Skip to content

Conversation

@HamedMP
Copy link
Owner

@HamedMP HamedMP commented Aug 20, 2024

Use any of the OpenRouter models with CursorLens, including claude-3.5-sonnet (by default Cursor doesn't support it when it's through OpenRouter)

image
image
image

"phi3:3.8b",
"phi3:14b",
],
openrouter: [
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Will need to refactor this after my PR goes in.

#5

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also means pricing metrics for these open router models will not be supported

Copy link
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, let's fix that, how we can do it?

Copy link
Contributor

@gruckion gruckion Aug 20, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

TLDR: option 3.

Option 1: we duplicate the entire object body, for each model configuration, onto the model that is available as a proxy via open router.

  • pro: no further refactors from my PR
  • con: repeated config - only if the provider model and proxied model identical in price.

Option 2: we instead put an attribute on the object body of models that have availability via proxies

  • pro: avoids repeated config
  • con: separation of concerns if the configs do differ in price between provider and proxies model

Option 3: we split the provider / model list config away from the config details and instead reference the config in the original provider / model and repeat it in the proxies (open router).

  • pro: can assign the default model config directly to the source providers offering. Can { ...spread, override: "specific values" } like price if they different slightly or keep identical default model config if the same
  • con: requires further change to our model config layout (not an issue)

Conclusion: we go with option 3. Easier to maintain. But the pricing appears to be the same for both providers / proxies offering.

image

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OpenAI pricing is same too. No additional markup for using it via open router (OR) proxy.

And other models like Nous Hermes don't offer their own hosted service. So those prices are what OR sets.

There are other proxies offering a similar service to OR like AI/ML API

https://aimlapi.com/models/nous-hermes-2-mistral-dpo-7b?43d56c14_page=2&6c9afc24_page=19

Their prices are higher than OR so I see no point in considering them.

@tm17-abcgen
Copy link

tm17-abcgen commented Aug 20, 2024

Already wrote it on X/Twitter but testing it worked 👍 Could use sonnet through OpenRouter. Great job man :)
Weirdly, tried to add Deepseek-Coder through openrouter, and that one did not work.

@HamedMP
Copy link
Owner Author

HamedMP commented Aug 20, 2024

Weirdly, tried to add Deepseek-Coder through openrouter, and that one did not work.

oh that’s interesting! I’ll debug and see what’s wrong. I’d love to have Deepseek supported

@tm17-abcgen
Copy link

tm17-abcgen commented Aug 20, 2024

Weirdly, tried to add Deepseek-Coder through openrouter, and that one did not work.

oh that’s interesting! I’ll debug and see what’s wrong. I’d love to have Deepseek supported

Yeah would love to use Deepseek-Coder than Sonnet 3.5 due to a similar performance but really 20x cheaper.

Edit: Thats on OpenRouter's side. The model can be used when you allow them to use your input data for model training.

@HamedMP
Copy link
Owner Author

HamedMP commented Aug 20, 2024

Thats on OpenRouter's side. The model can be used when you allow them to use your input data for model training.

so it’s fixed now? @tm17-abcgen

@tm17-abcgen
Copy link

Thats on OpenRouter's side. The model can be used when you allow them to use your input data for model training.

so it’s fixed now? @tm17-abcgen

yes

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants