Instructions to use cerebras/MiniMax-M2-REAP-162B-A10B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cerebras/MiniMax-M2-REAP-162B-A10B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="cerebras/MiniMax-M2-REAP-162B-A10B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cerebras/MiniMax-M2-REAP-162B-A10B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("cerebras/MiniMax-M2-REAP-162B-A10B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use cerebras/MiniMax-M2-REAP-162B-A10B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cerebras/MiniMax-M2-REAP-162B-A10B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cerebras/MiniMax-M2-REAP-162B-A10B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cerebras/MiniMax-M2-REAP-162B-A10B

SGLang

How to use cerebras/MiniMax-M2-REAP-162B-A10B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "cerebras/MiniMax-M2-REAP-162B-A10B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cerebras/MiniMax-M2-REAP-162B-A10B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "cerebras/MiniMax-M2-REAP-162B-A10B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cerebras/MiniMax-M2-REAP-162B-A10B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cerebras/MiniMax-M2-REAP-162B-A10B with Docker Model Runner:
```
docker model run hf.co/cerebras/MiniMax-M2-REAP-162B-A10B
```

thinking disables tools

by ktsaou - opened Nov 17, 2025

Discussion

ktsaou

Nov 17, 2025

When I send tools to it, I get:

{
  "error": {
    "message": "1 validation error for list[function-wrap[__log_extra_fields__()]]\n  Invalid JSON: expected value at line 1 column 1 [type=json_invalid, input_value='<think>The user is askin..., \"sort\": \"updated\"} }]', input_type=str]\n    For further information visit https://errors.pydantic.dev/2.12/v/json_invalid",
    "type": "BadRequestError",
    "param": null,
    "code": 400
  }
}

note this: type=json_invalid, input_value='<think>The user is askin...,

So, thinking interferes with tools. Any solution to this?

lazarevich

Cerebras org Nov 17, 2025

do you believe it's a specific issue with the REAP checkpoint? have you tried with vanilla MiniMax?

ktsaou

Nov 17, 2025

hm... cannot try the original. Actually this is the only one that fits into 2x rtx 6000 pro blackwell (96GB VRAM each).

ktsaou

Nov 17, 2025

Interesting. The above happens on non-stream connections. On stream connections thinking tags are sent properly, but it does not send tools at all.
I am downloading a quant to check if a 4bit version of the original works.

ktsaou

Nov 17, 2025

Found this vllm bug: https://github.com/vllm-project/vllm/issues/28852 - I commented there.

Thank you @lazarevich for the fast reply.

aaron-newsome

Nov 18, 2025

@ktsaou i've been using noctrex/MiniMax-M2-MXFP4_MOE-GGUF, HEAVILY over the last week on 2x RTX PRO 6000 Blackwell and it fits nicely. Using llama.cpp, only 92% VRAM used with --ctx-size 204800. I didn't specify any kv cache values in the start command, so I don't know what's being used. It's been SOLID, but I'll also checkout this REAP version to see how it works. When I say I've been using it HEAVILY, I mean exactly that. Mostly with OpenCode but also Cline and some others like Claude Code with CCR.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment