Open wyttime04 opened 3 months ago
Hey, I ran into this using the golang SDK. I'm sure you've moved on for now, but the solution for me was to use the FieldMappings option to map the chunkID to the Filename property. Then I can get the filename property back (which is actually the chunk_id) and perform another query against Cognitive Search using a filter. It's definitely less than ideal, but shouldn't break even after this is fixed.
Hi @taigrr , the method I use is the same as yours. I mark the chunk ID in the filepath
like path/to/file - Part 1
and the total chunk number in the metadata field, which is useful when I want to delete all chunks of one file. I think this approach is acceptable for now.
Btw, I found something strange, when setting the chunk size larger, the API response chunk_id
may be greater than 0. Does this mean the chat completion API might be automatically chunked? 🤔
Describe the bug I'm using
data_preparation.py
to upload my data to AI search index. Then I chat with AI and it response fine but then citationschunk_id
is always"0"
. How can I specify this citationschunk_id
?To Reproduce Steps to reproduce the behavior:
Prepare data with scripts/readme.md
Add
chunk_id
field indata_preparation.py
when create search indexAdd
chunk_id
field when chunk fileRun
data_preparation.py
withconfig.json
AI search result can see the
chunk_id
fieldStart App with this index
AI response message citations
chunk_id
always"0"
Expected behavior
chunk_id
should response AI search fields valueScreenshots If applicable, add screenshots to help explain your problem.
Configuration: Please provide the following
gpt-35-turbo
Azure AI Search
Additional context Add any other context about the problem here.
AZURE_OPENAI_PREVIEW_API_VERSION
: I have tried2024-02-15-preview
and2024-02-01