MiniMax-M2.1-NVFP4

Note: Updated for Transformers 5.0.0 release compatibility Jan 26th, 2026.

MiniMax-M2.1-NVFP4

NVFP4 quantized version of MiniMaxAI/MiniMax-M2.1 for efficient inference on NVIDIA Blackwell GPUs.

Model Details

Quantization Details

• Format: NVFP4 with two-level scaling (block-wise FP8 + global FP32)
• Scheme: compressed-tensors with nvfp4-pack-quantized format
• Target: All linear layers in attention and MoE experts
• Group Size: 16

Requirements

• NVIDIA Blackwell GPU (RTX 5090, RTX PRO 6000, etc.)
• vLLM with flashinfer-cutlass NVFP4 support
• ~130 GB VRAM (TP=2 recommended for dual GPU setups)

Usage with vLLM

from vllm import LLM, SamplingParams

llm = LLM(
    model="GadflyII/MiniMax-M2.1-NVFP4",
    tensor_parallel_size=2,
    max_model_len=4096,
    gpu_memory_utilization=0.90,
    trust_remote_code=True,
)

sampling_params = SamplingParams(
    temperature=0.7,
    top_p=0.9,
    max_tokens=1024,
)

outputs = llm.generate(["Your prompt here"], sampling_params)
print(outputs[0].outputs[0].text)

Performance

Tested on 2x RTX PRO 6000 Blackwell (96GB each):

License

Same as base model - see MiniMaxAI/MiniMax-M2.1 for details.

Acknowledgments

• MiniMax for the original MiniMax-M2.1 model
• vLLM team for NVFP4 quantization support