Models

The PaLM API is based on Google’s next generation model, PaLM 2, which excels at a variety of capabilities. PaLM 2 has been optimized for ease of use on key developer use cases and the ability to follow instructions with precision and nuance. It has variations that are trained for text and chat generation as well as text embeddings. This guide provides information about each variation to help you decide which is the best fit for your use case.

Intended use

This model is intended to be used for a wide variety of natural language processing (NLP) applications such as chat bots, text summarization, and question and answer. The embedding service allows additional NLP use cases such as document search. It is only available to use through the PaLM API or the MakerSuite web app. Your use of PaLM API is also subject to the Generative AI Prohibited Use Policy and the Additional terms of service.

Limitations

Large language models are powerful tools, but they are not without their limitations. Their versatility and applicability can sometimes lead to unexpected outputs, such as outputs that are inaccurate, biased, or offensive. Post-processing, and rigorous manual evaluation are essential to limit the risk of harm from such outputs. See the safety guidance for additional safe use suggestions.

Model

This section provides more specific details about the model and each model variation.

Model attributes

The table below describes the attributes of the PaLM 2 model which are common to all the model variations.

Attribute Description
Training data PaLM 2's knowledge cutoff date is mid-2021. Knowledge about events past that date will be limited.
Supported language English
Configurable model parameters
  • Top p
  • Top k
  • Temperature
  • Stop sequence
  • Max output length
  • Number of response candidates

See the model parameters section of the Intro to LLMs guide for information about each of these parameters.

Model variations

The PaLM API offers different models optimized for specific use cases. The following table describes attributes of each.

Variation Attribute Description
text-bison-001 Model last updated May 2023
Model size Bison
Model capabilities
  • Generates text.
  • Optimized for language tasks such as:
    • Code generation
    • Text generation
    • Text editing
    • Problem solving
    • Recommendations generation
    • Information extraction
    • Data extraction or generation
    • AI agent
  • Can handle zero, one, and few-shot tasks.
Model safety Adjustable safety settings for 6 dimensions of harm available to developers. See the safety settings topic for details.
Rate limit during preview 90 requests per minute
chat-bison-001 Model last updated May 2023
Model size Bison
Model capabilities
  • Generates text in a conversational format.
  • Optimized for dialog language tasks such as implementation of chat bots or AI agents.
  • Can handle zero, one, and few-shot tasks.
Model safety No adjustable safety settings.
Rate limit during preview 90 requests per minute
embedding-gecko-001 Model last updated May 2023
Model size Gecko
Model capabilities
  • Generates text embeddings for the input text.
  • Optimized for creating embeddings for text of up to 1024 tokens.
Model safety No adjustable safety settings.
Rate limit during preview 1500 requests per minute
embedding-gecko-002 Model last updated August 2023
Model size Gecko
Model capabilities
  • Supports instruction tuning of the input text
  • Generates text embeddings for the input text.
  • Optimized for creating embeddings for text of up to 1024 tokens.
Model safety No adjustable safety settings.
Rate limit during preview 1500 requests per minute

See the prompt gallery and the examples to see the capabilities of these model variations in action.

Model sizes

The model sizes are described by an animal name. The following table shows the available sizes and what they mean relative to each other.

Model size Description Services
Bison PaLM API's most capable model size.
  • text
  • chat
Gecko PaLM API's smallest, most efficient model size.
  • embeddings

Model metadata

Use the ModelService API to get additional metadata about the latest models such as input and output token limits. The following table displays the metadata for the text-bison-001 model.

Attribute Value
Display Name Text Bison
Name models/text-bison-001
Description Model targeted for text generation
Input token limit 8196
Output token limit 1024
Supported generation methods generateText
Temperature 0.7
top_p 0.95
top_k 40

Next steps