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

Enable CUDA graph support for llama 3.2 vision (#14917)

Browse Source 
Signed-off-by: Matt Ritter <100659061+mritterfigma@users.noreply.github.com>
This commit is contained in:
Matt Ritter
2025-03-19 23:29:16 -07:00
committed by GitHub
parent 2f726b241e
commit a8652f4f0f
3 changed files with 1 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/models/encoder_decoder/vision_language/test_mllama.py
View File

@@ -215,7 +215,6 @@ def _run_test(
max_num_seqs=2,
tensor_parallel_size=tensor_parallel_size,
distributed_executor_backend=distributed_executor_backend,
enforce_eager=True,
limit_mm_per_prompt={"image": _LIMIT_IMAGE_PER_PROMPT
}) as vllm_model:
vllm_outputs_per_image = [
@@ -425,7 +424,6 @@ def test_bnb_regression(
dtype=dtype,
max_model_len=4096,
max_num_seqs=2,
enforce_eager=True,
quantization="bitsandbytes",
load_format="bitsandbytes",
)
@@ -481,7 +479,6 @@ def test_explicit_implicit_prompt(
max_model_len=4096,
max_num_seqs=2,
tensor_parallel_size=1,
enforce_eager=True,
)
sampling_params = SamplingParams(
temperature=0,
@@ -513,7 +510,6 @@ def test_regression(vllm_runner, image_assets, model, dtype, max_tokens,
max_model_len=4096,
max_num_seqs=2,
tensor_parallel_size=1,
enforce_eager=True,
limit_mm_per_prompt={"image":
_LIMIT_IMAGE_PER_PROMPT}) as vllm_model: