v1

xtuner/configs/deepspeed/deepspeed_zero2.json

@@ -11,7 +11,7 @@
     "overlap_comm": true,
     "reduce_scatter": true
   },
-  "fp16": {
+  "bf16": {
     "enabled": true,
     "initial_scale_power": 16
   }

xtuner/configs/llava/llava_llama2_7b_chat_clip_vit_large_p14_e1.py

@@ -0,0 +1,176 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import torch
+from mmengine.dataset import DefaultSampler
+from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
+                            LoggerHook, ParamSchedulerHook)
+from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR
+from torch.optim import AdamW
+from transformers import (AutoModelForCausalLM, AutoTokenizer,
+                          CLIPImageProcessor, CLIPVisionModel)
+
+from xtuner.dataset import LLaVADataset
+from xtuner.dataset.collate_fns import default_collate_fn
+from xtuner.dataset.map_fns import llava_map_fn, template_map_fn_factory
+from xtuner.engine import DatasetInfoHook, EvaluateChatHook
+from xtuner.model import LLaVAModel
+from xtuner.utils import PROMPT_TEMPLATE
+
+#######################################################################
+#                          PART 1  Settings                           #
+#######################################################################
+# Model
+llm_name_or_path = 'meta-llama/Llama-2-7b-hf'
+visual_encoder_name_or_path = 'openai/clip-vit-large-patch14'
+
+# Data
+data_path = './data/llava_data/LLaVA-Pretrain/blip_laion_cc_sbu_558k.json'
+image_folder = './data/llava_data/LLaVA-Pretrain/images'
+prompt_template = PROMPT_TEMPLATE.llama2_chat
+max_length = 2048
+
+# Scheduler & Optimizer
+batch_size = 32  # per_device
+accumulative_counts = 1
+dataloader_num_workers = 4
+max_epochs = 1
+optim_type = AdamW
+lr = 1e-3
+betas = (0.9, 0.999)
+weight_decay = 0
+max_norm = 1  # grad clip
+
+# Evaluate the generation performance during the training
+evaluation_freq = 500
+SYSTEM = ''
+evaluation_images = ',/xtuner/xtuner/configs/llava/cloud.png'
+evaluation_inputs = ['请描述一下这张照片', 'Please describe this picture']
+
+#######################################################################
+#                 PART 2  Model & Tokenizer & Processor               #
+#######################################################################
+tokenizer = dict(
+    type=AutoTokenizer.from_pretrained,
+    pretrained_model_name_or_path=llm_name_or_path,
+    trust_remote_code=True,
+    padding_side='right')
+
+processor = dict(
+    type=CLIPImageProcessor.from_pretrained,
+    pretrained_model_name_or_path=visual_encoder_name_or_path,
+    trust_remote_code=True)
+
+model = dict(
+    type=LLaVAModel,
+    freeze_llm=True,
+    freeze_visual_encoder=True,
+    llm=dict(
+        type=AutoModelForCausalLM.from_pretrained,
+        pretrained_model_name_or_path=llm_name_or_path,
+        trust_remote_code=True,
+        torch_dtype=torch.float32),
+    visual_encoder=dict(
+        type=CLIPVisionModel.from_pretrained,
+        pretrained_model_name_or_path=visual_encoder_name_or_path))
+
+#######################################################################
+#                      PART 3  Dataset & Dataloader                   #
+#######################################################################
+llava_dataset = dict(
+    type=LLaVADataset,
+    data_path=data_path,
+    image_folder=image_folder,
+    tokenizer=tokenizer,
+    processor=processor,
+    dataset_map_fn=llava_map_fn,
+    template_map_fn=dict(
+        type=template_map_fn_factory, template=prompt_template),
+    max_length=max_length,
+    image_aspect_ratio='pad')
+
+train_dataloader = dict(
+    batch_size=batch_size,
+    num_workers=dataloader_num_workers,
+    dataset=llava_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=CosineAnnealingLR,
+    eta_min=lr * 0.1,
+    by_epoch=True,
+    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                           #
+#######################################################################
+# Log the dialogue periodically during the training process, optional
+custom_hooks = [
+    dict(type=DatasetInfoHook, tokenizer=tokenizer),
+    dict(
+        type=EvaluateChatHook,
+        tokenizer=tokenizer,
+        processor=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, interval=1),
+    # 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)

xtuner/configs/llava/llava_llama2_7b_chat_clip_vit_large_p14_e1_ft.py

@@ -0,0 +1,178 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import torch
+from mmengine.dataset import DefaultSampler
+from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
+                            LoggerHook, ParamSchedulerHook)
+from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR
+from torch.optim import AdamW
+from transformers import (AutoModelForCausalLM, AutoTokenizer,
+                          CLIPImageProcessor, CLIPVisionModel)
+
+from xtuner.dataset import LLaVADataset
+from xtuner.dataset.collate_fns import default_collate_fn
+from xtuner.dataset.map_fns import llava_map_fn, template_map_fn_factory
+from xtuner.engine import DatasetInfoHook, EvaluateChatHook
+from xtuner.model import LLaVAModel
+from xtuner.utils import PROMPT_TEMPLATE
+
+#######################################################################
+#                          PART 1  Settings                           #
+#######################################################################
+# Model
+llm_name_or_path = 'meta-llama/Llama-2-7b-hf'
+visual_encoder_name_or_path = 'openai/clip-vit-large-patch14'
+projector_pth = './work_dirs/llava_llama2_7b_chat_clip_vit_large_p14_e1/epoch_1.pth'  # noqa: E501
+
+# Data
+data_path = './data/llava_data/LLaVA-Instruct-150K/llava_v1_5_mix665k.json'
+image_folder = './data/llava_data/llava_images'
+prompt_template = PROMPT_TEMPLATE.llama2_chat
+max_length = 2048
+
+# Scheduler & Optimizer
+batch_size = 16  # per_device
+accumulative_counts = 1
+dataloader_num_workers = 4
+max_epochs = 1
+optim_type = AdamW
+lr = 2e-5
+betas = (0.9, 0.999)
+weight_decay = 0
+max_norm = 1  # grad clip
+
+# Evaluate the generation performance during the training
+evaluation_freq = 500
+SYSTEM = ''
+evaluation_images = './xtuner/configs/llava/cloud.png'
+evaluation_inputs = ['请描述一下这张照片', 'Please describe this picture']
+
+#######################################################################
+#                 PART 2  Model & Tokenizer & Processor               #
+#######################################################################
+tokenizer = dict(
+    type=AutoTokenizer.from_pretrained,
+    pretrained_model_name_or_path=llm_name_or_path,
+    trust_remote_code=True,
+    padding_side='right')
+
+processor = dict(
+    type=CLIPImageProcessor.from_pretrained,
+    pretrained_model_name_or_path=visual_encoder_name_or_path,
+    trust_remote_code=True)
+
+model = dict(
+    type=LLaVAModel,
+    freeze_llm=False,
+    freeze_visual_encoder=True,
+    projector_pth=projector_pth,
+    llm=dict(
+        type=AutoModelForCausalLM.from_pretrained,
+        pretrained_model_name_or_path=llm_name_or_path,
+        trust_remote_code=True,
+        torch_dtype=torch.float32),
+    visual_encoder=dict(
+        type=CLIPVisionModel.from_pretrained,
+        pretrained_model_name_or_path=visual_encoder_name_or_path))
+
+#######################################################################
+#                      PART 3  Dataset & Dataloader                   #
+#######################################################################
+llava_dataset = dict(
+    type=LLaVADataset,
+    data_path=data_path,
+    image_folder=image_folder,
+    tokenizer=tokenizer,
+    processor=processor,
+    dataset_map_fn=llava_map_fn,
+    template_map_fn=dict(
+        type=template_map_fn_factory, template=prompt_template),
+    max_length=max_length,
+    image_aspect_ratio='pad')
+
+train_dataloader = dict(
+    batch_size=batch_size,
+    num_workers=dataloader_num_workers,
+    dataset=llava_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=CosineAnnealingLR,
+    eta_min=lr * 0.1,
+    by_epoch=True,
+    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                           #
+#######################################################################
+# Log the dialogue periodically during the training process, optional
+custom_hooks = [
+    dict(type=DatasetInfoHook, tokenizer=tokenizer),
+    dict(
+        type=EvaluateChatHook,
+        tokenizer=tokenizer,
+        processor=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, interval=1),
+    # 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)

xtuner/dataset/__init__.py

@@ -1,10 +1,12 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from .concat_dataset import ConcatDataset
 from .huggingface import process_hf_dataset
+from .llava import LLaVADataset
 from .modelscope import process_ms_dataset
 from .moss_sft import MOSSSFTDataset
+from .utils import expand2square
 
 __all__ = [
     'process_hf_dataset', 'ConcatDataset', 'MOSSSFTDataset',
-    'process_ms_dataset'
+    'process_ms_dataset', 'LLaVADataset', 'expand2square'
 ]