https://www.boundaryml.com logo
Wondering if anyone is sharing my pain
# general
j
Wondering if anyone is sharing my pain I have a product that uses BAML and we support all the LLM Providers - but need to tell our end users how to configure their environment variables accordingly. 1 major sticky point is what model-id to use .. For our solution we required Image + Text input capabilties I have a list across all the providers I have compiled - 85 models.. (not fully exhaustive either) .. It seems most providers normally have the same models, but use differnt model id's .. (and slight cost differences sometimes). Im really struggling trying to tell this story to the end user in our docs (and give them a list of models they could use) > And keeping up to date across all the providers is a fun problem space (I'd love to build an agent but each provider has their own docs, and pricing are often listed elsewhere.. sometimes model capabilties (context window, input modalities) are not listed on the providers docs.. so extra research is needed too Really looking for any advice or ideas.
Partly tempted to just go build this agent - but curious if anyone knows of a system or datasource like this already? Been manually looking at each provider whenever news a new big model is released Supported Providers: Azure OpenAI AWS Bedrcok Gemini Google Vertex Anthropic OpenAI We support any openai-generic stack as well.. pandoras box over there - but majority of customers are gonna be using a provider above anyways (or adjecent OpenRouter/Together style openai inference that proxies simialrly)
n
OpenRouter?
It has an api for pooling metadata on model-providers. Since you are already using it, you should be able to get all information on how to set up a scrape and find all info you may need
j
If the issue is model identifier management, check out https://models.dev/
And +1 to OpenRouter as well I believe they do some thin schema normalization
n
this is nice! OpenRouter also gives you options for quantization from each provider.
j
Very nice! Thank you guys!! I found the LiteLLM has a pricing API I was gonna go explore how they do that too Model.dev looks nearly identical to my sheet too