[BUG] Index uses 2048-dimensional vectors, but Qdrant logs report receiving 1024

[qq1792569310]

Problem (one or two sentences)

I configured 2048 dimensions on roocode, but I don't know why Qdrant reports receiving 1024 dimensions.

Context (who is affected and when)

I use Alibaba Cloud's text-embedding-v4 model(base url:https://dashscope.aliyuncs.com/compatible-mode/v1), the model dimension is set to 2048, Qdrant uses the free product provided by qdrant.io. Indexing failed twice in a row, with dimension mismatch logs appearing.

I now manually set the dimension to 1024 on roocode, and indexing succeeded on the first try.

Reproduction steps

1、Configure Alibaba Cloud's text-embedding-v4 model, and set the dimension to 2048 on roocode
2、Configure qdrant.io's free product
3、Start indexing

Expected result

Indexing succeeds

Actual result

Indexing fails

Variations tried (optional)

No response

App Version

3.41.3

API Provider (optional)

None

Model Used (optional)

N/A

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I've analyzed this issue and identified the root cause:

Root Cause:
The configured dimensions parameter (2048 in your case) is correctly used when creating the Qdrant collection, but it's not being passed to the embedding API request. Some embedding models (like Alibaba Cloud's text-embedding-v4) support a dimensions parameter in the API to request embeddings of a specific size. Without passing this parameter, the API returns the model's default dimension (1024 for text-embedding-v4) instead of your configured 2048.

Technical Details:

1. config-manager.ts correctly loads codebaseIndexEmbedderModelDimension from settings
2. service-factory.ts correctly passes the dimension when creating the Qdrant collection
3. BUG: service-factory.ts does NOT pass the dimension to OpenAICompatibleEmbedder
4. BUG: openai-compatible.ts doesn't include dimensions in the embedding API request

Proposed Implementation Plan:

1. Modify OpenAICompatibleEmbedder constructor (src/services/code-index/embedders/openai-compatible.ts):

   • Add an optional dimensions parameter to the constructor
   • Store it as an instance variable

2. Update embedding API calls in openai-compatible.ts:

   • Add dimensions parameter to the SDK call (embeddingsClient.embeddings.create)
   • Add dimensions parameter to the direct HTTP request (makeDirectEmbeddingRequest)

3. Update service-factory.ts:

   • Pass config.modelDimension when creating the OpenAICompatibleEmbedder

4. Update tests to cover the new dimension parameter

This approach will ensure that when you configure 2048 dimensions, the embedding API will return 2048-dimensional vectors that match your Qdrant collection configuration.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[qq1792569310]

@roomote I checked the documentation, and text-embedding-v4's default dimension is indeed 1024. Consider whether adding the dimensions parameter will have a negative impact on other models.