codestral-embed: Fails with 3072 dimensions via OpenAI-compatible provider

[anojndr]

App Version

3.23.19

API Provider

OpenAI Compatible

Model Used

codestral-embed

🔁 Steps to Reproduce

1. Configure the OpenAI-compatible provider to use the codestral-embed model.
2. Set the embedding dimensions to 3072 in the settings.
3. Attempt to index or query for embeddings.
4. The process works correctly if the dimension is set to 1536.

💥 Outcome Summary

Expected: Embeddings should be generated with a dimension of 3072, as supported by the codestral-embed model.
Actual: The process fails. Qdrant logs show a dimension mismatch error (expected dim: 3072, got 1536), and a separate error is displayed in the UI.

📄 Relevant Logs or Errors

Qdrant Log:

2025-07-21T12:11:29.109031Z WARN collection::collection_manager::collection_updater: update operation declined: wrong input: vector dimension error: expected dim: 3072, got 1536

UI Error:

Error - Failed during initial scan: Indexing failed: Failed to process batch after 3 attempts: Bad Request