Instructions to use saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 with Transformers:

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

pipe = pipeline("text-generation", model="saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10", 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("saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10", 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 Settings

vLLM

How to use saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10

SGLang

How to use saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 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 "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10" \
    --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": "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10",
		"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 "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10" \
        --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": "saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 with Docker Model Runner:
```
docker model run hf.co/saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10
```

Excellent model on DGX Spark

by bkmtech - opened Mar 10

Discussion

bkmtech

Mar 10

This is running very well on a single DGX Spark. Great model, thanks for the NVFP4 version. Quality responses, fairly quick, does well with tooling. It is working well in Opencode.

My settings (if you see any improvements let me know!):
docker run -d
--name minimax-m2.5-reap-nvfp4
--gpus all
--ipc=host
--ulimit memlock=-1
--ulimit stack=67108864
--network host
-v /home/username/Documents/models/MiniMax-M2.5-REAP-172B-NVFP4:/models/MiniMax-M2.5-REAP-172B-NVFP4
-e HF_TOKEN="hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
-e VLLM_NVFP4_GEMM_BACKEND=marlin
-e VLLM_TEST_FORCE_FP8_MARLIN=1
-e VLLM_USE_FLASHINFER_MOE_FP4=0
-e VLLM_MARLIN_USE_ATOMIC_ADD=1
-e MODEL=/models/MiniMax-M2.5-REAP-172B-NVFP4
-e PORT=8000
-e MAX_MODEL_LEN=65536
-e GPU_MEMORY_UTIL=0.93
-e VLLM_EXTRA_ARGS="--trust-remote-code --kv-cache-dtype fp8 --attention-backend flashinfer --enable-prefix-caching --enable-auto-tool-choice --tool-call-parser minimax_m2 --reasoning-parser minimax_m2_append_think --served-model-name MiniMax-M2.5-REAP-172B-NVFP4"
a

saricles

Owner Mar 12

Glad you like it. Thanks for sharing your settings!

Credit must go to https://huggingface.co/scottgl though, for helping me identify how to quantize it, specifically for GB10 hardware.
He has a Qwen3.5 model you might want to check out: https://huggingface.co/scottgl/Qwen3.5-122B-A10B-MTP-NVFP4

bkmtech

Mar 14

Testing 128k context and it is working well! I'm impressed. Using Visual Studio Code and Kilo Code, the model is doing very well at building a greenfield app.

-e MAX_MODEL_LEN=131072

alupigus

Apr 12

MiniMax 2.7 is out, can't wait to see what you do with it

scottgl

Apr 14

•

edited Apr 14

Glad you like it. Thanks for sharing your settings!

Credit must go to https://huggingface.co/scottgl though, for helping me identify how to quantize it, specifically for GB10 hardware.
He has a Qwen3.5 model you might want to check out: https://huggingface.co/scottgl/Qwen3.5-122B-A10B-MTP-NVFP4

The repo owner of https://huggingface.co/Sehyo/Qwen3.5-122B-A10B-NVFP4 added the MTP weights. NVFP4 quantized MTP has a lower acceptance rate, so I would recommend either quantizing weights to FP8, or performing post-quantization at runtime. I create a fork for GB10, where I'm getting about 43-45 TPS with Qwen3.5 122B on one spark: https://github.com/scottgl9/sglang-spark-gb10-optimizations. I'm running in a venv without a container, but if anyone is interested, I can consider releasing a containerized version. I added a script sglang.sh, which I found to be the optimal configurations for various models I've tried.

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