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 16, 05:45 UTC
[CI/Build] Bump test transformers version (#10106)
Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
Isotr0py committed
Dec 6, 2024 at 00:05 UTC
998eeafe58c0263323b7fd8813c8b3d3f839bcbc
5 files changed
+3
-35
requirements-test.txt
+1
-1
@@ -550,7 +550,7 @@ tqdm==4.66.6
550
# transformers
551
tqdm-multiprocess==0.0.11
552
# via lm-eval
553
-transformers==4.45.2
553
+transformers==4.46.3
554
# via
555
# lm-eval
556
# peft
tests/models/decoder_only/vision_language/test_models.py
+1
-24
@@ -6,7 +6,6 @@ from pathlib import PosixPath
6
from typing import Type
7
8
import pytest
9
-import transformers
9
from transformers import AutoModelForVision2Seq
10
from transformers.utils import is_flash_attn_2_available
11
@@ -187,12 +186,6 @@ VLM_TEST_SETTINGS = {
186
comparator=check_outputs_equal,
187
max_tokens=8,
188
dtype="bfloat16",
190
- marks=[
191
- pytest.mark.skipif(
192
- transformers.__version__ < "4.46.2",
193
- reason="Model broken in HF, see huggingface/transformers#34379"
194
- ),
195
- ]
189
),
190
"fuyu": VLMTestInfo(
191
models=["adept/fuyu-8b"],
@@ -243,13 +236,7 @@ VLM_TEST_SETTINGS = {
236
max_model_len=8192,
237
max_num_seqs=2,
238
auto_cls=AutoModelForVision2Seq,
246
- marks=[
247
- pytest.mark.skipif(
248
- transformers.__version__ < "4.46.0",
249
- reason="Model introduced in HF >= 4.46.0"
250
- ),
251
- large_gpu_mark(min_gb=48),
252
- ],
239
+ marks=[large_gpu_mark(min_gb=48)],
240
),
241
"intern_vl": VLMTestInfo(
242
models=[
@@ -318,12 +305,6 @@ VLM_TEST_SETTINGS = {
305
auto_cls=AutoModelForVision2Seq,
306
vllm_output_post_proc=model_utils.llava_video_vllm_to_hf_output,
307
image_sizes=[((1669, 2560), (2560, 1669), (183, 488), (488, 183))],
321
- marks=[
322
- pytest.mark.skipif(
323
- transformers.__version__ < "4.46.2",
324
- reason="Model broken with changes in transformers 4.46"
325
- )
326
- ],
308
),
309
"minicpmv_25": VLMTestInfo(
310
models=["openbmb/MiniCPM-Llama3-V-2_5"],
@@ -404,10 +385,6 @@ VLM_TEST_SETTINGS = {
385
cuda_device_count_stateless() < 2,
386
reason="Need at least 2 GPUs to run the test.",
387
),
407
- pytest.mark.skipif(
408
- transformers.__version__ < "4.46.2",
409
- reason="Model broken in HF, see huggingface/transformers#34379"
410
- )
388
],
389
**COMMON_BROADCAST_SETTINGS # type: ignore
390
),
tests/models/decoder_only/vision_language/test_pixtral.py
+1
-1
@@ -228,7 +228,7 @@ def test_model_engine(vllm_runner, model: str, dtype: str) -> None:
228
name_1="output")
229
230
231
-@large_gpu_test(min_gb=24)
231
+@large_gpu_test(min_gb=48)
232
@pytest.mark.parametrize(
233
"prompt,expected_ranges",
234
[(_create_engine_inputs_hf(IMG_URLS[:1]), [{
tests/models/embedding/vision_language/test_llava_next.py
-4
@@ -2,7 +2,6 @@ from typing import List, Type
2
3
import pytest
4
import torch.nn.functional as F
5
-import transformers
5
from transformers import AutoModelForVision2Seq
6
7
from ....conftest import IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner
@@ -86,9 +85,6 @@ def _run_test(
85
)
86
87
89
-@pytest.mark.skipif(transformers.__version__.startswith("4.46"),
90
- reason="Model broken with changes in transformers 4.46")
91
-@pytest.mark.core_model
88
@pytest.mark.parametrize("model", MODELS)
89
@pytest.mark.parametrize("dtype", ["half"])
90
def test_models_text(
tests/models/test_initialization.py
-5
@@ -1,7 +1,6 @@
1
from unittest.mock import patch
2
3
import pytest
4
-import transformers
4
from transformers import PretrainedConfig
5
6
from vllm import LLM
@@ -11,10 +10,6 @@ from .registry import HF_EXAMPLE_MODELS
10
11
@pytest.mark.parametrize("model_arch", HF_EXAMPLE_MODELS.get_supported_archs())
12
def test_can_initialize(model_arch):
14
- if (model_arch in {"Idefics3ForConditionalGeneration", "GlmForCausalLM"}
15
- and transformers.__version__ < "4.46.0"):
16
- pytest.skip(reason="Model introduced in HF >= 4.46.0")
17
-
13
model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch)
14
if not model_info.is_available_online:
15
pytest.skip("Model is not available online")