Access Gemma on Hugging Face

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

To access Gemma on Hugging Face, you’re required to review and agree to Google’s usage license. To do this, please ensure you’re logged in to Hugging Face and click below. Requests are processed immediately.

DataGemma RAG model card

Resources and Technical Documentation:

Terms of Use: Terms

Authors: Google

Model Information

Description

DataGemma is a series of fine-tuned Gemma 2 models used to help LLMs access and incorporate reliable public statistical data from Data Commons into their responses. DataGemma RAG is used with Retrieval Augmented Generation, where it is trained to take a user query and generate natural language queries that can be understood by Data Commons' existing natural language interface. More information can be found in this research paper.

Inputs and outputs

Input: Text string containing a user query with a prompt to ask for statistical questions.
Output: A list of natural language queries that can be used to answer the user query and can be understood by Data Commons' existing natural language interface.

Here is an example of a prompt used to get statistical questions for the user query [User Query]:

Your role is that of a Question Generator.  Given Query below, come up with a
maximum of 25 Statistical Questions that help in answering Query.

These are the only forms of Statistical Questions you can generate:
1. What is $METRIC in $PLACE?
2. What is $METRIC in $PLACE $PLACE_TYPE?
3. How has $METRIC changed over time in $PLACE $PLACE_TYPE?

where,
- $METRIC should a metric on societal topics like demographics, economy, health,
  education, environment, etc.  Examples are unemployment rate and
  life expectancy.
- $PLACE is the name of a place like California, World, Chennai, etc.
- $PLACE_TYPE is an immediate child type within $PLACE, like counties, states,
  districts, etc.

Your response should only have questions, one per line, without any numbering
or bullet.

If you cannot come up with Statistical Questions to ask for a Query, return an
empty response.

Query: [User Query]
Statistical Questions:

Usage

Below we provide a code snippet to run the fine-tuned model, which is just one step in the complete RAG approach explained in the DataGemma paper. You can try out the end-to-end RAG flow in this colab notebook.

To run this model, first make sure to pip install -U transformers accelerate, then copy the code snippet from the following section.

Running the model on a single/multi GPU

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = 'google/datagemma-rag-27b-it'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map='auto',
    torch_dtype=torch.bfloat16,
)

input_text = """Your role is that of a Question Generator.  Given Query below, come up with a
maximum of 25 Statistical Questions that help in answering Query.

These are the only forms of Statistical Questions you can generate:
1. What is $METRIC in $PLACE?
2. What is $METRIC in $PLACE $PLACE_TYPE?
3. How has $METRIC changed over time in $PLACE $PLACE_TYPE?

where,
- $METRIC should be a metric on societal topics like demographics, economy, health,
  education, environment, etc.  Examples are unemployment rate and
  life expectancy.
- $PLACE is the name of a place like California, World, Chennai, etc.
- $PLACE_TYPE is an immediate child type within $PLACE, like counties, states,
  districts, etc.

Your response should only have questions, one per line, without any numbering
or bullet.

If you cannot come up with Statistical Questions to ask for a Query, return an
empty response.

Query: What are some interesting trends in Sunnyvale spanning gender, age, race, immigration, health conditions, economic conditions, crime and education?
Statistical Questions:"""
inputs = tokenizer(input_text, return_tensors='pt').to('cuda')

outputs = model.generate(**inputs, max_new_tokens=4096)
answer = tokenizer.batch_decode(outputs[:, inputs['input_ids'].shape[1]:], skip_special_tokens=True)[0].strip()
print(answer)

Example output

What is the population of Sunnyvale?
What is the population of Sunnyvale males?
What is the population of Sunnyvale females?
What is the population of Sunnyvale asians?
What is the population of Sunnyvale blacks?
What is the population of Sunnyvale whites?
What is the population of Sunnyvale males in their 20s?
What is the population of Sunnyvale females in their 20s?
What is the population of Sunnyvale males in their 30s?
What is the population of Sunnyvale females in their 30s?
What is the population of Sunnyvale males in their 40s?
What is the population of Sunnyvale females in their 40s?
What is the population of Sunnyvale males in their 50s?
What is the population of Sunnyvale females in their 50s?
What is the population of Sunnyvale males in their 60s?
What is the population of Sunnyvale females in their 60s?
How has the population of Sunnyvale changed over time?
How has the population of Sunnyvale males changed over time?
How has the population of Sunnyvale females changed over time?
How has the population of Sunnyvale asian people changed over time?
How has the population of Sunnyvale black people changed over time?
How has the population of Sunnyvale hispanic people changed over time?
How has the population of Sunnyvale white people changed over time?
How has the score on Sunnyvale schools changed over time?
How has the number of students enrolled in Sunnyvale schools changed over time?
How has the number of students enrolled in Sunnyvale charter schools changed over time?
How has the number of students enrolled in Sunnyvale private schools changed over time?

Run in 4-bit via bitsandbytes

To run this model, first make sure to pip install -U transformers bitsandbytes accelerate, then copy the code snippet from the following section.

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch
nf4_config = BitsAndBytesConfig(
   load_in_4bit=True,
   bnb_4bit_quant_type='nf4',
   bnb_4bit_compute_dtype=torch.bfloat16,
)

model_id = 'google/datagemma-rag-27b-it'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map='auto',
    quantization_config=nf4_config,
    torch_dtype=torch.bfloat16,
)
input_text = """Your role is that of a Question Generator.  Given Query below, come up with a
maximum of 25 Statistical Questions that help in answering Query.
These are the only forms of Statistical Questions you can generate:
1. What is $METRIC in $PLACE?
2. What is $METRIC in $PLACE $PLACE_TYPE?
3. How has $METRIC changed over time in $PLACE $PLACE_TYPE?
where,
- $METRIC should be a metric on societal topics like demographics, economy, health,
  education, environment, etc.  Examples are unemployment rate and
  life expectancy.
- $PLACE is the name of a place like California, World, Chennai, etc.
- $PLACE_TYPE is an immediate child type within $PLACE, like counties, states,
  districts, etc.

Your response should only have questions, one per line, without any numbering
or bullet.

If you cannot come up with Statistical Questions to ask for a Query, return an
empty response.

Query: What are some interesting trends in Sunnyvale spanning gender, age, race, immigration, health conditions, economic conditions, crime and education?
Statistical Questions:"""
inputs = tokenizer(input_text, return_tensors='pt').to('cuda')

outputs = model.generate(**inputs, max_new_tokens=4096)
answer = tokenizer.batch_decode(outputs[:, inputs['input_ids'].shape[1]:], skip_special_tokens=True)[0].strip()
print(answer)

Citation

@misc{radhakrishnan2024knowing,
      title={Knowing When to Ask - Bridging Large Language Models and Data}, 
      author={Prashanth Radhakrishnan and Jennifer Chen and Bo Xu and Prem Ramaswami and Hannah Pho and Adriana Olmos and James Manyika and R. V. Guha},
      year={2024},
      eprint={},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://datacommons.org/link/DataGemmaPaper}, 
}

Model Data

The base model was trained on a dataset of text data that includes a wide variety of sources, see the Gemma 2 documentation for more details. The DataGemma RAG model is fine-tuned on synthetically generated data. More details can be found in the DataGemma paper.

Implementation Information

Like Gemma, DataGemma RAG was trained on TPUv5e, using JAX.

Evaluation

Evaluation on the model was done as part of evaluation on the full RAG workflow and documented in the DataGemma paper.

Ethics and Safety

We are releasing an early version of the models. They are meant for academic and research purposes and are not ready for commercial or general public use. This version was trained on a very small corpus of examples and may exhibit unintended, and at times controversial or inflammatory, behavior. Please anticipate errors and limitations as we actively develop this LLM interface.

We red teamed and checked the Data Commons Natural Language interface pre-launch against a set of potentially dangerous queries that could result in misleading, controversial, or inflammatory results.
We ran these same queries against the outputs of the RIG and RAG models, finding a few examples where query responses were controversial, but not dangerous.
As this model is meant purely for academic and research purposes, it has not been subjected to our usual safety evaluations.

Usage and Limitations

These models have certain limitations that users should be aware of.

This is a very early version of DataGemma RAG. It is meant for trusted tester use (primarily for academic and research use) and not yet ready for commercial or general public use. This version was trained on a very small corpus of examples and may exhibit unintended, and at times controversial or inflammatory behavior. Please anticipate errors and limitations as we actively develop this large language model interface.

Your feedback and evaluations are critical to refining DataGemma's performance and will directly contribute to its training process. Known limitations are detailed in the DataGemma paper, and we encourage you to consult it for a comprehensive understanding of DataGemma's current capabilities.

Downloads last month: 10,784

Safetensors

Model size

27.2B params

Tensor type

BF16

Inference Examples

Text Generation

This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.