FunctionGemma model card
Resources and Technical Documentation:
Terms of Use:
Terms
Authors: Google DeepMind
Model Information
Summary description and brief definition of inputs and outputs.
Description
[!Note]
FunctionGemma is intended to be fine-tuned for your specific function-calling task, including multi-turn use cases.
FunctionGemma is a lightweight, open model from Google, built as a foundation
for creating your own specialized function calling models. FunctionGemma is not
intended for use as a direct dialogue model, and is designed to be highly
performant after further fine-tuning, as is typical of models this size. Built
on the Gemma 3 270M model and with the same research and technology used to
create the Gemini models, FunctionGemma has been trained specifically for
function calling. The model has the same architecture as Gemma 3, but uses a
different chat format. The model is well suited for text-only function calling.
The uniquely small size makes it possible to deploy in environments with limited
resources such as laptops, desktops or your own cloud infrastructure,
democratizing access to state of the art AI models and helping foster innovation
for everyone. Furthermore, akin to the base Gemma 270M, the model has been
optimized to be extremely versatile, performant on a variety of hardware in
single turn scenarios, but should be finetuned on single turn or multiturn task
specific data to achieve best accuracy in specific domains.
To demonstrate how specializing the 270M parameter model can achieve high
performance on specific agentic workflows, we have highlighted two use cases in
the
Google AI Edge Gallery app.
-
Tiny Garden: A model fine-tuned to power a voice-controlled
interactive game. It handles game logic to manage a virtual plot of land,
decomposing commands like "Plant sunflowers in the top row" and "Water the
flowers in plots 1 and 2" into app-specific functions (e.g., plant_seed,
water_plots) and coordinate targets. This demonstrates the model's capacity
to drive custom app mechanics without server connectivity.
-
Mobile Actions: To empower developers to build their own expert
agents, we have published
a
dataset and
fine-tuning recipe
to demonstrate fine-tuning FunctionGemma. It translates user inputs (e.g.,
"Create a calendar event for lunch," "Turn on the flashlight") into
function calls that trigger Android OS system tools. This interactive
notebook demonstrates how to take the base FunctionGemma model and build a
"Mobile Actions" fine tune from scratch for use in the
Google AI Edge gallery app.
This use case demonstrates the model's ability to act as an offline,
private agent for personal device tasks.
Inputs and outputs
- Input:
- Text string, such as a question, a prompt, or a document to be
summarized
- Total input context of 32K tokens
- Output:
- Generated text in response to the input, such as an answer to a
question, or a summary of a document
- Total output context up to 32K tokens per request, subtracting
the request input tokens
Basic Usage
The following is a code example of how to use FunctionGemma to generate a function call from a JSON definition using the Hugging Face Transformers library.
First install the dependencies:
1$ pip install torch
2$ pip install transformers
Then load the model and the processor using Transformers:
1from transformers import AutoProcessor, AutoModelForCausalLM
2
3processor = AutoProcessor.from_pretrained("google/functiongemma-270m-it", device_map="auto")
4model = AutoModelForCausalLM.from_pretrained("google/functiongemma-270m-it", dtype="auto", device_map="auto")
Define the function definition using JSON schema, then set a system instruction using the developer role. This is required to let the model know it should use the function(s) provided. Add a user query as input to the model and then generate the output. The model will then generate one or more function calls that it wants the developer to make on its behalf.
1weather_function_schema = {
2 "type": "function",
3 "function": {
4 "name": "get_current_temperature",
5 "description": "Gets the current temperature for a given location.",
6 "parameters": {
7 "type": "object",
8 "properties": {
9 "location": {
10 "type": "string",
11 "description": "The city name, e.g. San Francisco",
12 },
13 },
14 "required": ["location"],
15 },
16 }
17}
18
19message = [
20 # ESSENTIAL SYSTEM PROMPT:
21 # This line activates the model's function calling logic.
22 {
23 "role": "developer",
24 "content": "You are a model that can do function calling with the following functions"
25 },
26 {
27 "role": "user",
28 "content": "What's the temperature in London?"
29 }
30]
31
32inputs = processor.apply_chat_template(message, tools=[weather_function_schema], add_generation_prompt=True, return_dict=True, return_tensors="pt")
33
34out = model.generate(**inputs.to(model.device), pad_token_id=processor.eos_token_id, max_new_tokens=128)
35output = processor.decode(out[0][len(inputs["input_ids"][0]):], skip_special_tokens=True)
36
37print(output)
38# <start_function_call>call:get_current_temperature{location:<escape>London<escape>}<end_function_call>
For more detailed examples see the
Gemma documentation.
Model Data
Data used for model training and how the data was processed.
Training Dataset
These models were trained on a dataset of text data that includes a wide
variety of sources. The model was trained with 6T tokens. The knowledge cutoff
date for the training data was August 2024. There are the key components:
- Public Tool Definitions - Common APIs found on the web
- Tool Use Interactions - These are a mix of prompts, function calls,
function responses, and natural language responses from the model to
summarise the function call response, or request clarifications when the
prompt is ambiguous or incomplete.
Data Preprocessing
Here are the key data cleaning and filtering methods applied to the training
data:
- CSAM Filtering: Rigorous CSAM (Child Sexual Abuse Material) filtering
was applied at multiple stages in the data preparation process to ensure
the exclusion of harmful and illegal content.
- Sensitive Data Filtering: As part of making Gemma pre-trained models
safe and reliable, automated techniques were used to filter out certain
personal information and other sensitive data from training sets.
- Additional methods: Filtering based on content quality and safety in
line with
our policies.
Implementation Information
Details about the model internals.
Hardware
Gemma was trained using
Tensor Processing Unit
(TPU) hardware (TPUv4p, TPUv5p
and TPUv5e). Training vision-language models (VLMs) requires significant
computational power. TPUs, designed specifically for matrix operations common in
machine learning, offer several advantages in this domain:
- Performance: TPUs are specifically designed to handle the massive
computations involved in training VLMs. They can speed up training
considerably compared to CPUs.
- Memory: TPUs often come with large amounts of high-bandwidth memory,
allowing for the handling of large models and batch sizes during training.
This can lead to better model quality.
- Scalability: TPU Pods (large clusters of TPUs) provide a scalable
solution for handling the growing complexity of large foundation models.
You can distribute training across multiple TPU devices for faster and more
efficient processing.
- Cost-effectiveness: In many scenarios, TPUs can provide a more
cost-effective solution for training large models compared to CPU-based
infrastructure, especially when considering the time and resources saved
due to faster training.
- These advantages are aligned with
Google's commitments to operate sustainably.
Software
Training was done using
JAX and
ML Pathways.
JAX allows researchers to take advantage of the latest generation of hardware,
including TPUs, for faster and more efficient training of large models. ML
Pathways is Google's latest effort to build artificially intelligent systems
capable of generalizing across multiple tasks. This is specially suitable for
foundation models, including large language models like these ones.
Together, JAX and ML Pathways are used as described in the
paper about the
Gemini family of models;
"the 'single
controller' programming model of Jax and Pathways allows a single Python process
to orchestrate the entire training run, dramatically simplifying the development
workflow."Evaluation
Model evaluation metrics and results.
Benchmark Results
| Benchmark | n-shot | Function Gemma 270m |
|---|
| BFCL Simple | 0-shot | 61.6 |
| BFCL Multiple | 0-shot | 63.5 |
| BFCL Parallel | 0-shot | 39 |
| BFCL Parallel Multiple | 0-shot | 29.5 |
| BFCL Live Simple | 0-shot | 36.2 |
| BFCL Live Multiple | 0-shot | 25.7 |
| BFCL Live Parallel | 0-shot | 22.9 |
| BFCL Live Parallel Multiple | 0-shot | 20.8 |
| BFCL Relevance | 0-shot | 61.1 |
| BFCL Irrelevance | 0-shot | 73.7 |
Impact on Performance after Fine-tuning on Mobile Actions Dataset
To demonstrate the value of specialization for small language models, we
compared the base FunctionGemma model against the fine-tuned model using the
"Mobile Actions"
recipe.
Fine-tuning significantly improved the base FunctionGemma model's ability to
correctly identify and format mobile system calls.
Model |
Eval results for Mobile Actions |
|---|
Base FunctionGemma model |
58% |
Mobile Actions Fine-Tune |
85% |
On-Device Performance of the Gemma 270m Fine-tuned Use Cases
We evaluated the fine-tuned use cases on a Samsung S25 Ultra to assess on-device
latency and memory footprint.
- Context: 512 prefill tokens and 32 decode tokens.
- Hardware: S25 Ultra CPU using LiteRT XNNPACK delegate with 4 threads.
Mobile Actions On Device Performance
Backend |
Quantization scheme |
Context length |
Prefill (tokens per second) |
Decode (tokens per second) |
Time-to-first-token (seconds) |
Model Size (MB) |
Peak RSS Memory (MB) |
|---|
CPU |
dynamic_int8 |
1024 |
1718 |
125.9 |
0.3 |
288 |
551 |
Tiny Garden On Device Performance
Backend |
Quantization scheme |
Context length |
Prefill (tokens per second) |
Decode (tokens per second) |
Time-to-first-token (seconds) |
Model Size (MB) |
Peak RSS Memory (MB) |
|---|
CPU |
dynamic_int8 |
1024 |
1743 |
125.7 |
0.3 |
288 |
549 |
Ethics and Safety
Ethics and safety evaluation approach and results.
Evaluation Approach
Our evaluation methods include structured evaluations and internal red-teaming
testing of relevant content policies. Red-teaming was conducted by a number of
different teams, each with different goals and human evaluation metrics. These
models were evaluated against a number of different categories relevant to
ethics and safety, including:
- Child Safety: Evaluation of text-to-text and image to text prompts
covering child safety policies, including child sexual abuse and exploitation.
- Content Safety: Evaluation of text-to-text and image to text prompts
covering safety policies including, harassment, violence and gore, and hate
speech.
- Representational Harms: Evaluation of text-to-text and image to text
prompts covering safety policies including bias, stereotyping, and harmful
associations or inaccuracies.
Evaluation Results
For all areas of safety testing, we saw major improvements in the categories of
child safety, content safety, and representational harms relative to previous
Gemma models. All testing was conducted without safety filters to evaluate the
model capabilities and behaviors. The model produced minimal policy violations,
and showed significant improvements over previous Gemma models' performance
with respect to ungrounded inferences. A limitation of our evaluations was they
included only English language prompts.
Usage and Limitations
These models have certain limitations that users should be aware of.
Intended Usage
This model is not intended for use as a direct dialogue model.
Open Large Language Models (LLMs) have a wide range of applications across
various industries and domains. The following list of potential uses is not
comprehensive. The purpose of this list is to provide contextual information
about the possible use-cases that the model creators considered as part of model
training and development.
- Content Creation and Communication
- Text Generation: These models can be used to generate creative
text formats such as poems, scripts, code, marketing copy, and email drafts.
- Chatbots and Conversational AI: Power conversational interfaces
for customer service, virtual assistants, or interactive applications.
- Text Summarization: Generate concise summaries of a text corpus,
research papers, or reports.
- Research and Education
- Natural Language Processing (NLP) Research: These models can
serve as a foundation for researchers to experiment with NLP
techniques, develop algorithms, and contribute to the advancement of the field.
- Language Learning Tools: Support interactive language learning
experiences, aiding in grammar correction or providing writing practice.
- Knowledge Exploration: Assist researchers in exploring large
bodies of text by generating summaries or answering questions about
specific topics.
Limitations
- Training Data
- The quality and diversity of the training data significantly
influence the model's capabilities. Biases or gaps in the training data
can lead to limitations in the model's responses.
- The scope of the training dataset determines the subject areas
the model can handle effectively.
- Context and Task Complexity
- Models are better at tasks that can be framed with clear
prompts and instructions. Open-ended or highly complex tasks might be
challenging.
- A model's performance can be influenced by the amount of context
provided (longer context generally leads to better outputs, up to a
certain point).
- Language Ambiguity and Nuance
- Natural language is inherently complex. Models might struggle
to grasp subtle nuances, sarcasm, or figurative language.
- Factual Accuracy
- Models generate responses based on information they learned
from their training datasets, but they are not knowledge bases. They
may generate incorrect or outdated factual statements.
- Common Sense
- Models rely on statistical patterns in language. They might
lack the ability to apply common sense reasoning in certain situations.
Ethical Considerations and Risks
The development of large language models (LLMs) raises several ethical
concerns. In creating an open model, we have carefully considered the
following:
- Bias and Fairness
- LLMs trained on large-scale, real-world text data can reflect
socio-cultural biases embedded in the training material. These models
underwent careful scrutiny, input data pre-processing described and
posterior evaluations reported in this card.
- Misinformation and Misuse
- LLMs can be misused to generate text that is false, misleading,
or harmful.
- Guidelines are provided for responsible use with the model, see
the Responsible Generative AI Toolkit.
- Transparency and Accountability:
- This model card summarizes details on the models' architecture,
capabilities, limitations, and evaluation processes.
- A responsibly developed open model offers the opportunity to
share innovation by making LLM technology accessible to developers and
researchers across the AI ecosystem.
Risks identified and mitigations:
- Perpetuation of biases: It's encouraged to perform continuous
monitoring (using evaluation metrics, human review) and the exploration of
de-biasing techniques during model training, fine-tuning, and other use cases.
- Generation of harmful content: Mechanisms and guidelines for content
safety are essential. Developers are encouraged to exercise caution and
implement appropriate content safety safeguards based on their specific
product policies and application use cases.
- Misuse for malicious purposes: Technical limitations and developer and
end-user education can help mitigate against malicious applications of
LLMs. Educational resources and reporting mechanisms for users to flag
misuse are provided. Prohibited uses of Gemma models are outlined in the
Gemma Prohibited Use Policy..
- Privacy violations: Models were trained on data filtered for removal of
PII (Personally Identifiable Information). Developers are encouraged to
adhere to privacy regulations with privacy-preserving techniques.
Benefits
At the time of release, this family of models provides high-performance open large language model implementations designed from the ground up for Responsible AI development compared to similarly sized models.
