HN Explorer

Ask HN: What are some good APIs to interact with large language models?

19 points by yangikan ↗ HN
I am looking to do simple things like image classification/text classification using APIs without running the LLMs in my local machine. What are some APIs that provide a uniform interface to access different LLMs?

12 comments

[ 3.2 ms ] story [ 30.0 ms ] thread
logankeenan
I used runpod.io prior to buying a pair of 3090s. They make it easy to run vllm too so you can experiment with different models.

I also rented a GPU vm from them and ran huggingface models on it. That did require lot more coding and learning.

https://docs.runpod.io/serverless/workers/vllm/get-started

Tostino
As far as what 'API' to use, please just coalesce around the OpenAI API for your client software. You can start up an OpenAI compatible endpoint with vLLM for example. Just stick with that. You can use LiteLLM as a proxy to convert your client side requests to whatever server side format is expected for e.g. Claude.
anonzzzies
Openroute.ai works very well; we use it for everything.
anonzzzies
+r Sorry, didn't see till now ; openrouteR.ai
emmanueloga_
Going through middle layers has pros and cons. You'll likely find better docs and experience by going through the official APIs directly. Maybe start with OpenAI and/or Claude official libraries, both provide pretty good docs.

If you really must use a wrapper, a strategy I use is to look into open-source apps I like and see what they are using. Aider for instance seems to be using this "litellm" thing... If I were to need cross API support for AI, I would probably look into that [1]. llm uses opeanai directly [2]. Etc, etc.

--

1: https://github.com/Aider-AI/aider/blob/e76704e261647348fd7c1...

2: https://github.com/simonw/llm/blob/d654c9521235a737e59a4f1d7...

Gooblebrai
Replicate.com? Has a good variety of models
anon1094
I'm using Lemonfox for barebones speech-to-text translation.