A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI/Build] Sync multimodal tests (#23181)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Aug 20, 2025 at 13:06 UTC
de7b67a0232e35ae8e8ecd944aeddfc8cbc02631
2 files changed
+18
-16
tests/models/multimodal/processing/test_common.py
+7
-3
@@ -275,16 +275,17 @@ def _test_processing_correctness_one(
275
"google/gemma-3n-E2B-it",
276
"zai-org/glm-4v-9b",
277
"zai-org/GLM-4.1V-9B-Thinking",
278
+ "zai-org/GLM-4.5V",
279
"ibm-granite/granite-speech-3.3-2b",
280
"h2oai/h2ovl-mississippi-800m",
281
+ "naver-hyperclovax/HyperCLOVAX-SEED-Vision-Instruct-3B",
282
+ "HuggingFaceM4/Idefics3-8B-Llama3",
283
"internlm/Intern-S1",
284
"OpenGVLab/InternVL2-1B",
285
"OpenGVLab/InternVL3-1B",
283
- "HuggingFaceM4/Idefics3-8B-Llama3",
284
- "HuggingFaceTB/SmolVLM2-2.2B-Instruct",
286
+ "Kwai-Keye/Keye-VL-8B-Preview",
287
"moonshotai/Kimi-VL-A3B-Instruct",
288
"meta-llama/Llama-4-Scout-17B-16E-Instruct",
287
- "naver-hyperclovax/HyperCLOVAX-SEED-Vision-Instruct-3B",
289
"llava-hf/llava-1.5-7b-hf",
290
"llava-hf/llava-v1.6-mistral-7b-hf",
291
"llava-hf/LLaVA-NeXT-Video-7B-hf",
@@ -315,10 +316,13 @@ def _test_processing_correctness_one(
316
"Qwen/Qwen2-Audio-7B-Instruct",
317
"Qwen/Qwen2.5-Omni-3B",
318
"Skywork/Skywork-R1V-38B",
319
+ "HuggingFaceTB/SmolVLM2-2.2B-Instruct",
320
+ "stepfun-ai/step3",
321
"fixie-ai/ultravox-v0_5-llama-3_2-1b",
322
"openai/whisper-large-v3",
323
"omni-research/Tarsier-7b",
324
"omni-research/Tarsier2-Recap-7b",
325
+ "mistralai/Voxtral-Mini-3B-2507",
326
])
327
@pytest.mark.parametrize("hit_rate", [0.3, 0.5, 1.0])
328
@pytest.mark.parametrize("num_batches", [32])
tests/models/registry.py
+11
-13
@@ -215,9 +215,6 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
215
"HunYuanDenseV1ForCausalLM":_HfExamplesInfo("tencent/Hunyuan-7B-Instruct-0124",
216
trust_remote_code=True,
217
is_available_online=False),
218
- "HCXVisionForCausalLM": _HfExamplesInfo(
219
- "naver-hyperclovax/HyperCLOVAX-SEED-Vision-Instruct-3B",
220
- trust_remote_code=True),
218
"InternLMForCausalLM": _HfExamplesInfo("internlm/internlm-chat-7b",
219
trust_remote_code=True),
220
"InternLM2ForCausalLM": _HfExamplesInfo("internlm/internlm2-chat-7b",
@@ -298,8 +295,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
295
"StableLmForCausalLM": _HfExamplesInfo("stabilityai/stablelm-3b-4e1t"),
296
"Starcoder2ForCausalLM": _HfExamplesInfo("bigcode/starcoder2-3b"),
297
"Step3TextForCausalLM": _HfExamplesInfo("stepfun-ai/step3",
301
- trust_remote_code=True,
302
- is_available_online=False),
298
+ trust_remote_code=True),
299
"SolarForCausalLM": _HfExamplesInfo("upstage/solar-pro-preview-instruct",
300
trust_remote_code=True),
301
"TeleChat2ForCausalLM": _HfExamplesInfo("Tele-AI/TeleChat2-3B",
@@ -405,22 +401,24 @@ _MULTIMODAL_EXAMPLE_MODELS = {
401
hf_overrides={"architectures": ["GLM4VForCausalLM"]}), # noqa: E501
402
"Glm4vForConditionalGeneration": _HfExamplesInfo("zai-org/GLM-4.1V-9B-Thinking"), # noqa: E501
403
"Glm4vMoeForConditionalGeneration": _HfExamplesInfo("zai-org/GLM-4.5V",
408
- is_available_online=False), # noqa: E501
404
+ min_transformers_version="4.56"), # noqa: E501
405
"H2OVLChatModel": _HfExamplesInfo("h2oai/h2ovl-mississippi-800m",
406
trust_remote_code=True,
407
extras={"2b": "h2oai/h2ovl-mississippi-2b"}, # noqa: E501
408
max_transformers_version="4.48", # noqa: E501
409
transformers_version_reason="HF model is not compatible."), # noqa: E501
410
+ "HCXVisionForCausalLM": _HfExamplesInfo("naver-hyperclovax/HyperCLOVAX-SEED-Vision-Instruct-3B", # noqa: E501
411
+ trust_remote_code=True),
412
"Idefics3ForConditionalGeneration": _HfExamplesInfo("HuggingFaceM4/Idefics3-8B-Llama3", # noqa: E501
413
{"tiny": "HuggingFaceTB/SmolVLM-256M-Instruct"}, # noqa: E501
414
min_transformers_version="4.55.1",
415
transformers_version_reason="HF model broken in 4.55.0"), # noqa: E501
416
+ "InternS1ForConditionalGeneration": _HfExamplesInfo("internlm/Intern-S1",
417
+ trust_remote_code=True), # noqa: E501
418
"InternVLChatModel": _HfExamplesInfo("OpenGVLab/InternVL2-1B",
419
extras={"2B": "OpenGVLab/InternVL2-2B",
420
"3.0": "OpenGVLab/InternVL3-1B"}, # noqa: E501
421
trust_remote_code=True),
422
- "InternS1ForConditionalGeneration": _HfExamplesInfo("internlm/Intern-S1",
423
- trust_remote_code=True),
422
"KeyeForConditionalGeneration": _HfExamplesInfo("Kwai-Keye/Keye-VL-8B-Preview", # noqa: E501
423
trust_remote_code=True),
424
"KimiVLForConditionalGeneration": _HfExamplesInfo("moonshotai/Kimi-VL-A3B-Instruct", # noqa: E501
@@ -464,9 +462,10 @@ _MULTIMODAL_EXAMPLE_MODELS = {
462
transformers_version_reason="HF model is not compatible", # noqa: E501
463
extras={"1.6-llama": "AIDC-AI/Ovis1.6-Llama3.2-3B",
464
"1.6-gemma": "AIDC-AI/Ovis1.6-Gemma2-9B"}), # noqa: E501
467
- "Ovis2_5": _HfExamplesInfo("AIDC-AI/Ovis2.5-2B", trust_remote_code=True,
468
- max_transformers_version="4.53",
469
- transformers_version_reason="HF model is not compatible"), # noqa: E501
465
+ "Ovis2_5": _HfExamplesInfo("AIDC-AI/Ovis2.5-2B",
466
+ trust_remote_code=True,
467
+ max_transformers_version="4.53",
468
+ transformers_version_reason="HF model is not compatible"), # noqa: E501
469
"PaliGemmaForConditionalGeneration": _HfExamplesInfo("google/paligemma-3b-mix-224", # noqa: E501
470
extras={"v2": "google/paligemma2-3b-ft-docci-448"}), # noqa: E501
471
"Phi3VForCausalLM": _HfExamplesInfo("microsoft/Phi-3-vision-128k-instruct",
@@ -496,8 +495,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
495
min_transformers_version="4.55.1",
496
transformers_version_reason="HF model broken in 4.55.0"), # noqa: E501
497
"Step3VLForConditionalGeneration": _HfExamplesInfo("stepfun-ai/step3",
499
- trust_remote_code=True,
500
- is_available_online=False),
498
+ trust_remote_code=True),
499
"UltravoxModel": _HfExamplesInfo("fixie-ai/ultravox-v0_5-llama-3_2-1b", # noqa: E501
500
trust_remote_code=True),
501
"TarsierForConditionalGeneration": _HfExamplesInfo("omni-research/Tarsier-7b"), # noqa: E501