biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[VLM] Separate out profiling-related logic (#11746)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-01-06 16:02:21 +08:00
committed by GitHub
parent 2a622d704a
commit 996357e480
17 changed files with 1036 additions and 739 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/multimodal/test_processing.py
View File

@@ -586,9 +586,10 @@ def test_limit_mm_per_prompt_dummy(model_id, limit, num_supported, is_valid):
)
processor = processor_factory(ctx, cache=None)
profiler = processor.profiling_info
mock_supported_mm_limits = MagicMock(return_value={"image": num_supported})
processor.get_supported_mm_limits = mock_supported_mm_limits
profiler.get_supported_mm_limits = mock_supported_mm_limits
if is_valid:
exc_ctx = nullcontext()
@@ -596,7 +597,7 @@ def test_limit_mm_per_prompt_dummy(model_id, limit, num_supported, is_valid):
exc_ctx = pytest.raises(ValueError, match="this model only supports")
with exc_ctx:
processor._get_and_validate_dummy_mm_counts()
profiler.get_mm_limits()
@pytest.mark.parametrize("model_id", ["llava-hf/llava-v1.6-mistral-7b-hf"])
@@ -723,7 +724,7 @@ def _test_processing_cache_correctness(
}
mm_counts = {k: len(vs) for k, vs in mm_data.items()}
prompt = baseline_processor._get_dummy_processor_inputs(
prompt = baseline_processor.profiling_info.get_dummy_processor_inputs(
model_config.max_model_len,
mm_counts,
).prompt_text