Instructions to use LiquidAI/LFM2-24B-A2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use LiquidAI/LFM2-24B-A2B with Transformers:

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

pipe = pipeline("text-generation", model="LiquidAI/LFM2-24B-A2B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("LiquidAI/LFM2-24B-A2B")
model = AutoModelForCausalLM.from_pretrained("LiquidAI/LFM2-24B-A2B", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use LiquidAI/LFM2-24B-A2B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "LiquidAI/LFM2-24B-A2B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "LiquidAI/LFM2-24B-A2B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/LiquidAI/LFM2-24B-A2B

SGLang

How to use LiquidAI/LFM2-24B-A2B 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 "LiquidAI/LFM2-24B-A2B" \
    --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": "LiquidAI/LFM2-24B-A2B",
		"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 "LiquidAI/LFM2-24B-A2B" \
        --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": "LiquidAI/LFM2-24B-A2B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use LiquidAI/LFM2-24B-A2B with Docker Model Runner:
```
docker model run hf.co/LiquidAI/LFM2-24B-A2B
```

When LiquidAI/LFM2.5-24B-A2B ?

#18

by testamentaddress01 - opened Jun 23

Discussion

testamentaddress01

Jun 23

LiquidAI/LFM2.5-8B-A1B has reportedly undergone post-training with an expanded tokenizer vocabulary, but I still believe LiquidAI/LFM2-24B-A2B is a more powerful model in terms of tool calling and adherence to instructions. Therefore, I am eagerly awaiting the release of LiquidAI/LFM2.5-24B-A2B. Thank you.

space-spike

Jun 24

mlabonne

Liquid AI org Jun 24

Thanks for your messages! It's in the pipeline, coming soon-ish :)

buzzwang

Jul 2

Please please please do a 64k context, and adjust for tool use with Hermes Agent. I've said it before somewhere else, but I wanna make sure this gets through.

Your models absolutely SCREAM on my home hardware and I would love to implement them, but can't cuz of those two issues. I don't know what's all involved with this kind of adjustment, but I'm really hoping its something we can make happen!!

testamentaddress01

30 days ago

It has been quite some time since the last reply. I would like to ask @mlabonne Has LFM2.5-24B-A2B been cancelled release?

space-spike

30 days ago

•

edited 30 days ago

Could be that they are waiting for these hyper July AI launch weeks to pass for them to have a less saturated media space and then announce it. Who'll bat an eye on smaller models right now when new SOTA's are coming out left and right.

I'm still hoping they haven't given up on the small to medium tear models like the 24b (or even a bit larger if they go the route of competing with Qwen local models, seeing what ternary models are able to do right now).

mlabonne

Liquid AI org 29 days ago

We're still cooking 🧑‍🍳

Theory-of-mind

9 days ago

I really hope that the LFM2.5-24B-A2B will have an extended list of supported languages, like the 2.6B model. In any case, I wish you success in development.

cogitech

8 days ago

•

edited 8 days ago

I think the minimum appropriate context for a 24B model is 128k, ideally 256, but that's because I am using Hermes, which is a very greedy model.

You are probably close to done cooking now, but my request would be to prioritize instruction following and long-context fidelity over "world knowledge". These days an LLM-driven agent can look stuff up if they don't know the facts, but if they can't use skills/tools properly they are virtually useless (regardless how smart they are).

A junior employee who does things by the book (and looks up what they don't know) is a far better co-worker than a rogue genius who always assumes they know a "better" way.

krampenschiesser

2 days ago

@mlabonne : How is the cooking going 😃 I want to do some sft+grpo and am wondering if I should wait for the new release or use this one?

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