feat(model-providers): 16 Add Gemini Completion and Embedding Models #56
Conversation
mateobelanger
changed the title
mateobelanger
added
feat
model
…dd embeddings example
There was a problem hiding this comment.
Overall approved, but there's a consequence I wrote about related to parsing the additional_parameters into an explicit requests Model that is significant to think about. You might want to collect additional keys into it's own field and then merge that onto the reuqest after the fact (might be good to do this for the other providers in a later PR).
| for (key, value) in obj.iter().filter(|(_, v)| !v.is_null()) { | ||
| match key.as_str() { | ||
| "temperature" => { | ||
| if !value.is_null() { |
There was a problem hiding this comment.
don't you already check whether value is is null at line 545?
| } | ||
| } | ||
|
|
||
| impl TryFrom<serde_json::Value> for GenerationConfig { |
There was a problem hiding this comment.
You may not need this TryFrom implementation. Instead of doing GenerationConfig::try_from(completion_request.additional_params.unwrap_or_default())?;, you can do serde_json::from_value::<GenerationConfig>(completion_request.additional_params.unwrap_or_default())?. Since GenerationConfig has Option for each field, if the additional params don't contain the generation_config field, it won't fail.
This won't work though if field X of completion_request.additional_params is a string but field X of GenerationConfig expects a float for example.
There was a problem hiding this comment.
@mateobelanger is there a reason why you didn't change this?
…-embedding-models
…6-add-gemini-completion-embedding-models
mateobelanger
force-pushed
the
feat/model-provider/16-add-gemini-completion-embedding-models
branch
from
0e33fd8 to
3fb10d6
Compare
|
@garance-buricatu can you make sure your comments were addressed and approve? Would like to merge this before eod, thanks! |
🎯 Description & Goal
This pull request adds support for the Google Gemini LLM provider.
EmbeddingModelandCompletionModeltraits for Google Gemini APIDOCS Updates:
README.md: Added Google Gemini to the list of supported LLM providers. [1] [2]rig-core/README.md: Updated to include Google Gemini in the list of LLM providers. [1] [2]rig-core/examples/gemini_agent.rs: Added an example demonstrating how to use the Google Gemini client to create an agent and prompt it.Gemini Models Integration:
rig-core/src/providers/gemini/client.rs: Implemented the Google Gemini client with methods for creating embedding models, completion models, agents, and extractors.rig-core/src/providers/gemini/completion.rs: Added support for Google Gemini completion models and defined the necessary request and response structures.rig-core/src/providers/gemini/embedding.rs: Implemented the embedding model for Google Gemini, including methods for embedding documents.Note
async fn completion(). In OpenAI we add the preamble but not with Anthropic