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, 04:30 UTC
[CI/Build] Check transformers v4.47 (#10991)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Dec 11, 2024 at 13:02 UTC
2e33fe419186c65a18da6668972d61d7bbc31564
3 files changed
+3
-12
requirements-test.txt
+2
-2
@@ -506,7 +506,7 @@ tiktoken==0.7.0
506
# mistral-common
507
timm==1.0.11
508
# via -r requirements-test.in
509
-tokenizers==0.20.3
509
+tokenizers==0.21.0
510
# via transformers
511
torch==2.5.1
512
# via
@@ -534,7 +534,7 @@ tqdm==4.66.6
534
# transformers
535
tqdm-multiprocess==0.0.11
536
# via lm-eval
537
-transformers==4.46.3
537
+transformers==4.47.0
538
# via
539
# lm-eval
540
# peft
tests/models/decoder_only/vision_language/mm_processor_kwargs/test_idefics3.py
-9
@@ -3,7 +3,6 @@ from typing import Optional
3
4
import pytest
5
import torch
6
-import transformers
6
from transformers import AutoImageProcessor, AutoTokenizer
7
8
from vllm.inputs import InputContext, token_inputs
@@ -36,8 +35,6 @@ def get_max_idefics3_image_tokens():
35
return get_max_idefics3_image_tokens
36
37
39
-@pytest.mark.skipif(transformers.__version__ < "4.46.0",
40
- reason="Model introduced in HF >= 4.46.0")
38
@pytest.mark.parametrize("model", models)
39
@pytest.mark.parametrize("longest_edge", [None, 168, 336, 400, 2 * 336])
40
def test_input_mapper_override(model: str, image_assets: _ImageAssets,
@@ -77,8 +74,6 @@ def test_input_mapper_override(model: str, image_assets: _ImageAssets,
74
assert torch.all(hf_result["pixel_values"] == vllm_result["pixel_values"])
75
76
80
-@pytest.mark.skipif(transformers.__version__ < "4.46.0",
81
- reason="Model introduced in HF >= 4.46.0")
77
@pytest.mark.parametrize("model", models)
78
@pytest.mark.parametrize("longest_edge, expected_max_tokens", [
79
(None, 2873),
@@ -107,8 +102,6 @@ def test_max_tokens_override(get_max_idefics3_image_tokens, model: str,
102
assert expected_max_tokens == actual_max_tokens
103
104
110
-@pytest.mark.skipif(transformers.__version__ < "4.46.0",
111
- reason="Model introduced in HF >= 4.46.0")
105
@pytest.mark.parametrize("model", models)
106
@pytest.mark.parametrize("longest_edge, toks_per_img, num_imgs", [
107
(168, 169, 1),
@@ -143,8 +136,6 @@ def test_dummy_data_override(dummy_data_for_idefics3, model: str,
136
assert img_tok_count == toks_per_img * num_imgs
137
138
146
-@pytest.mark.skipif(transformers.__version__ < "4.46.0",
147
- reason="Model introduced in HF >= 4.46.0")
139
@pytest.mark.parametrize("model", models)
140
@pytest.mark.parametrize("longest_edge,expected_toks_per_img,num_imgs", [
141
(336, 169 * (1**2 + 1), 1),
tests/models/embedding/vision_language/test_llava_next.py
+1
-1
@@ -86,7 +86,7 @@ def _run_test(
86
)
87
88
89
-@pytest.mark.skipif(transformers.__version__.startswith("4.46"),
89
+@pytest.mark.skipif(transformers.__version__ >= "4.46",
90
reason="Model broken with changes in transformers 4.46")
91
@pytest.mark.core_model
92
@pytest.mark.parametrize("model", MODELS)