Instructions to use LiquidAI/LFM2.5-2.6B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use LiquidAI/LFM2.5-2.6B with Transformers:

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

pipe = pipeline("text-generation", model="LiquidAI/LFM2.5-2.6B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("LiquidAI/LFM2.5-2.6B")
model = AutoModelForCausalLM.from_pretrained("LiquidAI/LFM2.5-2.6B", 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/LFM2.5-2.6B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "LiquidAI/LFM2.5-2.6B"
# 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/LFM2.5-2.6B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/LiquidAI/LFM2.5-2.6B

SGLang

How to use LiquidAI/LFM2.5-2.6B 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/LFM2.5-2.6B" \
    --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/LFM2.5-2.6B",
		"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/LFM2.5-2.6B" \
        --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/LFM2.5-2.6B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use LiquidAI/LFM2.5-2.6B with Docker Model Runner:
```
docker model run hf.co/LiquidAI/LFM2.5-2.6B
```

Quick Thoughts on Liquid AI’s LFM 2.5-2.6B Release

by Ashacorporation - opened 2 days ago

Discussion

Ashacorporation

2 days ago

Huge shoutout to tim Liquid for the LFM2.5-2.6B drop. It’s wild seeing the AI Agent trend packed into such a compact model (Under 2.5 GB!). As someone who’s been keeping tabs on tim Liquid, this memory efficiency is honestly a breath of fresh air. Fingers crossed we get a solid Instruct model next with broader multilingual support, including Indonesian.

Skipping basic benchmarks like GPQA and MMLU makes total sense since the model is hyper-focused on agentic performance. Their diversification move swapping them for a competitive reasoning benchmark like AIME2026 is actually pretty clever and proves the logic is still pretty good.

While everyone else is busy chasing the giant parameter hype,, this model hits a totally different niche. Tim Liquid just proved that for on-device agents, being compact and functional is equally crucial as chasing massive numbers. Can't wait to see local agentic workflows running on this thing. Good job.

juiceb0xc0de

1 day ago

•

edited 1 day ago

It’s wild seeing the AI Agent trend packed into such a compact model (Under 2.5 GB!).

Hey just wanted to help out here, parameter size and file size are two different things. Perhaps you were looking at the version number when you point out "under 2.5 GB"? The file size is actually ~2x the parameter size for models. If you look in the Files and versions section for models you will see the safetensors files. Add those up and you will get an accurate file size. Hope this helps! 🤗

Ashacorporation

about 23 hours ago

@juiceb0xc0de Ah, good catch! You're totally right about the raw BF16 weights being ~5.4 GB.

My bad, I should've specified I was referencing Liquid AI’s official benchmarks for local, on-device setups. When you run the recommended Q4_K_M GGUF format, the file shrinks to 1.67 GB and runs smoothly under 2.5 GB RAM during active agent inference.

According to their technical paper, this is made possible by their hybrid architecture, which replaces most traditional attention layers with gated linear convolutions, cutting the massive KV cache footprint down significantly.

So yeah, raw files are beefier, but the real-world local agent workflow is fully optimized to live under that 2.5 GB limit. Appreciate the clarity!

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