-
Notifications
You must be signed in to change notification settings - Fork 1.6k
Description
Please describe the feature you want
Mistral AI made an embedding API for codestral-embed
(actually they made another one for mistral-embed
, but I don't expect it to be any relevant compared to codestral-embed
for this project). This contradicts the tabby docs if I'm not mistaken.
Would it be possible to support a new codestral/embedding
(or mistral/embedding
) API kind for this?
Additional context
I stumbled on this while trying to configure codestral-embed
in tabby.
Here is my current (failing) config for this model:
[model.embedding.http]
kind = "openai/embedding"
model_name = "codestral-embed"
api_endpoint = "https://api.mistral.ai/v1"
api_key = "my-key"
Here, I have been using the openai/embedding
API although it is not exectly the one expected by the model. Upon connection, I obtain this error:
Error 422: {"detail":[{"type":"model_attributes_type","loc":["body"],"msg":"Input should be a valid dictionary or object to extract fields from","input":"{\"input\":[\"hello Tabby\"],\"model\":\"codestral-embed\"}"}]}
Besides, this works perfectly fine in my terminal:
curl -X POST "https://api.mistral.ai/v1/embeddings" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $MISTRAL_API_KEY" \
-d "{\"input\": [\"hello Tabby\"], \"model\": \"codestral-embed\"}"
Thank you in advance!
Please reply with a 👍 if you want this feature.