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] Skip Mllama HF runner tests with Transformers v4.55.0 (#22659)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Aug 12, 2025 at 01:00 UTC
c90fb03df566cd76b0e69f91158108909da80c51
1 file changed
+17
tests/models/multimodal/generation/test_mllama.py
+17
@@ -6,6 +6,7 @@ from typing import Optional, overload
6
import pytest
7
import torch
8
from transformers import AutoConfig, AutoModelForImageTextToText, AutoTokenizer
9
+from transformers import __version__ as TRANSFORMERS_VERSION
10
11
from vllm import LLM, SamplingParams
12
from vllm.attention.backends.flash_attn import FlashAttentionMetadata
@@ -285,6 +286,10 @@ def clear_cache():
286
@pytest.mark.parametrize("max_tokens", [128])
287
@pytest.mark.parametrize("num_logprobs", [5])
288
@pytest.mark.parametrize("attn_backend", LIST_ENC_DEC_SUPPORTED_BACKENDS)
289
+@pytest.mark.skipif(
290
+ TRANSFORMERS_VERSION == "4.55.0",
291
+ reason="Transformers v4.55.0 has a regression issue on mllama, "
292
+ "see: https://github.com/huggingface/transformers/pull/40083")
293
def test_models_single_leading_image(hf_runner, vllm_runner, image_assets,
294
model, sizes, dtype, max_tokens,
295
num_logprobs,
@@ -313,6 +318,10 @@ def test_models_single_leading_image(hf_runner, vllm_runner, image_assets,
318
@pytest.mark.parametrize("max_tokens", [128])
319
@pytest.mark.parametrize("num_logprobs", [5])
320
@pytest.mark.parametrize("attn_backend", LIST_ENC_DEC_SUPPORTED_BACKENDS)
321
+@pytest.mark.skipif(
322
+ TRANSFORMERS_VERSION == "4.55.0",
323
+ reason="Transformers v4.55.0 has a regression issue on mllama, "
324
+ "see: https://github.com/huggingface/transformers/pull/40083")
325
def test_models_multi_leading_images(hf_runner, vllm_runner, image_assets,
326
model, dtype, max_tokens, num_logprobs,
327
attn_backend: _Backend) -> None:
@@ -362,6 +371,10 @@ def test_models_multi_leading_images(hf_runner, vllm_runner, image_assets,
371
@pytest.mark.parametrize("max_tokens", [128])
372
@pytest.mark.parametrize("num_logprobs", [5])
373
@pytest.mark.parametrize("attn_backend", LIST_ENC_DEC_SUPPORTED_BACKENDS)
374
+@pytest.mark.skipif(
375
+ TRANSFORMERS_VERSION == "4.55.0",
376
+ reason="Transformers v4.55.0 has a regression issue on mllama, "
377
+ "see: https://github.com/huggingface/transformers/pull/40083")
378
def test_models_interleaved_images(hf_runner, vllm_runner, image_assets, model,
379
dtype, max_tokens, num_logprobs,
380
attn_backend: _Backend) -> None:
@@ -402,6 +415,10 @@ def test_models_interleaved_images(hf_runner, vllm_runner, image_assets, model,
415
@pytest.mark.parametrize("dtype", ["bfloat16"])
416
@pytest.mark.parametrize("max_tokens", [64])
417
@pytest.mark.parametrize("num_logprobs", [5])
418
+@pytest.mark.skipif(
419
+ TRANSFORMERS_VERSION == "4.55.0",
420
+ reason="Transformers v4.55.0 has a regression issue on mllama, "
421
+ "see: https://github.com/huggingface/transformers/pull/40083")
422
def test_models_distributed(
423
hf_runner,
424
vllm_runner,