Dieser Browser wird nicht mehr unterstützt.
Führen Sie ein Upgrade auf Microsoft Edge aus, um die neuesten Funktionen, Sicherheitsupdates und technischen Support zu nutzen.
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(268.8, 63%, 36%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EAG%3C/text%3E%3C/svg%3E)
We're switching from OpenAI to Azure OpenAI which works fine so far. We're using Assitants with "File search".
Unfortunately the settings for chunk sizes (and overlaps) are being ignored creating vector stores. A different size of chunks, other than the default 800 and 400 for overlaps, is absolutely crucial to our process.
Is this feature not (yet) available? Must parameters be set differently, than this?:
chunking_strategy: { type: "static", static: { max_chunk_size_tokens: 2100, chunk_overlap_tokens: 1050 } }
Might there be an alternative for use with Assistants, if this feature is not available?
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(272, 65%, 36%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3ECO%3C/text%3E%3C/svg%3E)
Have you been able to fix this issue or have any workarounds? We are doing similar things and experiencing the same problem. It is also crucial to optimize our vector store to make the search usable/stable for our custom assistant.
Okay, so we figured it out today! When you create VectorStore, you need to set FileIds right away to apply the ChunkingStrategy. It doesn't work if you add the files to the already created VectorStore -> Those files will have the default "auto" strategy instead of the "static" one of the store.
Great news! I had to switch tasks, so I couldn't follow up. And also, unfortunately, even after a lengthy call with MS-support, we couldn't really figure it out.
Would you mind sharing this peace of code?
I really thought, I was doing what you suggest. This is what I do:
response = client.vector_stores.create(
parameters: {
name: "CV-Vector store for file #{run_data.filename}",
file_ids: [run_data.file_id],
chunking_strategy: { type: "static", static: { max_chunk_size_tokens: 3000, chunk_overlap_tokens: 1500 } }
}
)
Which boils down to a request to "/vector_stores" with those fields.