phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-06-29 11:24:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

sqlite-vec support for Responses file_search

Browse source 
This wires up the Files API optional dependency into sqlite_vec and
adds the localfs Files provider to our starter template, so that
Responses API file_search tool works out of the box for sqlite_vec in
that template.

Some additional testing with this provider plus some other inference
models led me to loosen the verification test results checking a bit -
not for the tool call, but just around the assistant response with the
file_search tool call. Some providers, such as OpenAI SaaS, make
multiple tool calls to resolve the query sometimes, especially when it
cannot find an answer so tries a few permutations before returning
empty results to the user in that test.

Signed-off-by: Ben Browning <bbrownin@redhat.com>
This commit is contained in:
Ben Browning 2025-06-13 14:01:34 -04:00
parent ec09524a91
commit 7a71d9ebd8
8 changed files with 30 additions and 24 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama_stack/providers/registry/vector_io.py
View file

@ -44,6 +44,7 @@ def available_providers() -> list[ProviderSpec]:
module="llama_stack.providers.inline.vector_io.sqlite_vec",
config_class="llama_stack.providers.inline.vector_io.sqlite_vec.SQLiteVectorIOConfig",
api_dependencies=[Api.inference],
optional_api_dependencies=[Api.files],
),
InlineProviderSpec(
api=Api.vector_io,
@ -53,6 +54,7 @@ def available_providers() -> list[ProviderSpec]:
config_class="llama_stack.providers.inline.vector_io.sqlite_vec.SQLiteVectorIOConfig",
deprecation_warning="Please use the `inline::sqlite-vec` provider (notice the hyphen instead of underscore) instead.",
api_dependencies=[Api.inference],
optional_api_dependencies=[Api.files],
),
remote_provider_spec(
Api.vector_io,