ktransformers/doc/en/MiniMax-M2.5.md

# Running MiniMax-M2.5 with SGLang and KT-Kernel

This tutorial demonstrates how to run MiniMax-M2.5 model inference using SGLang integrated with KT-Kernel for CPU-GPU heterogeneous inference. This setup enables efficient deployment of large MoE models by offloading experts to CPU.

## Table of Contents

- [Hardware Requirements](#hardware-requirements)
- [Prerequisites](#prerequisites)
- [Step 1: Download Model Weights](#step-1-download-model-weights)
- [Step 2: Launch SGLang Server](#step-2-launch-sglang-server)
- [Step 3: Send Inference Requests](#step-3-send-inference-requests)

## Hardware Requirements

**Minimum Configuration:**
- **GPU**: NVIDIA RTX 2x4090 48GB (or equivalent with at least total 48GB VRAM available)
- **CPU**: x86 CPU with AVX512BF16 support (e.g., Intel Sapphire Rapids)
- **RAM**: At least 200GB system memory
- **Storage**: ~200GB for model weights (FP8 weight, same weight folder for CPU and GPU)

## Prerequisites

Before starting, ensure you have:

1. **KT-Kernel installed**:

```
git clone https://github.com/kvcache-ai/ktransformers.git
git submodule update --init --recursive
cd kt-kernel && ./install.sh
```

2. **SGLang installed** - Install the kvcache-ai fork of SGLang (one of):

```bash
# Option A: One-click install (from ktransformers root)
./install.sh

# Option B: pip install
pip install sglang-kt
```

> Note: You may need to reinstall cudnn: `pip install nvidia-cudnn-cu12==9.16.0.29`

3. **CUDA toolkit** - Compatible with your GPU (CUDA 12.8+ recommended)
4. **Hugging Face CLI** - For downloading models:

   ```bash
   pip install huggingface-hub
   ```

## Step 1: Download Model Weights

```bash
# Create a directory for models
mkdir -p /path/to/models
cd /path/to/models

# Download MiniMax-M2.5 (FP8 for both CPU and GPU)
huggingface-cli download MiniMaxAI/MiniMax-M2.5 \
  --local-dir /path/to/minimax-m2.5
```

**Note:** Replace `/path/to/models` with your actual storage path throughout this tutorial.

## Step 2: Launch SGLang Server

Start the SGLang server with KT-Kernel integration for CPU-GPU heterogeneous inference.


### Launch Command (4x RTX 4090 Example)

```bash
python -m sglang.launch_server \
  --host 0.0.0.0 \
  --port 30005 \
  --model /path/to/minimax-m2.5 \
  --kt-weight-path /path/to/minimax-m2.5 \
  --kt-cpuinfer 96 \
  --kt-threadpool-count 2 \
  --kt-num-gpu-experts 30 \
  --kt-method FP8 \
  --kt-gpu-prefill-token-threshold 400 \
  --trust-remote-code \
  --mem-fraction-static 0.94 \
  --served-model-name MiniMax-M2.5 \
  --enable-mixed-chunk \
  --tensor-parallel-size 4 \
  --enable-p2p-check \
  --disable-shared-experts-fusion \
  --chunked-prefill-size 32658 \
  --max-total-tokens 50000 \
  --attention-backend flashinfer
```

It takes about 2~3 minutes to start the server.

See [KT-Kernel Parameters](https://github.com/kvcache-ai/ktransformers/tree/main/kt-kernel#kt-kernel-parameters) for detailed parameter tuning guidelines.

## Step 3: Send Inference Requests

Once the server is running, you can send inference requests using the OpenAI-compatible API.

### Basic Chat Completion Request

```bash
curl -s http://localhost:30005/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M2.5",
    "stream": false,
    "messages": [
      {"role": "user", "content": "hi, who are you?"}
    ]
  }'
```

### Example Response

```json
{
    "id": "e82360a51dd4465281a2b954d5237a06",
    "object": "chat.completion",
    "created": 1770980318,
    "model": "MiniMax-M2.5",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "The user is asking who I am. I should give a brief, friendly introduction about myself.\n</think>\n\nHi there! I'm MiniMax-M2.5, an AI assistant created by MiniMax. I'm here to help you with a wide range of tasks, including:\n\n- Answering questions\n- Writing and editing code\n- Explaining concepts\n- Brainstorming ideas\n- And much more!\n\nHow can I help you today?",
                "reasoning_content": null,
                "tool_calls": null
            },
            "logprobs": null,
            "finish_reason": "stop",
            "matched_stop": 200020
        }
    ],
    "usage": {
        "prompt_tokens": 44,
        "total_tokens": 138,
        "completion_tokens": 94,
        "prompt_tokens_details": null,
        "reasoning_tokens": 0
    },
    "metadata": {
        "weight_version": "default"
    }
}
```