Instructions to use liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8")
model = AutoModelForCausalLM.from_pretrained("liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8", device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8

SGLang

How to use liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8 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 "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8" \
    --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": "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8",
		"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 "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8" \
        --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": "liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8 with Docker Model Runner:
```
docker model run hf.co/liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8
```

Mellum2.1-12B-A2.5B-Thinking — FP8 (dynamic)

FP8 quantization of JetBrains/Mellum2.1-12B-A2.5B-Thinking, published by Liodon AI.

Quantized with llm-compressor using the FP8_DYNAMIC scheme: weights are cast to FP8 (E4M3) per-channel ahead of time, activations are quantized to FP8 dynamically per-token at inference time. No calibration dataset is needed for this scheme, so the quantized weights are numerically just a direct cast of the original — no calibration-set bias to worry about. lm_head is left unquantized (standard practice — negligible size, disproportionate quality impact if quantized).

Original size: 24.3 GB → Quantized: 12.6 GB.

Quick Start

vLLM

vllm serve liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8

Text Generation Inference (TGI)

docker run --gpus all -p 8080:80 ghcr.io/huggingface/text-generation-inference \
    --model-id liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8

SGLang

python -m sglang.launch_server --model-path liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8

FP8 execution requires an NVIDIA GPU with compute capability ≥ 8.9 (Ada/Hopper/Blackwell — RTX 40-series, L4/L40S, H100/H200, B100/B200/GB10). On older GPUs, vLLM/TGI will dequantize to run, which loses the speed/memory benefit.

Source

Model: JetBrains/Mellum2.1-12B-A2.5B-Thinking
License: other

Citation

@misc{liodonai_mellum2_1_12b_a2_5b_thinking_fp8,
  title        = {Mellum2.1-12B-A2.5B-Thinking — FP8},
  author       = {{Liodon AI}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8}},
  note         = {FP8 (dynamic) quantization of JetBrains/Mellum2.1-12B-A2.5B-Thinking}
}

Quantized by Liodon AI

Downloads last month: 2

Safetensors

Model size

12B params

Tensor type

BF16

F8_E4M3

Model tree for liodon-ai/Mellum2.1-12B-A2.5B-Thinking-FP8

Base model

JetBrains/Mellum2-12B-A2.5B-Base

Finetuned

JetBrains/Mellum2.1-12B-A2.5B-Thinking

Quantized

(9)

this model