README.md

---
license: mit
base_model:
- deepseek-ai/DeepSeek-R1-0528
---


# Model Overview

- **Model Architecture:** DeepSeek-R1-0528
  - **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](https://docs.sglang.ai/)/[vLLM](https://docs.vllm.ai/en/latest/)
- **Model Optimizer:** [AMD-Quark](https://quark.docs.amd.com/latest/index.html) (V0.10)
  - **Weight quantization:** OCP MXFP4, Static
  - **Activation quantization:** OCP MXFP4, Dynamic
- **Calibration Dataset:** [Pile](https://huggingface.co/datasets/mit-han-lab/pile-val-backup)

This model was built with deepseek-ai DeepSeek-R1-0528 model by applying [AMD-Quark](https://quark.docs.amd.com/latest/index.html) for MXFP4 quantization.

# Model Quantization

The model was quantized from [deepseek-ai/DeepSeek-R1-0528](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528) using [AMD-Quark](https://quark.docs.amd.com/latest/index.html). Both weights and activations were quantized to MXFP4 format. 

**Preprocessing requirement:**

Before executing the quantization script below, the original FP8 model must first be dequantized to BFloat16.
You can either perform the dequantization manually using this [conversion script](https://github.com/deepseek-ai/DeepSeek-V3/blob/main/inference/fp8_cast_bf16.py), or use the pre-converted BFloat16 model available at [amd/DeepSeek-R1-0528-BF16](https://huggingface.co/amd/DeepSeek-R1-0528-BF16).

**Quantization scripts:**
```
cd Quark/examples/torch/language_modeling/llm_ptq/
exclude_layers="*lm_head model.layers.61.*"
python3 quantize_quark.py --model_dir $MODEL_DIR \
                          --quant_scheme w_mxfp4_a_mxfp4 \
                          --group_size 32 \
                          --num_calib_data 128 \
                          --exclude_layers $exclude_layers \
                          --skip_evaluation \
                          --multi_gpu \
                          --model_export hf_format \
                          --output_dir amd/DeepSeek-R1-0528-MXFP4-V2
```

# Deployment

This model can be deployed efficiently using the [SGLang](https://docs.sglang.ai/) and [vLLM](https://docs.vllm.ai/en/latest/) backends.

## Evaluation

The model was evaluated on AIME24, and GSM8K benchmarks using the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) framework. 

### Accuracy

<table>
  <tr>
   <td><strong>Benchmark</strong>
   </td>
   <td><strong>DeepSeek-R1-0528-MXFP4-V2 (non MTP) </strong>
   </td>
   <td><strong>DeepSeek-R1-0528-MXFP4-V2 (MTP=3)</strong>
   </td>
  </tr>
  <tr>
   <td>AIME24 
   </td>
   <td>80.00
   </td>
   <td>83.33
   </td>
  </tr>
  <tr>
   <td>GSM8K 
   </td>
   <td>95.00
   </td>
   <td>95.30
   </td>
  </tr>
</table>

### Reproduction

The results of AIME24 and GSM8K, were obtained using forked [lm-evaluation-harness](https://github.com/BowenBao/lm-evaluation-harness/tree/cot).

### Launch Server
```
#!/bin/bash
MODEL=/models/amd/DeepSeek-R1-0528-MXFP4-V2
LOG="sglang-serving.log"

SGLANG_AITER_MLA_PERSIST=1 \
python3 -m sglang.launch_server \
--model-path $MODEL \
--tensor-parallel-size 8 \
--trust-remote-code \
--chunked-prefill-size 131072 \
--host 0.0.0.0 \
--port 8321 \
--disable-radix-cache \
--mem-fraction-static 0.8 \
--max-running-requests 64 \
--attention-backend aiter 2>&1 | tee $LOG
```

### AIME24
```
lm_eval --model local-completions \
    --model_args model=/models/amd/DeepSeek-R1-0528-MXFP4-V2,base_url=http://0.0.0.0:8321/v1/completions,num_concurrent=999999,timeout=999999,tokenized_requests=False,max_length=32000,temperature=0.6,top_p=0.95 \
    --tasks aime24 \
    --num_fewshot 0 \
    --gen_kwargs "do_sample=True,temperature=0.6,top_p=0.95,max_tokens=32000" \
    --batch_size auto 2>&1 | tee aime24.log
```

### GSM8K
```
lm_eval --model local-completions \
    --model_args model=/models/amd/DeepSeek-R1-0528-MXFP4-V2,base_url=http://0.0.0.0:8321/v1/completions,num_concurrent=256,max_retries=10,max_gen_toks=2048,tokenized_requests=False \
    --tasks gsm8k \
    --num_fewshot 5 \
    --batch_size auto 2>&1 | tee gsm8k.log
```

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