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
[Misc] Replace flaky image urls in pixtral test (#22574)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Aug 10, 2025 at 21:18 UTC
049c245143ef0f8fd338fc3200f51a18fc53b403
2 files changed
+13
-14
tests/models/multimodal/generation/test_pixtral.py
+11
-13
@@ -18,7 +18,7 @@ from vllm.multimodal.inputs import PlaceholderRange
18
from vllm.sequence import Logprob, SampleLogprobs
19
20
from ....utils import VLLM_PATH, large_gpu_test
21
-from ...utils import check_logprobs_close
21
+from ...utils import check_logprobs_close, dummy_hf_overrides
22
23
if TYPE_CHECKING:
24
from _typeshed import StrPath
@@ -29,10 +29,10 @@ MISTRAL_SMALL_3_1_ID = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
29
MODELS = [PIXTRAL_ID, MISTRAL_SMALL_3_1_ID]
30
31
IMG_URLS = [
32
- "https://picsum.photos/id/237/400/300",
33
- "https://picsum.photos/id/231/200/300",
34
- "https://picsum.photos/id/27/500/500",
35
- "https://picsum.photos/id/17/150/600",
32
+ "https://huggingface.co/datasets/Isotr0py/mistral-test-images/resolve/main/237-400x300.jpg",
33
+ "https://huggingface.co/datasets/Isotr0py/mistral-test-images/resolve/main/231-200x300.jpg",
34
+ "https://huggingface.co/datasets/Isotr0py/mistral-test-images/resolve/main/27-500x500.jpg",
35
+ "https://huggingface.co/datasets/Isotr0py/mistral-test-images/resolve/main/17-150x600.jpg",
36
]
37
PROMPT = "Describe each image in one short sentence."
38
@@ -110,11 +110,6 @@ MSGS = [
110
_create_msg_format(IMG_URLS[:2]),
111
_create_msg_format(IMG_URLS),
112
]
113
-ENGINE_INPUTS = [
114
- _create_engine_inputs(IMG_URLS[:1]),
115
- _create_engine_inputs(IMG_URLS[:2]),
116
- _create_engine_inputs(IMG_URLS),
117
-]
113
114
SAMPLING_PARAMS = SamplingParams(max_tokens=512, temperature=0.0, logprobs=5)
115
LIMIT_MM_PER_PROMPT = dict(image=4)
@@ -195,7 +190,6 @@ def test_chat(
190
name_1="output")
191
192
198
-@large_gpu_test(min_gb=48)
193
@pytest.mark.parametrize("prompt,expected_ranges",
194
[(_create_engine_inputs_hf(IMG_URLS[:1]),
195
[PlaceholderRange(offset=11, length=494)]),
@@ -204,7 +198,7 @@ def test_chat(
198
PlaceholderRange(offset=277, length=1056),
199
PlaceholderRange(offset=1333, length=418)
200
])])
207
-def test_multi_modal_placeholders(vllm_runner, prompt,
201
+def test_multi_modal_placeholders(vllm_runner, prompt: TextPrompt,
202
expected_ranges: list[PlaceholderRange],
203
monkeypatch) -> None:
204
@@ -215,6 +209,8 @@ def test_multi_modal_placeholders(vllm_runner, prompt,
209
"mistral-community/pixtral-12b",
210
max_model_len=8192,
211
limit_mm_per_prompt=LIMIT_MM_PER_PROMPT,
212
+ load_format="dummy",
213
+ hf_overrides=dummy_hf_overrides,
214
) as vllm_model:
215
outputs = vllm_model.llm.generate(prompt)
216
@@ -230,5 +226,7 @@ def test_multi_modal_placeholders(vllm_runner, prompt,
226
expected_ranges), f"{image_placeholder_ranges=}"
227
for real_range, expected_range in zip(image_placeholder_ranges,
228
expected_ranges):
233
- assert real_range == expected_range, \
229
+ assert real_range.offset == expected_range.offset, \
230
+ f"{real_range=} {expected_range=}"
231
+ assert real_range.length == expected_range.length, \
232
f"{real_range=} {expected_range=}"
tests/models/utils.py
+2
-1
@@ -357,7 +357,8 @@ class RerankModelInfo(NamedTuple):
357
358
def dummy_hf_overrides(
359
hf_config: PretrainedConfig,
360
- model_arch: str,
360
+ *,
361
+ model_arch: str = "",
362
exist_overrides: Optional[dict[str, Any]] = None,
363
) -> PretrainedConfig:
364
"""