From a4ee6277058be405fcd97e5edab467b7e543a7a5 Mon Sep 17 00:00:00 2001 From: Isotr0py <2037008807@qq.com> Date: Thu, 7 Nov 2024 13:11:19 +0800 Subject: [PATCH 1/4] bump test transformers version to 4.46.2 Signed-off-by: Isotr0py <2037008807@qq.com> --- requirements-test.txt | 4 ++-- .../vision_language/test_models.py | 21 ------------------- .../vision_language/test_llava_next.py | 3 --- vllm/model_executor/models/idefics3.py | 3 +-- 4 files changed, 3 insertions(+), 28 deletions(-) diff --git a/requirements-test.txt b/requirements-test.txt index 518e81021cbc..416a42a0b176 100644 --- a/requirements-test.txt +++ b/requirements-test.txt @@ -490,7 +490,7 @@ tiktoken==0.8.0 # via lm-eval timm==1.0.11 # via -r requirements-test.in -tokenizers==0.20.1 +tokenizers==0.20.3 # via transformers torch==2.5.1 # via @@ -518,7 +518,7 @@ tqdm==4.66.6 # transformers tqdm-multiprocess==0.0.11 # via lm-eval -transformers==4.45.2 +transformers==4.46.2 # via # lm-eval # peft diff --git a/tests/models/decoder_only/vision_language/test_models.py b/tests/models/decoder_only/vision_language/test_models.py index 3dbfaafb781a..8cd33b00c2b3 100644 --- a/tests/models/decoder_only/vision_language/test_models.py +++ b/tests/models/decoder_only/vision_language/test_models.py @@ -6,7 +6,6 @@ from typing import Type import pytest -import transformers from transformers import AutoModelForVision2Seq from vllm.platforms import current_platform @@ -155,12 +154,6 @@ comparator=check_outputs_equal, max_tokens=8, dtype="bfloat16", - marks=[ - pytest.mark.skipif( - transformers.__version__.startswith("4.46"), - reason="Model broken in HF, see huggingface/transformers#34379" - ) - ] ), "fuyu": VLMTestInfo( models=["adept/fuyu-8b"], @@ -273,12 +266,6 @@ auto_cls=AutoModelForVision2Seq, vllm_output_post_proc=model_utils.llava_video_vllm_to_hf_output, image_sizes=[((1669, 2560), (2560, 1669), (183, 488), (488, 183))], - marks=[ - pytest.mark.skipif( - transformers.__version__.startswith("4.46"), - reason="Model broken with changes in transformers 4.46" - ) - ], ), "minicpmv": VLMTestInfo( models=["openbmb/MiniCPM-Llama3-V-2_5"], @@ -336,10 +323,6 @@ max_num_seqs=2, auto_cls=AutoModelForVision2Seq, marks=[ - pytest.mark.skipif( - transformers.__version__ < "4.46.0", - reason="Model introduced in HF >= 4.46.0" - ), large_gpu_mark(min_gb=48), ], ), @@ -361,10 +344,6 @@ cuda_device_count_stateless() < 2, reason="Need at least 2 GPUs to run the test.", ), - pytest.mark.skipif( - transformers.__version__.startswith("4.46"), - reason="Model broken in HF, see huggingface/transformers#34379" - ) ], **COMMON_BROADCAST_SETTINGS # type: ignore ), diff --git a/tests/models/embedding/vision_language/test_llava_next.py b/tests/models/embedding/vision_language/test_llava_next.py index 9fab5898a06b..52aef8c34d6f 100644 --- a/tests/models/embedding/vision_language/test_llava_next.py +++ b/tests/models/embedding/vision_language/test_llava_next.py @@ -2,7 +2,6 @@ import pytest import torch.nn.functional as F -import transformers from transformers import AutoModelForVision2Seq from ....conftest import IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner @@ -86,8 +85,6 @@ def _run_test( ) -@pytest.mark.skipif(transformers.__version__.startswith("4.46"), - reason="Model broken with changes in transformers 4.46") @pytest.mark.parametrize("model", MODELS) @pytest.mark.parametrize("dtype", ["half"]) def test_models_text( diff --git a/vllm/model_executor/models/idefics3.py b/vllm/model_executor/models/idefics3.py index e4c98f22fb16..8671e4bc8d94 100644 --- a/vllm/model_executor/models/idefics3.py +++ b/vllm/model_executor/models/idefics3.py @@ -21,8 +21,7 @@ import torch.utils.checkpoint from PIL import Image from torch import nn -# Temporary solution for transformers below 4.46.0. -from transformers import PretrainedConfig as Idefics3Config +from transformers import Idefics3Config from vllm.attention import AttentionMetadata from vllm.config import CacheConfig, MultiModalConfig From 82477202c7a173642587f921eac3a479207bbf6d Mon Sep 17 00:00:00 2001 From: Isotr0py <2037008807@qq.com> Date: Thu, 7 Nov 2024 14:14:05 +0800 Subject: [PATCH 2/4] remove FIXME Signed-off-by: Isotr0py <2037008807@qq.com> --- tests/models/decoder_only/vision_language/test_models.py | 1 - 1 file changed, 1 deletion(-) diff --git a/tests/models/decoder_only/vision_language/test_models.py b/tests/models/decoder_only/vision_language/test_models.py index 8cd33b00c2b3..460dc9b3f3db 100644 --- a/tests/models/decoder_only/vision_language/test_models.py +++ b/tests/models/decoder_only/vision_language/test_models.py @@ -256,7 +256,6 @@ runner_mm_key="videos", )], ), - # FIXME "llava_next_video": VLMTestInfo( models=["llava-hf/LLaVA-NeXT-Video-7B-hf"], test_type=VLMTestType.VIDEO, From 9e3b23369bf09b89d39456b545446361089e8500 Mon Sep 17 00:00:00 2001 From: DarkLight1337 Date: Thu, 5 Dec 2024 10:59:27 +0000 Subject: [PATCH 3/4] Bump transformers patch version Signed-off-by: DarkLight1337 --- requirements-test.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/requirements-test.txt b/requirements-test.txt index 322868e224e9..19369254dbe2 100644 --- a/requirements-test.txt +++ b/requirements-test.txt @@ -550,7 +550,7 @@ tqdm==4.66.6 # transformers tqdm-multiprocess==0.0.11 # via lm-eval -transformers==4.46.2 +transformers==4.46.3 # via # lm-eval # peft From bfb03a39d4e1f03b49d96be8210ff50889b94df6 Mon Sep 17 00:00:00 2001 From: DarkLight1337 Date: Thu, 5 Dec 2024 14:09:09 +0000 Subject: [PATCH 4/4] Update Signed-off-by: DarkLight1337 --- tests/models/decoder_only/vision_language/test_pixtral.py | 2 +- tests/models/test_initialization.py | 5 ----- 2 files changed, 1 insertion(+), 6 deletions(-) diff --git a/tests/models/decoder_only/vision_language/test_pixtral.py b/tests/models/decoder_only/vision_language/test_pixtral.py index 6233860747b9..90c0fab99054 100644 --- a/tests/models/decoder_only/vision_language/test_pixtral.py +++ b/tests/models/decoder_only/vision_language/test_pixtral.py @@ -228,7 +228,7 @@ def test_model_engine(vllm_runner, model: str, dtype: str) -> None: name_1="output") -@large_gpu_test(min_gb=24) +@large_gpu_test(min_gb=48) @pytest.mark.parametrize( "prompt,expected_ranges", [(_create_engine_inputs_hf(IMG_URLS[:1]), [{ diff --git a/tests/models/test_initialization.py b/tests/models/test_initialization.py index 2a072737db04..3b728f2744fc 100644 --- a/tests/models/test_initialization.py +++ b/tests/models/test_initialization.py @@ -1,7 +1,6 @@ from unittest.mock import patch import pytest -import transformers from transformers import PretrainedConfig from vllm import LLM @@ -11,10 +10,6 @@ @pytest.mark.parametrize("model_arch", HF_EXAMPLE_MODELS.get_supported_archs()) def test_can_initialize(model_arch): - if (model_arch in {"Idefics3ForConditionalGeneration", "GlmForCausalLM"} - and transformers.__version__ < "4.46.0"): - pytest.skip(reason="Model introduced in HF >= 4.46.0") - model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch) if not model_info.is_available_online: pytest.skip("Model is not available online")