Instructions to use dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B with Transformers:

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

pipe = pipeline("text-generation", model="dervig/m51Lab-MiniMax-M2.7-REAP-139B-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("dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("dervig/m51Lab-MiniMax-M2.7-REAP-139B-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 dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "dervig/m51Lab-MiniMax-M2.7-REAP-139B-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": "dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B

SGLang

How to use dervig/m51Lab-MiniMax-M2.7-REAP-139B-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 "dervig/m51Lab-MiniMax-M2.7-REAP-139B-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": "dervig/m51Lab-MiniMax-M2.7-REAP-139B-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 "dervig/m51Lab-MiniMax-M2.7-REAP-139B-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": "dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B with Docker Model Runner:
```
docker model run hf.co/dervig/m51Lab-MiniMax-M2.7-REAP-139B-A10B
```

NVFP4 quantization of m51Lab-MiniMax-M2.7-REAP-139B-A10B

by scottgl - opened Apr 15

Discussion

scottgl

Apr 15

Could you release an NVFP4 quant of this model? Thanks!

dervig changed discussion status to closed Apr 15

dervig changed discussion status to open Apr 15

dervig

Owner Apr 15

We will be releasing FP8, NVFP4, and AWQ tomorrow =)

ineedquants

Apr 15

Hey, i was waiting for a reap and are happy that you are providing us a good reap! However i dont know if it is llama.cpps fault but loading in iqs or q 4 k m takes a lot more ram than i think it should. I am restricted to 96 ram and both barely fit with 60 k context.

dervig

Owner Apr 15

Thanks for the feedback! You're absolutely right! Q4_K_M at 84 GB + default FP16 KV cache at ctx=60K does push very close to the 96 GB ceiling. The KV cache alone is ~15 GB at that context.

Quick fix: KV cache quantization adds 10-12 GB of free RAM with essentially no quality loss:
./llama-server -m ...-Q4_K_M.gguf
--cache-type-k q8_0 --cache-type-v q8_0 \
-c 60000 -ngl 99

Or for even more headroom: q4_0 cuts KV by 75 %. If you mainly want long context over absolute quality, IQ4_XS (74 GB) + q8_0 KV gives you ~15 GB spare room at ctx=60K. I'll update the model cards with an explicit memory budget table, this is important guidance that was missing. Thanks again for flagging!

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