feat(py): Python GoogleAi plugin #2173
Conversation
github-actions
bot
added
docs
kamilkorski
force-pushed
the
korski/py-googleai-plugin
branch
from
e9f4018 to
fbdee64
Compare
kamilkorski
requested review from
yesudeep,
pavelgj,
kirgrim and
timere-nemo
There was a problem hiding this comment.
Please also introduce some tests verifying it works structure-wise, you can copy and adapt tests from https://github.com/firebase/genkit/pull/1828/files#diff-285168c0aaeff7b2a46f86c5cfd05113fdd9f050e75d40d9c4c9c2b9a085561eR12
| return f'googleai/{name}' | ||
|
|
||
|
|
||
| class GoogleAiPluginOptions(pydantic.BaseModel): |
There was a problem hiding this comment.
please move schemas to the separate file, e.g. models.py
|
|
||
|
|
||
| class GoogleAi(plugin_abc.Plugin): | ||
| def __init__(self, plugin_params=GoogleAiPluginOptions): |
There was a problem hiding this comment.
annotate it as GoogleAiPluginOptions | None and assign to None by default -> if its None set it to GoogleAiPluginOptions. This will be the safest option to handle the default behaviour:
def __init__(self, plugin_params: GoogleAiPluginOptions | None = None): self.plugin_params = plugin_params or GoogleAiPluginOptions
| def _create_callback( | ||
| self, model: str | ||
| ) -> Callable[[GenerateRequest], GenerateResponse]: | ||
| def model_callback(request: GenerateRequest) -> GenerateResponse: |
There was a problem hiding this comment.
we are currently approaching defining callbacks via dedicated Model handler that create in initialise, I believe that will make code more readable and easier to maintain, example: https://github.com/firebase/genkit/pull/1828/files#diff-e147b482a7fce3760b369ebb507f3d27e8b6f5d44629772c295fcf016eb218f6R114
There was a problem hiding this comment.
also consider handling streaming API by passing ctx: ActionRunContext | None = None parameter and calling send_chunk explicitly
| reqest_msgs.append( | ||
| genai.types.Content(parts=message_parts, role=msg.role) | ||
| ) | ||
| response = self._client.models.generate_content( |
There was a problem hiding this comment.
as we are aiming at async-first solution, is it possible to use this client SDK in an async version?
Created google ai plugin
Checklist (if applicable):