together-haystack provides Haystack components to use Together.ai models in Haystack pipelines.

At the moment, the integration is limited to embedder components that use Together Embeddings Inference APIs.

• TogetherAITextEmbedder to embed strings / text
• TogetherAIDocumentEmbedder to embed a collection of Haystack documents.

Refer to Haystack's documentation to see how these can be plugged-in in Haystack Pipelines to build whatever you need to.

Support for Generators (LLM) and other resources offered by together will be added soon™.

1. Install from pypi using your python dependency manager, e.g. pip install togetherai-haystack
2. Import and use as any haystack component:

   >>> from togetherai_haystack.components.embedders import TogetherAITextEmbedder
   
   # by default the API KEY is read from the TOGETHER_API_KEY environment variable
   >>> text_embedder = TogetherTextEmbedder(model="togethercomputer/m2-bert-80M-32k-retrieval")
   >>> text_embedder.run("Together.ai provides optimized inference endpoints")
   {'embedding': [-0.3592394, 0.1824189, ...]}
   

This project uses pdm as a package manager and ruff as a formatter and linter.

The test suite is currently quite bad, there are just two "integration" tests which require a real API key and will invoke the real Together endpoints. These means that:

• You need an account and to set up an API key to run these tests
• These tests require an internet connection and won't work if the environment can't reach https://api.together.xyz/v1 or if Together itself is nonresponsive.
• These tests will use your Together.ai account credit. Though, given their low prices and the minuscule number of tokens used in the tests, each run should cost you ~$0.0000008, meaning you'd have to run them ~12500 times before 1cent is taken from your account.