Instructions to use JetBrains/Mellum2-12B-A2.5B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use JetBrains/Mellum2-12B-A2.5B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="JetBrains/Mellum2-12B-A2.5B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("JetBrains/Mellum2-12B-A2.5B-Instruct")
model = AutoModelForCausalLM.from_pretrained("JetBrains/Mellum2-12B-A2.5B-Instruct")
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 JetBrains/Mellum2-12B-A2.5B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "JetBrains/Mellum2-12B-A2.5B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "JetBrains/Mellum2-12B-A2.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/JetBrains/Mellum2-12B-A2.5B-Instruct

SGLang

How to use JetBrains/Mellum2-12B-A2.5B-Instruct 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 "JetBrains/Mellum2-12B-A2.5B-Instruct" \
    --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": "JetBrains/Mellum2-12B-A2.5B-Instruct",
		"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 "JetBrains/Mellum2-12B-A2.5B-Instruct" \
        --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": "JetBrains/Mellum2-12B-A2.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use JetBrains/Mellum2-12B-A2.5B-Instruct with Docker Model Runner:
```
docker model run hf.co/JetBrains/Mellum2-12B-A2.5B-Instruct
```

Request: MXFP6 Quantization :)

by dopeydon - opened 2 days ago

Discussion

dopeydon

2 days ago

Hi everyone,

First of all, thank you so much for releasing the Mellum 2 models! The architecture is incredibly promising, especially for local software development on consumer hardware.

I noticed that there are already excellent MXFP4 quantizations available, but 4-bit precision can sometimes cross the threshold where critical syntax logic or subtle coding nuances are lost.

I would be incredibly grateful if someone from the team or the community could create an MXFP6 quantization of the model.

The Ultimate Sweet Spot: MXFP6 bridges the gap perfectly between MXFP4 and FP8. It offers significantly better perplexity and code-syntax retention than 4-bit, while remaining much lighter than 8-bit.

Consumer Hardware Friendly: An MXFP6 version would fit beautifully into the VRAM of standard consumer GPUs 16GB cards, leaving headroom for the context window without sacrificing the model's capabilities.

If anyone with the compute resources could cook an MXFP6 version (compatible with to Radeon), it would be an absolute game-changer for independent developers running this locally. I (and I'm sure many others in the community) would appreciate your time and effort immensely!

Thank you so much for your hard work and dedication to open-source LLMs!

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