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
```

<|tool_call_start|> and <|tool_call_end|> The biggest problem with this model.

#18

by koloved - opened 12 days ago

Discussion

koloved

12 days ago

About 50% of my tool calls end with output <|tool_call_start|> and <|tool_call_end|>

I don't even know what to do with this. I'm using the official Jin template.

koloved

12 days ago

When it happens
Non-deterministic / context-dependent: the same prompt usually yields a clean text answer, but after several tool-call rounds in the conversation history (≈5+), the model "gets stuck" in tool-calling mode and emits one more tool call instead of the final text answer — even when that round has no tools available.
The chat template is not the cause: it only renders tool_calls that already exist in the assistant history; the stray pair above is generated by the model itself.
Native tool calling works correctly when tools are passed: the server returns structured tool_calls with finish_reason: "tool_calls".
What would help
Is there a reliable way to force a text-only answer (a stop sequence, a system instruction, or a special token) so the model stops emitting tool calls once no tools are available?
Does the model have a known failure mode at low quantization (Q6_K) with tool-calling under --reasoning? Is a higher-precision checkpoint or a different reasoning setting expected to mitigate it?
Would you recommend a dedicated "final answer" prompt/template for rounds where tools must not be called?

nathanrchn

Liquid AI org 11 days ago

Thanks for the report. One clarifying question: is there anything between the two tokens (a function name and arguments), or is it literally just the empty pair? Also could you share some conversations when this happens?

mtasic85

3 days ago

I get this behavior from time to time in Pi coding agent. It is hard to replicate.

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