---
base_model:
- MiniMaxAI/MiniMax-M2.7
language:
- en
library_name: transformers
license: other
license_name: modified-mit
license_link: https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/main/LICENSE
---
# Model Overview

- **Model Architecture:** MiniMaxM2ForCausalLM
  - **Input:** Text
  - **Output:** Text
- **Supported Hardware Microarchitecture:** AMD MI300/MI350/MI355 (emulation)
- **ROCm:** 7.2.2
- **PyTorch**: 2.10.0
- **Transformers**: 5.2.0
- **Operating System(s):** Linux
- **Inference Engine:** [vLLM](https://docs.vllm.ai/en/latest/)
- **Model Optimizer:** [AMD-Quark](https://quark.docs.amd.com/latest/index.html) (v0.12)
- **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)
  - **Quantized layers:** `experts`
  - **Weight quantization:** NVFP4, Static 
  - **Activation quantization:** NVFP4, Dynamic


# Model Quantization

The model was quantized from [amd/MiniMax-M2.7-BF16](https://huggingface.co/amd/MiniMax-M2.7-BF16), originally from [MiniMax/MiniMax-M2.7](https://huggingface.co/MiniMax/MiniMax-M2.7), using [AMD-Quark](https://quark.docs.amd.com/latest/index.html). The weights and activations are quantized to NVFP4.


**Quantization scripts:**
```
cd Quark/examples/torch/language_modeling/llm_ptq
exclude_layers="lm_head *block_sparse_moe.gate* *self_attn*"
python3 quantize_quark.py --model_dir amd/MiniMax-M2.7-BF16 \
                          --quant_scheme nvfp4 \
                          --exclude_layers $exclude_layers \
                          --num_calib_data 128 \
                          --model_export hf_format \
                          --multi_gpu balanced \
                          --trust_remote_code \
                          --output_dir amd/MiniMax-M2.7-NVFP4 
```


For further details or issues, please refer to the AMD-Quark documentation or contact the respective developers.
# Deployment
## Evaluation
The model was evaluated on gsm8k benchmarks using the vllm framework.

### Accuracy

<table>
  <tr>
   <td><strong>Benchmark</strong>
   </td>
   <td><strong>MiniMaxAI/MiniMax-M2.7 </strong>
   </td>
   <td><strong>amd/MiniMax-M2.7-NVFP4(this model)</strong>
   </td>
   <td><strong>Recovery</strong>
   </td>
  </tr>
  <tr>
   <td>gsm8k (flexible-extract) 
   </td>
   <td>91.81
   </td>
   <td>92.20
   </td>
   <td>100.04%
   </td>
  </tr>
</table>


### Reproduction

The GSM8K result was obtained using the `lm-evaluation-harness` framework, based on the Docker image `rocm/vllm-dev:nightly_main_20260603`.

Install the lm-eval `(Version: 0.4.12)` in container first.
```
pip install lm-eval
pip install lm-eval[api]
```
#### Launching server
```
vllm serve \
    --model amd/MiniMax-M2.7-NVFP4 \
    --trust-remote-code \
    --host 0.0.0.0 \
    --port 8011 \
    --tensor-parallel-size 4 \
    --enable-auto-tool-choice \
    --tool-call-parser minimax_m2 \
    --reasoning-parser minimax_m2_append_think
```


#### Evaluating model in a new terminal
```
python3 vllm/tests/evals/gsm8k/gsm8k_eval.py --host http://0.0.0.0 --port 8011
```


# License
Modifications Copyright(c) 2026 Advanced Micro Devices, Inc. All rights reserved.