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] Clean up deadcode for legacy processing pipeline (#24153)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Sep 3, 2025 at 16:34 UTC
9c99e4871f12aa57072b3dfd14157018d30e327b
2 files changed
+1
-7
tests/models/multimodal/processing/test_tensor_schema.py
-3
@@ -41,9 +41,6 @@ ARCH_NEEDS_EXTRAS = [
41
]
42
REPO_ID_TO_SKIP = {
43
"nm-testing/pixtral-12b-FP8-dynamic": "duplicated test",
44
- # FIXME(Isotr0py): enable GPT-OSS based InternVL3.5 model
45
- # after support PP for GPT-OSS
46
- "OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview": "Broken model",
44
}
45
46
ImageInput = list[Image.Image]
vllm/multimodal/utils.py
+1
-4
@@ -378,10 +378,7 @@ def group_mm_inputs_by_modality(
378
elif len(mm_input) == 1:
379
return next(iter(mm_input.keys()))
380
381
- # FIXME(Isotr0py): Modality of mm_input from legacy pipeline is empty,
382
- # this is used to make InternVL with legacy pipeline still work with v1.
383
- else:
384
- return ""
381
+ raise AssertionError("This line should be unreachable.")
382
383
return [
384
list(group) for _, group in groupby(mm_inputs, key=modality_group_func)