MiniMax-M2 Model Repository

This is the official MiniMax-M2 model repository containing a 230B parameter MoE model with 10B active parameters, optimized for coding and agentic workflows.

Model Information

• Model Type: Mixture of Experts (MoE)
• Total Parameters: 230B
• Active Parameters: 10B
• Architecture: Transformer-based MoE
• License: Modified MIT
• Pipeline Tag: text-generation

Usage

This model can be used with various inference frameworks:

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("your-username/MiniMax-M2")
tokenizer = AutoTokenizer.from_pretrained("your-username/MiniMax-M2")

vLLM

from vllm import LLM, SamplingParams

llm = LLM(model="your-username/MiniMax-M2")

SGLang

from sglang import function, system, user, assistant, gen, select

@function
def multi_turn_question(s, question):
    s += system("You are a helpful assistant.")
    s += user(question)
    s += assistant(gen("answer", max_tokens=256))
    return s["answer"]

Model Details

• Context Length: 128K tokens
• Thinking Format: Uses <think>...</think> tags for reasoning
• Recommended Parameters:
  • Temperature: 1.0
  • Top-p: 0.95
  • Top-k: 40

Deployment Guides

See the docs/ directory for detailed deployment guides:

• Transformers Guide
• vLLM Guide
• SGLang Guide
• MLX Guide

License

This model is released under the Modified MIT License. See the license file for details.