88 lines
2.3 KiB
Python
88 lines
2.3 KiB
Python
_base_ = [
|
|
'../_base_/default_runtime.py',
|
|
]
|
|
|
|
# model settings
|
|
model = dict(
|
|
type='Otter',
|
|
tokenizer=dict(type='LlamaTokenizer', name_or_path='huggyllama/llama-7b'),
|
|
vision_encoder=dict(
|
|
type='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=(
|
|
'https://download.openmmlab.com/mmclassification/v0/clip/'
|
|
'vit-large-p14_clip-openai-pre_3rdparty_20230517-95e2af0b.pth'),
|
|
),
|
|
lang_encoder=dict(
|
|
base=dict(
|
|
type='AutoModelForCausalLM',
|
|
name_or_path='huggyllama/llama-7b',
|
|
local_files_only=True),
|
|
adapter=dict(
|
|
type='FlamingoLMAdapter',
|
|
vis_hidden_size=1024,
|
|
cross_attn_every_n_layers=4,
|
|
use_media_placement_augmentation=False,
|
|
only_attend_previous=True,
|
|
),
|
|
),
|
|
task='caption',
|
|
final_prompt_tmpl='<image>User:Please describe the image. GPT:<answer>',
|
|
generation_cfg=dict(
|
|
num_beams=3, max_new_tokens=24, no_repeat_ngram_size=3),
|
|
)
|
|
|
|
# data settings
|
|
data_preprocessor = dict(
|
|
type='MultiModalDataPreprocessor',
|
|
mean=[122.770938, 116.7460125, 104.09373615],
|
|
std=[68.5005327, 66.6321579, 70.32316305],
|
|
to_rgb=True,
|
|
)
|
|
|
|
test_pipeline = [
|
|
dict(type='LoadImageFromFile'),
|
|
dict(
|
|
type='ResizeEdge',
|
|
scale=224,
|
|
interpolation='bicubic',
|
|
backend='pillow'),
|
|
dict(type='CenterCrop', crop_size=(224, 224)),
|
|
dict(
|
|
type='PackInputs',
|
|
algorithm_keys=['gt_caption'],
|
|
meta_keys=['image_id'],
|
|
),
|
|
]
|
|
|
|
val_dataloader = dict(
|
|
batch_size=8,
|
|
num_workers=8,
|
|
dataset=dict(
|
|
type='COCOCaption',
|
|
data_root='data/coco',
|
|
ann_file='annotations/coco_karpathy_val.json',
|
|
pipeline=test_pipeline,
|
|
),
|
|
sampler=dict(type='DefaultSampler', shuffle=False),
|
|
persistent_workers=True,
|
|
)
|
|
|
|
val_evaluator = dict(
|
|
type='COCOCaption',
|
|
ann_file='data/coco/annotations/coco_karpathy_val_gt.json')
|
|
|
|
# If you want standard test, please manually configure the test dataset
|
|
test_dataloader = val_dataloader
|
|
test_evaluator = val_evaluator
|
|
|
|
# schedule settings
|
|
val_cfg = dict()
|
|
test_cfg = dict()
|