github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-05-06 07:28:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

[format] applied code formatting on changed files in pull request 3300 (#3302)

Browse Source 
Co-authored-by: github-actions <github-actions@github.com>
This commit is contained in:
github-actions[bot] 2023-03-29 09:28:24 +08:00 committed by GitHub
parent 31c78f2be3
commit cb413ccf28
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 36 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
applications/Chat/examples/train_prompts.py
View File

@ -1,19 +1,20 @@
import argparse import argparse
import pandas as pd import pandas as pd
import torch import torch
import torch.distributed as dist import torch.distributed as dist
from coati.models.bloom import BLOOMActor, BLOOMRM, BLOOMCritic from coati.dataset import DataCollatorForSupervisedDataset, PromptDataset, SupervisedDataset
from coati.models.gpt import GPTActor, GPTRM, GPTCritic from coati.models.bloom import BLOOMRM, BLOOMActor, BLOOMCritic
from coati.models.opt import OPTActor, OPTRM, OPTCritic from coati.models.gpt import GPTRM, GPTActor, GPTCritic
from coati.models.llama import LlamaActor, LlamaRM, LlamaCritic from coati.models.llama import LlamaActor, LlamaCritic, LlamaRM
from coati.models.opt import OPTRM, OPTActor, OPTCritic
from coati.trainer import PPOTrainer from coati.trainer import PPOTrainer
from coati.trainer.strategies import ColossalAIStrategy, DDPStrategy, NaiveStrategy from coati.trainer.strategies import ColossalAIStrategy, DDPStrategy, NaiveStrategy
from coati.utils import prepare_llama_tokenizer_and_embedding
from torch.optim import Adam from torch.optim import Adam
from torch.utils.data import DataLoader from torch.utils.data import DataLoader
from torch.utils.data.distributed import DistributedSampler from torch.utils.data.distributed import DistributedSampler
from transformers import AutoTokenizer, BloomTokenizerFast, LlamaTokenizer, GPT2Tokenizer from transformers import AutoTokenizer, BloomTokenizerFast, GPT2Tokenizer, LlamaTokenizer
from coati.dataset import SupervisedDataset, DataCollatorForSupervisedDataset, PromptDataset
from coati.utils import prepare_llama_tokenizer_and_embedding
from colossalai.nn.optimizer import HybridAdam from colossalai.nn.optimizer import HybridAdam
@ -62,7 +63,6 @@ def main(args):
else: else:
raise ValueError(f'Unsupported reward model "{rm_model_name}"') raise ValueError(f'Unsupported reward model "{rm_model_name}"')
if args.rm_path is not None: if args.rm_path is not None:
reward_model.load_state_dict(state_dict) reward_model.load_state_dict(state_dict)
@ -132,12 +132,19 @@ def main(args):
prompt_dataset = PromptDataset(tokenizer=tokenizer, data_path=args.prompt_path, max_datasets_size=16384) prompt_dataset = PromptDataset(tokenizer=tokenizer, data_path=args.prompt_path, max_datasets_size=16384)
if dist.is_initialized() and dist.get_world_size() > 1: if dist.is_initialized() and dist.get_world_size() > 1:
prompt_sampler = DistributedSampler(prompt_dataset, shuffle=True, seed=42, drop_last=True) prompt_sampler = DistributedSampler(prompt_dataset, shuffle=True, seed=42, drop_last=True)
prompt_dataloader = DataLoader(prompt_dataset, shuffle=(prompt_sampler is None), sampler=prompt_sampler, batch_size=args.train_batch_size) prompt_dataloader = DataLoader(prompt_dataset,
shuffle=(prompt_sampler is None),
sampler=prompt_sampler,
batch_size=args.train_batch_size)
pretrain_dataset = SupervisedDataset(tokenizer=tokenizer, data_path=args.pretrain_dataset, max_datasets_size=16384) pretrain_dataset = SupervisedDataset(tokenizer=tokenizer, data_path=args.pretrain_dataset, max_datasets_size=16384)
if dist.is_initialized() and dist.get_world_size() > 1: if dist.is_initialized() and dist.get_world_size() > 1:
pretrain_sampler = DistributedSampler(pretrain_dataset, shuffle=True, seed=42, drop_last=True) pretrain_sampler = DistributedSampler(pretrain_dataset, shuffle=True, seed=42, drop_last=True)
pretrain_dataloader = DataLoader(pretrain_dataset, shuffle=(pretrain_sampler is None), sampler=pretrain_sampler, batch_size=args.ptx_batch_size, collate_fn=data_collator) pretrain_dataloader = DataLoader(pretrain_dataset,
shuffle=(pretrain_sampler is None),
sampler=pretrain_sampler,
batch_size=args.ptx_batch_size,
collate_fn=data_collator)
def tokenize_fn(texts): def tokenize_fn(texts):
# MUST padding to max length to ensure inputs of all ranks have the same length # MUST padding to max length to ensure inputs of all ranks have the same length
@ -145,8 +152,7 @@ def main(args):
batch = tokenizer(texts, return_tensors='pt', max_length=96, padding='max_length', truncation=True) batch = tokenizer(texts, return_tensors='pt', max_length=96, padding='max_length', truncation=True)
return {k: v.to(torch.cuda.current_device()) for k, v in batch.items()} return {k: v.to(torch.cuda.current_device()) for k, v in batch.items()}
(actor, actor_optim), (critic, critic_optim) = strategy.prepare( (actor, actor_optim), (critic, critic_optim) = strategy.prepare((actor, actor_optim), (critic, critic_optim))
(actor, actor_optim), (critic, critic_optim))
# configure trainer # configure trainer
trainer = PPOTrainer( trainer = PPOTrainer(
@ -192,7 +198,8 @@ if __name__ == '__main__':
parser.add_argument('--pretrain_dataset', type=str, default=None, help='path to the pretrained dataset') parser.add_argument('--pretrain_dataset', type=str, default=None, help='path to the pretrained dataset')
parser.add_argument('--strategy', parser.add_argument('--strategy',
choices=['naive', 'ddp', 'colossalai_gemini', 'colossalai_zero2'], choices=['naive', 'ddp', 'colossalai_gemini', 'colossalai_zero2'],
default='naive', help='strategy to use') default='naive',
help='strategy to use')
parser.add_argument('--model', default='gpt2', choices=['gpt2', 'bloom', 'opt', 'llama']) parser.add_argument('--model', default='gpt2', choices=['gpt2', 'bloom', 'opt', 'llama'])
parser.add_argument('--pretrain', type=str, default=None) parser.add_argument('--pretrain', type=str, default=None)
parser.add_argument('--rm_model', default=None, choices=['gpt2', 'bloom', 'opt', 'llama']) parser.add_argument('--rm_model', default=None, choices=['gpt2', 'bloom', 'opt', 'llama'])