Instructions to use trl-lib/LFM2.5-8B-A1B-SFT-function-calling with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use trl-lib/LFM2.5-8B-A1B-SFT-function-calling with Transformers:

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

pipe = pipeline("text-generation", model="trl-lib/LFM2.5-8B-A1B-SFT-function-calling")
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("trl-lib/LFM2.5-8B-A1B-SFT-function-calling")
model = AutoModelForCausalLM.from_pretrained("trl-lib/LFM2.5-8B-A1B-SFT-function-calling", 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 trl-lib/LFM2.5-8B-A1B-SFT-function-calling with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "trl-lib/LFM2.5-8B-A1B-SFT-function-calling"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "trl-lib/LFM2.5-8B-A1B-SFT-function-calling",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/trl-lib/LFM2.5-8B-A1B-SFT-function-calling

SGLang

How to use trl-lib/LFM2.5-8B-A1B-SFT-function-calling 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 "trl-lib/LFM2.5-8B-A1B-SFT-function-calling" \
    --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": "trl-lib/LFM2.5-8B-A1B-SFT-function-calling",
		"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 "trl-lib/LFM2.5-8B-A1B-SFT-function-calling" \
        --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": "trl-lib/LFM2.5-8B-A1B-SFT-function-calling",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use trl-lib/LFM2.5-8B-A1B-SFT-function-calling with Docker Model Runner:
```
docker model run hf.co/trl-lib/LFM2.5-8B-A1B-SFT-function-calling
```

LFM2.5-8B-A1B-SFT-function-calling

LiquidAI/LFM2.5-8B-A1B fine-tuned with TRL on the 60k xLAM samples of argilla/apigen-function-calling: a query, the tools it may use, and the one or several parallel calls that answer it. All weights are trained, with the loss on the assistant turn only. One epoch on 8 H100s, with the 32 experts split 8 ways (expert parallelism) and everything else FSDP-sharded. The training script is 1_sft.py.

938 steps in 16 minutes. The loss goes from 1.16 to 0.03 and the token accuracy from 90.5% to 99.3%.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

from trl.chat_template_utils import add_response_schema, parse_response


def get_weather(city: str, unit: str) -> str:
    """
    Gets the current weather in a city.

    Args:
        city: The name of the city.
        unit: The temperature unit, "celsius" or "fahrenheit".
    """


model_id = "trl-lib/LFM2.5-8B-A1B-SFT-function-calling"
tokenizer = add_response_schema(AutoTokenizer.from_pretrained(model_id))
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16, device_map="cuda")

messages = [{"role": "user", "content": "Is it warmer in Lisbon or in Oslo right now? Answer in Fahrenheit."}]
prompt = tokenizer.apply_chat_template(messages, tools=[get_weather], add_generation_prompt=True)["input_ids"]
output = model.generate(torch.tensor([prompt], device="cuda"), max_new_tokens=128, do_sample=False)
print(parse_response(tokenizer, output[0, len(prompt) :].tolist(), prefix=prompt))

{'role': 'assistant', 'tool_calls': [{'type': 'function', 'function': {'name': 'get_weather', 'arguments': {'city': 'Lisbon', 'unit': 'fahrenheit'}}}, {'type': 'function', 'function': {'name': 'get_weather', 'arguments': {'city': 'Oslo', 'unit': 'fahrenheit'}}}], 'content': ''}

Downloads last month: 200

Safetensors

Model size

8B params

Tensor type

F32

BF16

Model tree for trl-lib/LFM2.5-8B-A1B-SFT-function-calling

Base model

LiquidAI/LFM2.5-8B-A1B-Base

Finetuned

LiquidAI/LFM2.5-8B-A1B

Finetuned

(45)

this model

Dataset used to train trl-lib/LFM2.5-8B-A1B-SFT-function-calling

Paper for trl-lib/LFM2.5-8B-A1B-SFT-function-calling

APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets

Paper • 2406.18518 • Published Jun 26, 2024 • 26