scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-08 03:08:45 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 116

test(vertex_ai): pin gemini-embedding-2 spend assertions to the bundled cost map

Browse source 
The four TestProcessEmbedContentResponseUsage cases that assert spend read
the gemini-embedding-2 rates from litellm.model_cost, which is fetched from
main at import time. main now prices that model per modality token and no
longer carries input_cost_per_image / input_cost_per_audio_per_second /
input_cost_per_video_per_second, so every embed-content cost assertion in
this class fails on this branch regardless of the code under test.

Use the existing local_model_cost_map fixture, which pins the bundled
in-repo map, exactly as the other pricing tests do.

Fixes #41224
This commit is contained in:
François Bossière 2026-09-15 14:11:52 +02:00
parent e319bf270c
commit a217303766
1 changed files with 7 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
tests/unit/llms/vertex_ai/gemini_embeddings/test_batch_embed_content_transformation.py
View file

@ -308,10 +308,17 @@ class TestProcessResponse:
)
@pytest.mark.usefixtures("local_model_cost_map")
class TestProcessEmbedContentResponseUsage:
"""Gemini Embedding 2 embedContent usageMetadata must drive spend.
Regression for multimodal calls recording prompt_tokens=0 / spend=$0.
The spend assertions read the ``gemini-embedding-2`` rates, so they must be
pinned to the bundled in-repo cost map: the default network-fetched ``main``
copy is a different branch and already re-prices this model per modality
token (no ``input_cost_per_image`` / ``input_cost_per_*_per_second``), which
makes these cases fail for reasons unrelated to the transformation.
"""
MODEL = "gemini-embedding-2"