Instructions to use nex-agi/Nex-N2-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nex-agi/Nex-N2-mini with Transformers:

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

pipe = pipeline("text-generation", model="nex-agi/Nex-N2-mini")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("nex-agi/Nex-N2-mini")
model = AutoModelForMultimodalLM.from_pretrained("nex-agi/Nex-N2-mini", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use nex-agi/Nex-N2-mini with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nex-agi/Nex-N2-mini"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nex-agi/Nex-N2-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nex-agi/Nex-N2-mini

SGLang

How to use nex-agi/Nex-N2-mini 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 "nex-agi/Nex-N2-mini" \
    --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": "nex-agi/Nex-N2-mini",
		"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 "nex-agi/Nex-N2-mini" \
        --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": "nex-agi/Nex-N2-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nex-agi/Nex-N2-mini with Docker Model Runner:
```
docker model run hf.co/nex-agi/Nex-N2-mini
```

Very good model, punches above its weight.

#13

by tarruda - opened Jul 14

Discussion

tarruda

Jul 14

There's something fundamentally different about this model vs other finetunes, and I noticed you trained this one on top of Qwen 3.5 35B Base which is even more impressive. Any chance you could release the dataset?

Also, do you have any plans to train other Qwen 3.5 variants? A Nex N2 Flash based on Qwen 3.5 122B would be amazing.

BingoBird

Aug 28

There's something fundamentally different about this model vs other finetunes

Please describe your evaluation in more detail. Thank you.

tarruda

Aug 28

Please describe your evaluation in more detail. Thank you.

I have a local benchmark where I give the LLM and harness a llama.cpp checkout and ask a question in a way that requires the model to dive into the code and fully understand its nuances. Before Nex N2 Pro, only GPT 5.x passed this consistently. Nex N2 Pro also managed to pass it (a 2-bit quant BTW) and Nex N2 mini also passes it very frequently (though sometimes it gets stuck in a loop). Even models being released today have trouble with this benchmark. Since it is a private evaluation, it can't have leaked into the models training data.

I attribute the "fundamentally different" part to how this model (and Nex N2 Pro) reasons. That "caveman" thinking style is starting to get more popular, and even deepseek v4 0731 and qwen 3.8 seem to mimic it sometimes. Nex N2 models were the first ones I saw using this style.

BingoBird

Aug 28

Before Nex N2 Pro, only GPT 5.x passed this consistently.

Thank you. I am sure I and some other people will benefit from this.

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