File size: 11,158 Bytes
1bbd0a2 |
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 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 |
import torch
from mmengine.dataset import DefaultSampler
from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
LoggerHook, ParamSchedulerHook)
from transformers import (AutoModelForCausalLM, AutoTokenizer,
BitsAndBytesConfig,
CLIPImageProcessor, CLIPVisionModel,
SiglipVisionModel, SiglipImageProcessor, AutoProcessor)
from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR, LinearLR
from peft import LoraConfig
from torch.optim import AdamW
from xtuner.dataset import LLaVADataset, CambrianDataset, ConcatDataset
from xtuner.dataset.collate_fns import default_collate_fn
from xtuner.dataset.map_fns import llava_map_fn, cambrian_map_fn, template_map_fn_factory
from xtuner.dataset.samplers import LengthGroupedSampler
from xtuner.engine import DatasetInfoHook, EvaluateChatHook
from xtuner.model import LLaVAModel, PikaModel
from xtuner.utils import PROMPT_TEMPLATE
#######################################################################
# PART 1 Settings #
#######################################################################
# Model
llm_name_or_path = 'meta-llama/Meta-Llama-3-8B-Instruct'
visual_encoder_name_or_path = 'google/siglip-so400m-patch14-384'
pretrained_pth = '/data/wenhao/projects/xtuner/work_dirs/final_new_p/projector'
prompt_template = PROMPT_TEMPLATE.llama3_chat
max_length = 4096
size = 378
batch_size = 1 # per_device
accumulative_counts = 32
lr = 4e-5
dataloader_num_workers = 0
max_epochs = 1
optim_type = AdamW
betas = (0.9, 0.999)
weight_decay = 0
max_norm = 1 # grad clip
warmup_ratio = 0.03
sf = False
# Save
save_steps = 200
save_total_limit = 2 # Maximum checkpoints to keep (-1 means unlimited)
#######################################################################
# PART 2 Model & Tokenizer & Image Processor #
#######################################################################
tokenizer = dict(
type=AutoTokenizer.from_pretrained,
pretrained_model_name_or_path=llm_name_or_path,
trust_remote_code=True,
padding_side='right')
image_processor = dict(
type=CLIPImageProcessor.from_pretrained,
pretrained_model_name_or_path='laion/CLIP-ViT-bigG-14-laion2B-39B-b160k',
trust_remote_code=True,
size=size,
crop_size=size)
model = dict(
type=PikaModel,
sf=sf,
freeze_llm=True,
freeze_visual_encoder=False,
pretrained_pth=pretrained_pth,
llm=dict(
type=AutoModelForCausalLM.from_pretrained,
pretrained_model_name_or_path=llm_name_or_path,
trust_remote_code=True,
torch_dtype=torch.float16,),
visual_encoder=dict(
type=SiglipVisionModel.from_pretrained,
pretrained_model_name_or_path=visual_encoder_name_or_path))
#######################################################################
# PART 3 Dataset & Dataloader #
#######################################################################
m3it_data_root = '/data/wenhao/projects/xtuner/data/m3it/'
m3it_data_path = m3it_data_root + 'm3it.jsonl'
m3it_image_folder = m3it_data_root
m3it_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/m3it/pre_token_llama31',
image_folder=m3it_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
chatterbox_data_root = '/data/wenhao/projects/xtuner/data/ChatterBox/'
chatterbox_data_path = chatterbox_data_root + 'chatterbox_76k.jsonl'
chatterbox_image_folder = chatterbox_data_root
chatterbox_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/ChatterBox/pre_token_llama31',
image_folder=chatterbox_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
laion_data_root = '/data/wenhao/projects/xtuner/data/LLaVA-Pretrain/'
laion_data_path = laion_data_root + 'laion_558k.jsonl'
laion_image_folder = laion_data_root
laion_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/LLaVA-Pretrain/pre_token_llama31',
image_folder=laion_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
face_data_root = '/data/wenhao/projects/xtuner/data/FaceCaption-15M/'
face_data_path = face_data_root + 'FaceCaption-100K.jsonl'
face_image_folder = face_data_root + 'full_data'
face_processed_text_folder = face_data_root + 'pre_token_llama3'
face_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder=face_processed_text_folder,
image_folder=face_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
cost_data_root = '/data/wenhao/projects/xtuner/data/COST/'
cost_data_path = cost_data_root + 'cost.jsonl'
cost_image_folder = cost_data_root
cost_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/COST/pre_token_llama31',
# tokenizer=tokenizer,
# data_path='/data/wenhao/projects/xtuner/data/COST/cost.jsonl',
image_folder=cost_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
sharept_data_root = '/data/wenhao/projects/xtuner/data/ShareGPT4V/'
sharept_data_path = sharept_data_root + 'sharegpt4v_pt.jsonl'
sharept_image_folder = '/data/wenhao/projects/xtuner/data/'
sharept_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/ShareGPT4V/pre_token_llama31',
# tokenizer=tokenizer,
# data_path='/data/wenhao/projects/xtuner/data/ShareGPT4V/sharegpt4v_pt.jsonl',
image_folder=sharept_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
llavaone_data_root = '/data/wenhao/projects/xtuner/data/onevision/'
llavaone_data_path = '/data/wenhao/projects/xtuner/data/LLaVA-OneVision-Data/llava_onevision.jsonl'
llavaone_image_folder = llavaone_data_root + 'images'
llavaone_dataset = dict(
type=CambrianDataset,
offline_processed_text_folder='/data/wenhao/projects/xtuner/data/onevision/pre_token_llama31',
# tokenizer=tokenizer,
# data_path='/data/wenhao/projects/xtuner/data/LLaVA-OneVision-Data/llava_onevision.jsonl',
image_folder=llavaone_image_folder,
image_processor=image_processor,
dataset_map_fn=cambrian_map_fn,
template_map_fn=dict(
type=template_map_fn_factory, template=prompt_template),
max_length=max_length,
pad_image_to_square=True)
train_dataset = dict(
type=ConcatDataset,
datasets=[m3it_dataset, chatterbox_dataset, laion_dataset, face_dataset, cost_dataset, sharept_dataset, llavaone_dataset],
)
train_dataloader = dict(
batch_size=batch_size,
num_workers=dataloader_num_workers,
dataset=train_dataset,
sampler=dict(type=DefaultSampler, shuffle=True),
collate_fn=dict(type=default_collate_fn))
#######################################################################
# PART 4 Scheduler & Optimizer #
#######################################################################
# optimizer
optim_wrapper = dict(
type=AmpOptimWrapper,
optimizer=dict(
type=optim_type, lr=lr, betas=betas, weight_decay=weight_decay),
clip_grad=dict(max_norm=max_norm, error_if_nonfinite=False),
accumulative_counts=accumulative_counts,
loss_scale='dynamic',
dtype='float16')
# learning policy
# More information: https://github.com/open-mmlab/mmengine/blob/main/docs/en/tutorials/param_scheduler.md # noqa: E501
param_scheduler = [
dict(
type=LinearLR,
start_factor=1e-5,
by_epoch=True,
begin=0,
end=warmup_ratio * max_epochs,
convert_to_iter_based=True),
dict(
type=CosineAnnealingLR,
eta_min=0.0,
by_epoch=True,
begin=warmup_ratio * max_epochs,
T_max=max_epochs,
convert_to_iter_based=True)
]
# train, val, test setting
train_cfg = dict(by_epoch=True, max_epochs=max_epochs, val_interval=1)
#######################################################################
# PART 5 Runtime #
#######################################################################
# Evaluate the generation performance during the training
evaluation_freq = 100
SYSTEM = ''
evaluation_images = 'https://llava-vl.github.io/static/images/view.jpg'
evaluation_inputs = ['请描述一下这张照片', 'Please describe this picture']
# Log the dialogue periodically during the training process, optional
custom_hooks = [
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
image_processor=image_processor,
every_n_iters=evaluation_freq,
evaluation_inputs=evaluation_inputs,
evaluation_images=evaluation_images,
system=SYSTEM,
prompt_template=prompt_template)
]
# configure default hooks
default_hooks = dict(
# record the time of every iteration.
timer=dict(type=IterTimerHook),
# print log every 100 iterations.
logger=dict(type=LoggerHook, interval=10),
# enable the parameter scheduler.
param_scheduler=dict(type=ParamSchedulerHook),
# save checkpoint per epoch.
checkpoint=dict(
type=CheckpointHook,
by_epoch=False,
interval=save_steps,
max_keep_ckpts=save_total_limit),
# set sampler seed in distributed evrionment.
sampler_seed=dict(type=DistSamplerSeedHook),
)
# configure environment
env_cfg = dict(
# whether to enable cudnn benchmark
cudnn_benchmark=False,
# set multi process parameters
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
# set distributed parameters
dist_cfg=dict(backend='nccl'),
)
# set visualizer
visualizer = None
# set log level
log_level = 'INFO'
# load from which checkpoint
load_from = None
# whether to resume training from the loaded checkpoint
resume = False
# Defaults to use random seed and disable `deterministic`
randomness = dict(seed=None, deterministic=False) |