Instructions to use google/functiongemma-270m-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/functiongemma-270m-it with Transformers:

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

pipe = pipeline("text-generation", model="google/functiongemma-270m-it")
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("google/functiongemma-270m-it")
model = AutoModelForCausalLM.from_pretrained("google/functiongemma-270m-it", 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 google/functiongemma-270m-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "google/functiongemma-270m-it"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "google/functiongemma-270m-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/google/functiongemma-270m-it

SGLang

How to use google/functiongemma-270m-it 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 "google/functiongemma-270m-it" \
    --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": "google/functiongemma-270m-it",
		"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 "google/functiongemma-270m-it" \
        --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": "google/functiongemma-270m-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use google/functiongemma-270m-it with Docker Model Runner:
```
docker model run hf.co/google/functiongemma-270m-it
```

Any plans to release 1b-it model?

#16

by uasan - opened Jan 8

Discussion

uasan

Jan 8

The 270m-it model delivers excellent performance, but for many use cases, a second, larger model is needed, resulting in lower overall performance.

For us, and perhaps for many others, the 270m isn't enough to cover all real-world use cases, but we've assessed its potential. In my opinion, the larger 1b version should cover all our use cases without the need for additional, larger models.

Thank you for your work.

pannaga10

Google org Jan 13

Hi @uasan
We're glad to hear the 270m model is serving you well and appreciate your detailed feedback on the specific need for a 1B version. I have passed your request along to the engineering team. Thanks

A-L-01

2 days ago

I agree with this, a bigger model would be very useful, although not TOO big, just slightly bigger, enough to make the accuracy scores that are achievable with fine tuning get close to stuff like 98%. I would say if I had to guess, the sweet spot may be either 0.5B or 1B, not more, unless you give us a version with QAT that can be small and use little memory while retaining the performances of the full precision version.

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