSeek-V3.2-mxfp4 online for free by clicking the link below.

amd DeepSeek-V3.2-mxfp4 online free url in huggingface.co:

https://huggingface.co/amd/DeepSeek-V3.2-mxfp4

DeepSeek-V3.2-mxfp4 install

DeepSeek-V3.2-mxfp4 is an open source model from GitHub that offers a free installation service, and any user can find DeepSeek-V3.2-mxfp4 on GitHub to install. At the same time, huggingface.co provides the effect of DeepSeek-V3.2-mxfp4 install, users can directly use DeepSeek-V3.2-mxfp4 installed effect in huggingface.co for debugging and trial. It also supports api for free installation.

DeepSeek-V3.2-mxfp4 install url in huggingface.co:

https://huggingface.co/amd/DeepSeek-V3.2-mxfp4

Url of DeepSeek-V3.2-mxfp4

DeepSeek-V3.2-mxfp4 huggingface.co Url

Provider of DeepSeek-V3.2-mxfp4 huggingface.co

amd
ORGANIZATIONS

Other API from amd

huggingface.co

Total runs: 275.6K
Run Growth: 180.1K
Growth Rate: 65.32%
Updated:April 14 2026
huggingface.co

Total runs: 92.5K
Run Growth: -30.5K
Growth Rate: -32.96%
Updated:July 17 2026
huggingface.co

Total runs: 62.0K
Run Growth: 2.1K
Growth Rate: 3.35%
Updated:July 01 2026
huggingface.co

Total runs: 47.1K
Run Growth: -74.1K
Growth Rate: -157.17%
Updated:June 19 2026
huggingface.co

Total runs: 8.7K
Run Growth: -3.4K
Growth Rate: -39.56%
Updated:August 12 2025
huggingface.co

Total runs: 8.4K
Run Growth: -3.3K
Growth Rate: -39.14%
Updated:October 09 2024
huggingface.co

Total runs: 7.0K
Run Growth: -17.1K
Growth Rate: -244.37%
Updated:July 27 2026
huggingface.co

Total runs: 2.4K
Run Growth: 544
Growth Rate: 22.66%
Updated:June 19 2026
huggingface.co

Total runs: 1.2K
Run Growth: -116
Growth Rate: -9.29%
Updated:May 27 2026
huggingface.co

Total runs: 1.2K
Run Growth: -6.5K
Growth Rate: -536.76%
Updated:June 19 2026
huggingface.co

Total runs: 1.1K
Run Growth: 736
Growth Rate: 64.96%
Updated:January 22 2026
huggingface.co

Total runs: 874
Run Growth: -560
Growth Rate: -64.07%
Updated:June 19 2026
huggingface.co

Total runs: 849
Run Growth: 841
Growth Rate: 99.06%
Updated:June 26 2025
huggingface.co

Total runs: 770
Run Growth: 529
Growth Rate: 68.70%
Updated:November 03 2025
huggingface.co

Total runs: 725
Run Growth: 476
Growth Rate: 65.66%
Updated:November 15 2025
huggingface.co

Total runs: 514
Run Growth: 432
Growth Rate: 84.05%
Updated:November 18 2025