Supported LLMs
AI Studio supports a variety of LLMs, enabling you to choose the right model for a given task. Whenever possible we provision redundant resources spanning different cloud regions and providers to maximize reliability. For an overview of how AI studio interacts with LLMs please review the section on Prompt AI Behaviors.
In order to insure proper resource allocation, some LLMs require you to contact Quiq Support to be enabled.
LLM | Description | External API Reference | Example Functions | Contact Support? |
---|---|---|---|---|
gpt-35-turbo-0613 | ChatCompletion style built by OpenAI, available through both OpenAI and Azure cloud offerings. Fast given the level of accuracy. - completion maps to /choices[0]["message"]["content"] in the response- 16k model variant used automatically when necessary | Link | Link | No |
gpt-35-turbo-1106 | ChatCompletion style built by OpenAI, available through both OpenAI and Azure cloud offerings. Fast given the level of accuracy. - completion maps to /choices[0]["message"]["content"] in the response | Link | Link | No |
gpt-35-turbo-0125 | ChatCompletion style built by OpenAI, available through both OpenAI and Azure cloud offerings. Fast given the level of accuracy. - completion maps to /choices[0]["message"]["content"] in the response | Link | Link | No |
gpt-4-1106 | ChatCompletion style built by OpenAI, available through both OpenAI and Azure cloud offerings. Slower but generally more accurate than 35-turbo models. - completion maps to /choices[0]["message"]["content"] in the response | Link | Link | Yes |
gpt-4-vision-preview | ChatCompletion style that supports answering questions about images. Built by OpenAI and available through both OpenAI and Azure cloud offerings. Slower but generally more accurate than 35-turbo models - completion maps to /choices[0]["message"]["content"] in the response | Link | Link | Yes |
llama3-8b-instruct-v1 | The 8B parameter variant of the Llama 3 Chat family of LLMs built by Meta. Hosted on AWS Bedrock. - completion maps to /generation in the response | Link | Link | Yes |
llama3-70b-instruct-v1 | The 70B parameter variant of the Llama 3 Chat family of LLMs built by Meta. Hosted on AWS Bedrock. - completion maps to /generation in the response | Link | Link | Yes |
claude-3-haiku | Claude V3 Haiku model by Anthropic - hosted on AWS Bedrock. - completion maps to /content/text in the response | Link | Link | No |
claude-3-sonnet | Claude V3 Sonnet model by Anthropic - hosted on AWS Bedrock. - completion maps to /content/text in the response | Link | Link | Yes |
Notes
- When building prompt payloads in the Function Editor you can assume that the
request
object should generally mimic the request payload of the underlying LLM according to its docs, but the framework may override certain parameters. See the example functions - Some LLM request payloads specify the model in the payload. AI studio will implicitly set/override these parts of your prompt objects
- When an LLM is invoked on behalf of a resource transformation we may override settings such as temperature to produce the most deterministic result
We continually uptake new LLMs and have preliminary support for additional models not listed in this documentation. Contact Quiq support or your account rep if you have specific LLM requirements
Custom LLMs
Custom LLMs enable you to communicate with your own LLM resources rather than Quiq's managed resources. For example, if you've created your own fine-tuned ChatCompletion model hosted by OpenAI, you can configure AI studio to use your fine-tuned resource. Configuring a Custom LLM requires the following information:
- The URL where the LLM is hosted
- Appropriate credentials enabling AI Studio to securely access the LLM
- The family/style of LLM resource (for purposes of matching the API contract)
In order to register a Custom LLM, contact Quiq Support.
Embeddings Models
The following table lists the supported models for embedding your AI Resources for purposes of semantic search.
Model | Description | Contact Support? |
---|---|---|
text-ada-002-embedding | Built by OpenAI, available through both OpenAI and Azure cloud offerings | No |
titan-embed-text-v1 | Built by Amazon and and available through AWS | Yes |
text-embedding-3-small | Built by OpenAI, available through both OpenAI and Azure cloud offerings | Yes |
text-embedding-3-large | Built by OpenAI, available through both OpenAI and Azure cloud offerings | Yes |
Updated 4 months ago