File size: 2,471 Bytes
256a159 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 |
from opencompass.multimodal.models.openflamingo import OpenFlamingoVQAPromptConstructor
# dataloader settings
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='mmpretrain.ResizeEdge',
scale=224,
interpolation='bicubic',
backend='pillow'),
dict(type='CenterCrop', crop_size=(224, 224)),
dict(
type='mmpretrain.PackInputs',
algorithm_keys=['question', 'gt_answer', 'gt_answer_weight'],
meta_keys=['question_id', 'image_id'],
)
]
dataset = dict(
type='mmpretrain.TextVQA',
data_root='data/textvqa',
ann_file='annotations/TextVQA_0.5.1_val.json',
pipeline=val_pipeline,
data_prefix='images/train_images',
)
openflamingo_textvqa_dataloader = dict(
batch_size=8,
num_workers=4,
dataset=dataset,
sampler=dict(type='DefaultSampler', shuffle=False),
collate_fn=dict(type='default_collate'),
persistent_workers=True,
)
# model settings
openflamingo_textvqa_model = dict(
type='openflamingo',
data_preprocessor=dict(
type='mmpretrain.MultiModalDataPreprocessor',
mean=[122.770938, 116.7460125, 104.09373615],
std=[68.5005327, 66.6321579, 70.32316305],
to_rgb=True,
),
tokenizer=dict(type='mmpretrain.LlamaTokenizer',
name_or_path='decapoda-research/llama-7b-hf'),
vision_encoder=dict(
type='mmpretrain.VisionTransformer',
arch='l',
patch_size=14,
pre_norm=True,
norm_cfg=dict(type='LN', eps=1e-5),
layer_cfgs=dict(act_cfg=dict(type='mmpretrain.QuickGELU')),
final_norm=False,
out_type='raw',
pretrained= # noqa: E251
'/path/to/vision/encoder', # noqa
),
lang_encoder=dict(
base=dict(type='mmpretrain.AutoModelForCausalLM',
name_or_path=
'decapoda-research/llama-7b-hf',
local_files_only=True),
adapter=dict(type='mmpretrain.FlamingoLMAdapter',
vis_hidden_size=1024,
cross_attn_every_n_layers=4,
use_media_placement_augmentation=False),
),
task='vqa',
generation_cfg=dict(num_beams=3, max_new_tokens=20, length_penalty=-2.0),
prompt_constructor=dict(type=OpenFlamingoVQAPromptConstructor)
)
# evaluation settings
openflamingo_textvqa_evaluator = [dict(type='mmpretrain.VQAAcc')]
openflamingo_load_from = '/path/to/pretrained/weights' # noqa
|