[CI] Shard quantization, pooling, multimodal APIs, core kernels and PCP eval automatically (#61477)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com> Co-authored-by: Codex <noreply@openai.com>
This commit is contained in:
@@ -69,6 +69,7 @@ def test_inject_into_mm_cache(
|
||||
mm_processor_cache_type,
|
||||
caplog_vllm,
|
||||
vllm_runner,
|
||||
monkeypatch,
|
||||
):
|
||||
"""Test that inject_into_mm_cache() injects pre-processed mm_kwargs into
|
||||
the processor cache and MM cache hit metrics are updated correctly.
|
||||
@@ -87,6 +88,9 @@ def test_inject_into_mm_cache(
|
||||
limit_mm_per_prompt={"image": 2},
|
||||
mm_processor_cache_type=mm_processor_cache_type,
|
||||
) as runner:
|
||||
# The first LLM initialization can reset logger propagation.
|
||||
monkeypatch.setattr(logging.getLogger("vllm"), "propagate", True)
|
||||
|
||||
# Step 1: Normal requests to populate the cache
|
||||
runner.llm.chat(_make_messages(image_urls[0]))
|
||||
assert _get_mm_cache_stats(runner.llm.get_metrics()) == (1, 0)
|
||||
|
||||
Reference in New Issue
Block a user