feat: Add NVIDIA as a Provider and expose it to Goose #132
Conversation
Signed-off-by: Praateek <praateekm@gmail.com>
Signed-off-by: Praateek <praateekm@gmail.com>
Signed-off-by: Praateek <praateekm@gmail.com>
Signed-off-by: Praateek <praateekm@gmail.com>
praateekmahajan
changed the title
| @@ -19,6 +19,8 @@ packages = [{ include = "goose", from = "src" }] | |||
| [tool.hatch.build.targets.wheel] | |||
| packages = ["src/goose"] | |||
|
|
|||
|
|
|||
There was a problem hiding this comment.
Was this an accidental change?
src/goose/cli/config.py
Outdated
| } | ||
| processor, accelerator = recommended.get(provider, ("gpt-4o", "gpt-4o-mini")) | ||
| processor, accelerator = recommended.get(provider, recommended["nvidia"]) |
There was a problem hiding this comment.
I think this line should be reverted. Is there a reason to change the default values to nvidia's values?
Your change on line 105 should work to set the default configuration to nvidia when it detects NVIDIA_API_KEY.
| self.client = client | ||
|
|
||
| @classmethod | ||
| def from_env(cls: Type["NVIDIAProvider"]) -> "NVIDIAProvider": |
There was a problem hiding this comment.
this works! however @elenazherdeva recently introduced cls.check_env_vars() pattern so these can be checked upfront so we can avoid the need for a try/except block
|
@praateekmahajan you ok if I take this on and finish it up to get it in? |
|
@michaelneale hey sorry for the late response, I had marked it as draft since it was a POC, and I was trying to get it to work with LLAMA 3.1 but wasn't able to. If you've been able to that's great. And thanks for taking it over, I haven't had the bandwidth to look at this and make it work yet. |
michaelneale
added
enhancement
|
@praateekmahajan no worries - yes at some point we will iterate a bit on the newer llamas (main thing was tool calling was not great generally yet - but will get better) |
Creates a provider in
exchange.In goose, adds a config for the NVIDIA provider (using
llama-3.1-405b) todefault_model_configuration