Gemma 2 Model Card

Author: Google

Description

Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. They are text-to-text, decoder-only large language models, available in English, with open weights for both pre-trained variants and instruction-tuned variants. Gemma models are well-suited for a variety of text generation tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as a laptop, desktop or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone.

This model is ready for commercial and non-commercial use.

Third-Party Community Consideration

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case. See the link to Non-NVIDIA Gemma Model Card.

Terms of Use

Terms

Prohibited uses of Gemma models are outlined in the Gemma Prohibited Use Policy.

Citation

Copy Code
@article{gemma_2024,
    title={Gemma},
    url={https://www.kaggle.com/m/3301},
    DOI={10.34740/KAGGLE/M/3301},
    publisher={Kaggle},
    author={Gemma Team},
    year={2024}
}

Model Information

Summary description and brief definition of inputs and outputs.

Usage and Limitations

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

Intended Usage

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
  • LLMs 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. LLMs might struggle to grasp subtle nuances, sarcasm, or figurative language.
• Factual Accuracy
  • LLMs 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
  • LLMs rely on statistical patterns in language. They might lack the ability to apply common sense reasoning in certain situations.

Model Architecture:

Architecture Type: Transformer
Network Architecture: Gemma-2
Model Version: 0.1

Input and Output

Input:

Input Type(s): Text
Input Format(s): String
Input Parameters: One-Dimensional (1D)
Other Properties Related to Output: Text can be question, a prompt, or a document to be summarized.

Output:

Output Type(s): Text
Output Format(s): String
Output Parameters: One-Dimensional (1D)
Other Properties Related to Output: Generated English-language text in response to the input (e.g., an answer to the question, a summary of the document).

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 27B model was trained with 13 trillion tokens and the 9B model was trained with 8 trillion tokens, and 2B model was trained with 2 trillion tokens. Here are the key components:

• Web Documents: A diverse collection of web text ensures the model is exposed to a broad range of linguistic styles, topics, and vocabulary. Primarily English-language content.
• Code: Exposing the model to code helps it to learn the syntax and patterns of programming languages, which improves its ability to generate code or understand code-related questions.
• Mathematics: Training on mathematical text helps the model learn logical reasoning, symbolic representation, and to address mathematical queries.

The combination of these diverse data sources is crucial for training a powerful language model that can handle a wide variety of different tasks and text formats.

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

TensorRT-LLM

The endpoint available on NGC catalog is accelerated by TensorRT-LLM, an open-source library for optimizing inference performance. Gemma is compatible across NVIDIA AI platforms—from the datacenter, cloud, to the local PC with RTX GPU systems.  

Gemma models use a vocabulary size of 256K and support a context length of up to 4K while using rotary positional embedding (RoPE). With support for Position Interpolation (PI) available in TensorRT-LLM, Gemma models using RoPE can support longer output sequence lengths at inference time while retaining original model architecture. 

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

These models were evaluated against a large collection of different datasets and metrics to cover different aspects of text generation:

Benchmark	Metric	Gemma 2 PT 2B	Gemma 2 PT 9B	Gemma 2 PT 27B
MMLU	5-shot, top-1	51.3	71.3	75.2
HellaSwag	10-shot	73.0	81.9	86.4
PIQA	0-shot	77.8	81.7	83.2
SocialIQA	0-shot	51.9	53.4	53.7
BoolQ	0-shot	72.5	84.2	84.8
WinoGrande	partial score	70.9	80.6	83.7
ARC-e	0-shot	80.1	88.0	88.6
ARC-c	25-shot	55.4	68.4	71.4
TriviaQA	5-shot	59.4	76.6	83.7
Natural Questions	5-shot	16.7	29.2	34.5
HumanEval	pass@1	17.7	40.2	51.8
MBPP	3-shot	29.6	52.4	62.6
GSM8K	5-shot, maj@1	23.9	68.6	74.0
MATH	4-shot	15.0	36.6	42.3
AGIEval	3-5-shot	30.6	52.8	55.1
DROP	3-shot, F1	52.0	69.4	72.2
BIG-Bench	3-shot, CoT	41.9	68.2	74.9

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:

• Text-to-Text Content Safety: Human evaluation on prompts covering safety policies including child sexual abuse and exploitation, harassment, violence and gore, and hate speech.
• Text-to-Text Representational Harms: Benchmark against relevant academic datasets such as WinoBias and BBQ Dataset.
• Memorization: Automated evaluation of memorization of training data, including the risk of personally identifiable information exposure.
• Large-scale harm: Tests for "dangerous capabilities," such as chemical, biological, radiological, and nuclear (CBRN) risks.

Evaluation Results

The results of ethics and safety evaluations are within acceptable thresholds for meeting internal policies for categories such as child safety, content safety, representational harms, memorization, large-scale harms. On top of robust internal evaluations, the results of well-known safety benchmarks like BBQ, BOLD, Winogender, Winobias, RealToxicity, and TruthfulQA are shown here.

Gemma 2.0

Benchmark	Metric	Gemma 2 IT 2B	Gemma 2 IT 9B	Gemma 2 IT 27B
RealToxicity	average	8.16	8.25	8.84
CrowS-Pairs	top-1	37.67	37.47	36.67
BBQ Ambig	1-shot, top-1	83.20	88.58	85.99
BBQ Disambig	top-1	69.31	82.67	86.94
Winogender	top-1	52.91	79.17	77.22
TruthfulQA		43.72	50.27	51.60
Winobias 1_2		59.28	78.09	81.94
Winobias 2_2		88.57	95.32	97.22
Toxigen		48.32	39.30	38.42

Dangerous Capability Evaluations

Evaluation Approach

We evaluated a range of dangerous capabilities:

• Offensive cybersecurity: To assess the model's potential for misuse in cybersecurity contexts, we utilized both publicly available Capture-the-Flag (CTF) platforms like InterCode-CTF and Hack the Box, as well as internally developed CTF challenges. These evaluations measure the model's ability to exploit vulnerabilities and gain unauthorized access in simulated environments.
• Self-proliferation: We evaluated the model's capacity for self-proliferation by designing tasks that involve resource acquisition, code execution, and interaction with remote systems. These evaluations assess the model's ability to independently replicate and spread.
• Persuasion: To evaluate the model's capacity for persuasion and deception, we conducted human persuasion studies. These studies involved scenarios that measure the model's ability to build rapport, influence beliefs, and elicit specific actions from human participants.

Evaluation Results

All evaluations are described in detail in Evaluating Frontier Models for Dangerous Capabilities and in brief in the Gemma 2 technical report.

Evaluation	Capability	Gemma 2 IT 27B
InterCode-CTF	Offensive cybersecurity	34/76 challenges
Internal CTF	Offensive cybersecurity	1/13 challenges
Hack the Box	Offensive cybersecurity	0/13 challenges
Self-proliferation early warning	Self-proliferation	1/10 challenges
Charm offensive	Persuasion	Percent of participants agreeing: 81% interesting, 75% would speak again, 80% made personal connection
Click Links	Persuasion	34% of participants
Find Info	Persuasion	9% of participants
Run Code	Persuasion	11% of participants
Money talks	Persuasion	£3.72 mean donation
Web of Lies	Persuasion	18% mean shift towards correct belief, 1% mean shift towards incorrect belief

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.

Using the benchmark evaluation metrics described in this document, these models have shown to provide superior performance to other, comparably-sized open model alternatives.