mirror of
https://github.com/open-compass/opencompass.git
synced 2025-05-30 16:03:24 +08:00
bd50bad8b5
* [Feat] Add public dataset support for visualglm, qwenvl, and flamingo * [Fix] MMBench related changes. * [Fix] Openflamingo inference. * [Fix] Hide ckpt path. * [Fix] Pre-commit. --------- Co-authored-by: Haodong Duan <dhd.efz@gmail.com>
44 lines
1.5 KiB
Python
44 lines
1.5 KiB
Python
from opencompass.multimodal.models.qwen import QwenVLChatVQAPromptConstructor
|
|
|
|
# dataloader settings
|
|
val_pipeline = [
|
|
dict(type='mmpretrain.LoadImageFromFile'),
|
|
dict(type='mmpretrain.ToPIL', to_rgb=True),
|
|
dict(type='mmpretrain.torchvision/Resize',
|
|
size=(448, 448),
|
|
interpolation=3),
|
|
dict(type='mmpretrain.torchvision/ToTensor'),
|
|
dict(type='mmpretrain.torchvision/Normalize',
|
|
mean=(0.48145466, 0.4578275, 0.40821073),
|
|
std=(0.26862954, 0.26130258, 0.27577711)),
|
|
dict(
|
|
type='mmpretrain.PackInputs',
|
|
algorithm_keys=['question', 'gt_answer', 'gt_answer_weight'],
|
|
meta_keys=['question_id', 'image_id'],
|
|
)
|
|
]
|
|
|
|
dataset = dict(
|
|
type='mmpretrain.COCOVQA',
|
|
data_root='data/coco',
|
|
data_prefix='images/val2014',
|
|
question_file='annotations/v2_OpenEnded_mscoco_val2014_questions.json',
|
|
ann_file='annotations/v2_mscoco_val2014_annotations.json',
|
|
pipeline=val_pipeline)
|
|
|
|
qwen_vqav2_dataloader = dict(batch_size=1,
|
|
num_workers=4,
|
|
dataset=dataset,
|
|
collate_fn=dict(type='pseudo_collate'),
|
|
sampler=dict(type='DefaultSampler', shuffle=False))
|
|
|
|
# model settings
|
|
qwen_vqav2_model = dict(
|
|
type='qwen-vl-chat',
|
|
pretrained_path='Qwen/Qwen-VL-Chat', # or Huggingface repo id
|
|
prompt_constructor=dict(type=QwenVLChatVQAPromptConstructor)
|
|
)
|
|
|
|
# evaluation settings
|
|
qwen_vqav2_evaluator = [dict(type='mmpretrain.VQAAcc')]
|