Vectorname in Collection must match embedding

Poppi2 commented

2025-02-12 15:36:43 +00:00

(Migrated from github.com)

In qdrant, collections' vectorname must match the name of the embedding. E.g. using sentence-transformers/all-MiniLM-L6-v2 it requires the vector name to be fast-all-MiniLM-L6-v2. There seems to be no way to change this. It makes working with existing collections impossible.
Using MCP inspector:

Tool Result: Error

Unexpected Response: 400 (Bad Request)
Raw response content:
b'{"status":{"error":"Wrong input: Vector with name fast-all-minilm-l6-v2 is not configured in this collection"},"time":0.00008094}'

It works if we create a new collection with fast-all-minilm-l6-v2 as its vector name but since collection names cant be changed after creation this mcp server is not really useable. We tested on multiple machines.
We would appreciate any help regarding this, personal efforts have not proven successful so far. Only got it to work without fast embed. Thank you.

In qdrant, collections' vectorname must match the name of the embedding. E.g. using sentence-transformers/all-MiniLM-L6-v2 it requires the vector name to be fast-all-MiniLM-L6-v2. There seems to be no way to change this. It makes working with existing collections impossible. Using MCP inspector: ``` Tool Result: Error Unexpected Response: 400 (Bad Request) Raw response content: b'{"status":{"error":"Wrong input: Vector with name fast-all-minilm-l6-v2 is not configured in this collection"},"time":0.00008094}' ``` It works if we create a new collection with fast-all-minilm-l6-v2 as its vector name but since collection names cant be changed after creation this mcp server is not really useable. We tested on multiple machines. We would appreciate any help regarding this, personal efforts have not proven successful so far. Only got it to work without fast embed. Thank you.

kacperlukawski commented

2025-02-14 11:22:11 +00:00

(Migrated from github.com)

@Poppi2 We could possibly allow the collection and vector names to be set. However, the payload structure is also important, as we assume vectors are created with FastEmbed, and it has a specific payload structure. I guess that's something we could implement, but I'd like to understand what exactly has to be done.

What is the model you use for the embeddings? Is it FastEmbed-compatible?

@Poppi2 We could possibly allow the collection and vector names to be set. However, the payload structure is also important, as we assume vectors are created with FastEmbed, and it has a specific payload structure. I guess that's something we could implement, but I'd like to understand what exactly has to be done. What is the model you use for the embeddings? Is it FastEmbed-compatible?

Poppi2 commented

2025-02-19 09:22:36 +00:00

(Migrated from github.com)

@kacperlukawski We currently use nomic-ai/nomic-embed-text-v1.5 and weren't aware that vectors needed to be upserted with FastEmbed. So thank you for that clarification. According to FastEmbeds supported models nomic v1.5 should be a viable option. If I can be of any further assistance please don't hesitate.

kacperlukawski commented

2025-02-19 11:18:28 +00:00

(Migrated from github.com)

@Poppi2 It's not strictly required, but FastEmbed creates payloads in a certain way that you would need to reflect. I'm happy to extend the server to support different embedding models, as some people might prefer models FastEmbed does not support, and that shouldn't be a

Poppi2 commented

2025-02-19 15:02:41 +00:00

(Migrated from github.com)

@kacperlukawski From a user side it's detrimental to be able to work with existing collections & embeddings that don't use FastEmbed - so I'm glad to hear it's a possibility.

kacperlukawski commented

2025-03-10 16:49:41 +00:00

(Migrated from github.com)

@Poppi2 We're now ready to extend the MCP server to accept some other providers. Which ones would you like to use? I'm happy to give them the highest priority!

Poppi2 commented

2025-03-19 09:42:05 +00:00

(Migrated from github.com)

@kacperlukawski We are locally hosting (ollama) nomic-embed-text-v1.5
Thank you for adressing the issue!

@kacperlukawski We are locally hosting (ollama) nomic-embed-text-v1.5 Thank you for adressing the issue!

francismiko commented

2025-03-21 06:17:15 +00:00

(Migrated from github.com)

Similarly, using ollama's locally hosted nomic-embed-text will result in the error Vector with name fast-nomic-embed-text-v1 is not configured in this collection 🫡

Similarly, using ollama's locally hosted nomic-embed-text will result in the error `Vector with name fast-nomic-embed-text-v1 is not configured in this collection` 🫡

Rows
Columns

Vectorname in Collection must match embedding #10