Skip to main content

Generative Search - PaLM

LICENSEย Weaviate on Stackoverflow badgeย Weaviate issues on GitHub badgeย Weaviate version badgeย Weaviate total Docker pulls badgeย Go Report Card

In shortโ€‹

  • The Generative PaLM (generative-palm) module generates responses based on the data stored in your Weaviate instance.
  • The module can generate a response for each returned object, or a single response for a group of objects.
  • The module adds a generate {} parameter to the GraphQL _additional {} property of the Get {} queries.
  • Added in Weaviate v1.19.1.
  • You need an API key for a PaLM API to use this module.
  • The default model is chat-bison.

Introductionโ€‹

generative-palm generates responses based on the data stored in your Weaviate instance.

The module works in two steps:

  1. (Weaviate) Run a search query in Weaviate to find relevant objects.
  2. (PaLM) Use a PaLM model to generate a response based on the results (from the previous step) and the provided prompt or task.
note

You can use the Generative PaLM module with non-PaLM upstream modules. For example, you could use text2vec-openai, text2vec-cohere or text2vec-huggingface to vectorize and query your data, but then rely on the generative-palm module to generate a response.

The generative module can provide results for:

  • each returned object, using singleResult{ prompt }
  • the group of all results together, using groupedResult{ task }

You need to input both a query and a prompt (for individual responses) or a task (for all responses).

Inference API keyโ€‹

Important: Provide PaLM API key to Weaviate

As the generative-palm uses a PaLM API endpoint, you must provide a valid PaLM API key to weaviate.

For Google Cloud usersโ€‹

This is called an access token in Google Cloud.

If you have the Google Cloud CLI tool installed and set up, you can view your token by running the following command:

gcloud auth print-access-token

Providing the key to Weaviateโ€‹

You can provide your PaLM API key by providing "X-Palm-Api-Key" through the request header. If you use the Weaviate client, you can do so like this:

import weaviate

client = weaviate.Client(
url = "https://some-endpoint.weaviate.network/",
additional_headers = {
"X-Palm-Api-Key": "YOUR-PALM-API-KEY", # Replace with your API key
}
)

Optionally (not recommended), you can provide the PaLM API key as an environment variable.

How to provide the PaLM API key as an environment variable

During the configuration of your Docker instance, by adding PALM_APIKEY under environment to your docker-compose file, like this:

environment:
PALM_APIKEY: 'your-key-goes-here' # Setting this parameter is optional; you can also provide the key at runtime.
...

Token expiry for Google Cloud usersโ€‹

Important

Google Cloud's OAuth 2.0 access tokens are configured to have a standard lifetime of 1 hour.

Therefore, you must periodically replace the token with a valid one and supply it to Weaviate by re-instantiating the client with the new key.

You can do this manually.

Automating this is a complex, advanced process that is outside the scope of our control. However, here are a couple of possible options for doing so:

With Google Cloud CLI

If you are using the Google Cloud CLI, you could run this through your preferred programming language, and extract the results.


For example, you could periodically run:

client = re_instantiate_weaviate()

Where re_instantiate_weaviate is something like:

import subprocess
import weaviate

def refresh_token() -> str:
result = subprocess.run(["gcloud", "auth", "print-access-token"], capture_output=True, text=True)
if result.returncode != 0:
print(f"Error refreshing token: {result.stderr}")
return None
return result.stdout.strip()

def re_instantiate_weaviate() -> weaviate.Client:
token = refresh_token()

client = weaviate.Client(
url = "https://some-endpoint.weaviate.network", # Replace with your Weaviate URL
additional_headers = {
"X-Palm-Api-Key": token,
}
)
return client

# Run this every ~60 minutes
client = re_instantiate_weaviate()
With google-auth

Another way is through Google's own authentication library google-auth.


See the links to google-auth in Python and Node.js libraries.


You can, then, periodically the refresh function (see Python docs) to obtain a renewed token, and re-instantiate the Weaviate client.

For example, you could periodically run:

client = re_instantiate_weaviate()

Where re_instantiate_weaviate is something like:

from google.auth.transport.requests import Request
from google.oauth2.service_account import Credentials
import weaviate

def get_credentials() -> Credentials:
credentials = Credentials.from_service_account_file('path/to/your/service-account.json', scopes=['openid'])
request = Request()
credentials.refresh(request)
return credentials

def re_instantiate_weaviate() -> weaviate.Client:
credentials = get_credentials()
token = credentials.token

client = weaviate.Client(
url = "https://some-endpoint.weaviate.network", # Replace with your Weaviate URL
additional_headers = {
"X-Palm-Api-Key": token,
}
)
return client

# Run this every ~60 minutes
client = re_instantiate_weaviate()

The service account key shown above can be generated by following this guide.

Module configurationโ€‹

Not applicable to WCS

This module is enabled and pre-configured on Weaviate Cloud Services.

Configuration file (Weaviate open source only)โ€‹

You can enable the Generative Palm module in your configuration file (e.g. docker-compose.yaml). Add the generative-palm module (alongside any other module you may need) to the ENABLE_MODULES property, like this:

ENABLE_MODULES: 'text2vec-palm,generative-palm'
See a full example of a Docker configuration with generative-palm

Here is a full example of a Docker configuration, which uses the generative-palm module in combination with text2vec-palm, and provides the API key:

---
version: '3.4'
services:
weaviate:
command:
- --host
- 0.0.0.0
- --port
- '8080'
- --scheme
- http
image:
semitechnologies/weaviate:1.19.6
ports:
- 8080:8080
restart: on-failure:0
environment:
QUERY_DEFAULTS_LIMIT: 25
AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: 'true'
PERSISTENCE_DATA_PATH: '/var/lib/weaviate'
DEFAULT_VECTORIZER_MODULE: 'text2vec-palm'
ENABLE_MODULES: 'text2vec-palm,generative-palm'
PALM_APIKEY: sk-foobar # Setting this parameter is optional; you can also provide the key at runtime.
CLUSTER_HOSTNAME: 'node1'

Schema configurationโ€‹

You can define settings for this module in the schema, including the API endpoint and project information, as well as optional model parameters.

Note that the projectId parameter is required.

Example schemaโ€‹

For example, the following schema configuration will set the PaLM API information, as well as the optional parameters.

  • The "projectId" is REQUIRED, and may be something like "cloud-large-language-models"
  • The "apiEndpoint" is optional, and may be something like: "us-central1-aiplatform.googleapis.com", and
  • The "modelId" is optional, and may be something like "chat-bison".
{
"classes": [
{
"class": "Document",
"description": "A class called document",
...,
"moduleConfig": {
"generative-palm": {
"projectId": "YOUR-GOOGLE-CLOUD-PROJECT-ID", // Required. Replace with your value: (e.g. "cloud-large-language-models")
"apiEndpoint": "YOUR-API-ENDPOINT", // Optional. Defaults to "us-central1-aiplatform.googleapis.
"modelId": "YOUR-GOOGLE-CLOUD-ENDPOINT-ID", // Optional. Defaults to "chat-bison"
"temperature": 0.2, // Optional
"maxOutputTokens": 512, // Optional
"topK": 3, // Optional
"topP": 0.95, // Optional
}
}
}
]
}

See the relevant PaLM API documentation for further details on these parameters.

New to Weaviate Schemas?

If you are new to Weaviate, check out the Weaviate schema tutorial.

How to useโ€‹

This module extends the _additional {...} property with a generate operator.

generate takes the following arguments:

FieldData TypeRequiredExampleDescription
singleResult {prompt}stringnoSummarize the following in a tweet: {summary}Generates a response for each individual search result. You need to include at least one result field in the prompt, between braces.
groupedResult {task}stringnoExplain why these results are similar to each otherGenerates a single response for all search results

Example of properties in the promptโ€‹

When piping the results to the prompt, at least one field returned by the query must be added to the prompt. If you don't add any fields, Weaviate will throw an error.

For example, assume your schema looks like this:

{
Article {
title
summary
}
}

You can add both title and summary to the prompt by enclosing them in curly brackets:

{
Get {
Article {
title
summary
_additional {
generate(
singleResult: {
prompt: """
Summarize the following in a tweet:

{title} - {summary}
"""
}
) {
singleResult
error
}
}
}
}
}

Example - single resultโ€‹

Here is an example of a query where:

  • we run a vector search (with nearText) to find articles about "Italian food"
  • then we ask the generator module to describe each result as a Facebook ad.
    • the query asks for the summary field, which it then includes in the prompt argument of the generate operator.
{
Get {
Article(
nearText: {
concepts: ["Italian food"]
}
limit: 1
) {
title
summary
_additional {
generate(
singleResult: {
prompt: """
Describe the following as a Facebook Ad: {summary}
"""
}
) {
singleResult
error
}
}
}
}
}

Example response - single resultโ€‹

{
"data": {
"Get": {
"Article": [
{
"_additional": {
"generate": {
"error": null,
"singleResult": "This Facebook Ad will explore the fascinating history of Italian food and how it has evolved over time. Learn from Dr Eva Del Soldato and Diego Zancani, two experts in Italian food history, about how even the emoji for pasta isn't just pasta -- it's a steaming plate of spaghetti heaped with tomato sauce on top. Discover how Italy's complex history has shaped the Italian food we know and love today."
}
},
"summary": "Even the emoji for pasta isn't just pasta -- it's a steaming plate of spaghetti heaped with tomato sauce on top. But while today we think of tomatoes as inextricably linked to Italian food, that hasn't always been the case. \"People tend to think Italian food was always as it is now -- that Dante was eating pizza,\" says Dr Eva Del Soldato , associate professor of romance languages at the University of Pennsylvania, who leads courses on Italian food history. In fact, she says, Italy's complex history -- it wasn't unified until 1861 -- means that what we think of Italian food is, for the most part, a relatively modern concept. Diego Zancani, emeritus professor of medieval and modern languages at Oxford University and author of \"How We Fell in Love with Italian Food,\" agrees.",
"title": "How this fruit became the star of Italian cooking"
}
]
}
}
}

Example - grouped resultโ€‹

Here is an example of a query where:

  • we run a vector search (with nearText) to find publications about finance,
  • then we ask the generator module to explain why these articles are about finance.
{
Get {
Publication(
nearText: {
concepts: ["magazine or newspaper about finance"]
certainty: 0.75
}
) {
name
_additional {
generate(
groupedResult: {
task: "Explain why these magazines or newspapers are about finance"
}
) {
groupedResult
error
}
}
}
}
}

Example response - grouped resultโ€‹

{
"data": {
"Get": {
"Publication": [
{
"_additional": {
"generate": {
"error": null,
"groupedResult": "The Financial Times, Wall Street Journal, and The New York Times Company are all about finance because they provide news and analysis on the latest financial markets, economic trends, and business developments. They also provide advice and commentary on personal finance, investments, and other financial topics."
}
},
"name": "Financial Times"
},
{
"_additional": {
"generate": null
},
"name": "Wall Street Journal"
},
{
"_additional": {
"generate": null
},
"name": "The New York Times Company"
}
]
}
}
}

Additional informationโ€‹

Supported modelsโ€‹

The chat-bison model is used by default. The model has the following properties:

  • Max input token: 8,192
  • Max output tokens: 1,024
  • Training data: Up to Feb 2023

More resourcesโ€‹

If you can't find the answer to your question here, please look at the:

  1. Frequently Asked Questions. Or,
  2. Knowledge base of old issues. Or,
  3. For questions: Stackoverflow. Or,
  4. For more involved discussion: Weaviate Community Forum. Or,
  5. We also have a Slack channel.