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

[CI/Build] Split up models tests (#10069)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-11-10 03:39:14 +08:00
committed by GitHub
parent b09895a618
commit 51c2e1fcef
21 changed files with 115 additions and 129 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
tests/models/decoder_only/vision_language/test_models.py
View File

@@ -130,8 +130,8 @@ VLM_TEST_SETTINGS = {
max_num_seqs=2,
auto_cls=AutoModelForVision2Seq,
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
image_size_factors=[(), (0.25,), (0.25, 0.25, 0.25), (0.25, 0.2, 0.15)],
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
#### Extended model tests
"blip2": VLMTestInfo(
@@ -159,9 +159,9 @@ VLM_TEST_SETTINGS = {
dtype="bfloat16",
marks=[
pytest.mark.skipif(
transformers.__version__.startswith("4.46"),
transformers.__version__ < "4.46.2",
reason="Model broken in HF, see huggingface/transformers#34379"
)
),
]
),
"fuyu": VLMTestInfo(
@@ -185,8 +185,8 @@ VLM_TEST_SETTINGS = {
max_num_seqs=2,
dtype="bfloat16",
get_stop_token_ids=lambda tok: [151329, 151336, 151338],
marks=[large_gpu_mark(min_gb=48)],
patch_hf_runner=model_utils.glm_patch_hf_runner,
marks=[large_gpu_mark(min_gb=48)],
),
"h2ovl": VLMTestInfo(
models = [
@@ -205,6 +205,22 @@ VLM_TEST_SETTINGS = {
use_tokenizer_eos=True,
patch_hf_runner=model_utils.h2ovl_patch_hf_runner,
),
"idefics3": VLMTestInfo(
models=["HuggingFaceM4/Idefics3-8B-Llama3"],
test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
prompt_formatter=lambda img_prompt:f"<|begin_of_text|>User:{img_prompt}<end_of_utterance>\nAssistant:", # noqa: E501
img_idx_to_prompt=lambda idx: "<image>",
max_model_len=8192,
max_num_seqs=2,
auto_cls=AutoModelForVision2Seq,
marks=[
pytest.mark.skipif(
transformers.__version__ < "4.46.0",
reason="Model introduced in HF >= 4.46.0"
),
large_gpu_mark(min_gb=48),
],
),
"intern_vl": VLMTestInfo(
models=[
"OpenGVLab/InternVL2-1B",
@@ -263,7 +279,6 @@ VLM_TEST_SETTINGS = {
runner_mm_key="videos",
)],
),
# FIXME
"llava_next_video": VLMTestInfo(
models=["llava-hf/LLaVA-NeXT-Video-7B-hf"],
test_type=VLMTestType.VIDEO,
@@ -275,7 +290,7 @@ VLM_TEST_SETTINGS = {
image_sizes=[((1669, 2560), (2560, 1669), (183, 488), (488, 183))],
marks=[
pytest.mark.skipif(
transformers.__version__.startswith("4.46"),
transformers.__version__ < "4.46.2",
reason="Model broken with changes in transformers 4.46"
)
],
@@ -316,6 +331,7 @@ VLM_TEST_SETTINGS = {
max_model_len=8192,
max_num_seqs=2,
auto_cls=AutoModelForVision2Seq,
marks=[large_gpu_mark(min_gb=48)],
),
"qwen": VLMTestInfo(
models=["Qwen/Qwen-VL"],
@@ -327,22 +343,6 @@ VLM_TEST_SETTINGS = {
vllm_output_post_proc=model_utils.qwen_vllm_to_hf_output,
prompt_path_encoder=model_utils.qwen_prompt_path_encoder,
),
"idefics3": VLMTestInfo(
models=["HuggingFaceM4/Idefics3-8B-Llama3"],
test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
prompt_formatter=lambda img_prompt:f"<|begin_of_text|>User:{img_prompt}<end_of_utterance>\nAssistant:", # noqa: E501
img_idx_to_prompt=lambda idx: "<image>",
max_model_len=8192,
max_num_seqs=2,
auto_cls=AutoModelForVision2Seq,
marks=[
pytest.mark.skipif(
transformers.__version__ < "4.46.0",
reason="Model introduced in HF >= 4.46.0"
),
large_gpu_mark(min_gb=48),
],
),
### Tensor parallel / multi-gpu broadcast tests
"broadcast-chameleon": VLMTestInfo(
models=["facebook/chameleon-7b"],
@@ -362,7 +362,7 @@ VLM_TEST_SETTINGS = {
reason="Need at least 2 GPUs to run the test.",
),
pytest.mark.skipif(
transformers.__version__.startswith("4.46"),
transformers.__version__ < "4.46.2",
reason="Model broken in HF, see huggingface/transformers#34379"
)
],