feat: add dedicated Mistral AI provider with latest model variants#118
feat: add dedicated Mistral AI provider with latest model variants#118ThomsenDrake wants to merge 4 commits into
Conversation
ThomsenDrake
requested review from
raphamorim and
tauraamui
and removed request for
a team
|
Aren't |
devstral-medium-latest is the endpoint for Devstral 2 and devstral-small-latest is the endpoint for Devstral Small 2 |
|
would love to see this merged! |
|
@ThomsenDrake thanks for the PR, it looks like this pr needs a rebase and also have you made sure that the changes work as expected in crush? You can test it locally by running the catwalk server This will help me validate the configuration too :) |
- Added Mistral AI provider configuration with 32 models - Extracted all Mistral models from OpenRouter config - Added -latest variants for all main model sizes (large, medium, small) - Added -latest variants for devstral models (medium, small) - Updated providers.go to include mistralProvider - Set API endpoint to https://api.mistral.ai/v1 - Uses tILXtdeWUWgbTX3Uwg0Flp4SVLEahdyH environment variable Generated by Mistral Vibe. Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
- Updated Frontier Models - Generalist pricing: - Mistral Large 3: /bin/sh.5/M input, .5/M output - Mistral Medium 3.1: /bin/sh.4/M input, /M output - Mistral Small 3.2: /bin/sh.1/M input, /bin/sh.3/M output - Ministral 3 14B: /bin/sh.2/M input, /bin/sh.2/M output - Ministral 3 8B: /bin/sh.15/M input, /bin/sh.15/M output - Ministral 3 3B: /bin/sh.1/M input, /bin/sh.1/M output - Updated Other Models pricing: - Devstral Medium 1.0: /bin/sh.4/M input, /M output - Devstral Small 1.1: /bin/sh.1/M input, /bin/sh.3/M output - Mistral Medium 3: /bin/sh.4/M input, /M output - Mistral Large 2.1: /M input, /M output (already correct) - Pixtral Large: /M input, /M output (already correct) - Mistral Nemo 12B: /bin/sh.15/M input, /bin/sh.15/M output (already correct) - Codestral: /bin/sh.3/M input, /bin/sh.9/M output - Updated -latest variants to match their corresponding models Generated by Mistral Vibe. Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
- Changed default_large_model_id from mistral-large to devstral-medium-latest - Changed default_small_model_id from mistral-small to devstral-small-latest - This makes Devstral models the default choice when using the Mistral provider Generated by Mistral Vibe. Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
- Remove redundant "Mistral: " prefix from non-flagship model names - Add TypeMistral to known provider types - Add InferenceProviderMistral to known inference providers - Register Mistral in KnownProviders() and KnownProviderTypes()
ThomsenDrake
force-pushed
the
add-mistral-provider
branch
from
cb72877 to
1d1c545
Compare
|
Thanks for the review! I've rebased on the latest main and tested locally with crush. Testing results:
One issue discovered: When testing in crush with This would need a separate fix in the crush repo to add Mistral API key format validation. The catwalk configuration itself is working correctly. |
|
@kujtimiihoxha did you had time to review the changes here? For me i'm getting this error when starting crush with this PR catwalk server running: > CATWALK_URL=http://localhost:8080/ crush
ERROR
Failed to initialize coder agent: provider type not supported: "mistral". |
|
I big fan of Mistral as well. Would love to see this merged. What is the current status? |
| "name": "Mistral AI", | ||
| "id": "mistral", | ||
| "api_key": "$MISTRAL_API_KEY", | ||
| "api_endpoint": "https://api.mistral.ai/v1", |
There was a problem hiding this comment.
Looks like there is a /models endpoint available:
https://docs.mistral.ai/api/endpoint/models#operation-list_models_v1_models_get
In those cases, we prefer to have a generator to fetch and update the list for us.
AIs are usually good do create one taking the others as examples. They are all under cmd/.
There was a problem hiding this comment.
Hey @andreynering, I just took at stab at it here: #332
|
Closing in favor of: |
7 participants
Generated by Mistral Vibe.