bf16_vs_fp8 / fastchat /train /train_flant5.py
zjasper666's picture
Upload folder using huggingface_hub
8655a4b verified
raw
history blame
15 kB
# Adapted from tatsu-lab@stanford_alpaca. Below is the original copyright:
# Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from collections import defaultdict
import copy
import os
from dataclasses import dataclass, field
import random
import json
import logging
import pathlib
from typing import Dict, Optional, Sequence
import torch
import torch.distributed as dist
import transformers
from torch.utils.data import Dataset
from transformers import Trainer, AddedToken
from fastchat.model.model_adapter import get_conversation_template
default_conversation = get_conversation_template("t5")
# TODO: import and use code from ../data/dataset.py
IGNORE_INDEX = -100
DEFAULT_PAD_TOKEN = "[PAD]"
DEFAULT_EOS_TOKEN = "</s>"
DEFAULT_BOS_TOKEN = "</s>"
DEFAULT_UNK_TOKEN = "</s>"
@dataclass
class ModelArguments:
model_name_or_path: Optional[str] = field(default="facebook/opt-125m")
@dataclass
class DataArguments:
data_path: str = field(
default=None, metadata={"help": "Path to the training data."}
)
lazy_preprocess: bool = False
num_data: int = -1
preprocessed_path: str = field(
default=None, metadata={"help": "Path to the preprocessed training data."}
)
@dataclass
class TrainingArguments(transformers.TrainingArguments):
cache_dir: Optional[str] = field(default=None)
optim: str = field(default="adamw_torch")
model_max_length: int = field(
default=2048,
metadata={
"help": "Maximum sequence length. Sequences will be right padded (and possibly truncated)."
},
)
def safe_save_model_for_hf_trainer(trainer: transformers.Trainer, output_dir: str):
"""Collects the state dict and dump to disk."""
state_dict = trainer.model.state_dict()
if trainer.args.should_save:
cpu_state_dict = {key: value.cpu() for key, value in state_dict.items()}
del state_dict
trainer._save(output_dir, state_dict=cpu_state_dict) # noqa
def smart_tokenizer_and_embedding_resize(
special_tokens_dict: Dict,
other_tokens,
tokenizer: transformers.PreTrainedTokenizer,
model: transformers.PreTrainedModel,
):
"""Resize tokenizer and embedding.
Note: This is the unoptimized version that may make your embedding size not be divisible by 64.
"""
num_new_tokens = tokenizer.add_special_tokens(special_tokens_dict)
for new_token in other_tokens:
num_new_tokens += tokenizer.add_tokens(AddedToken(new_token, normalized=False))
model.resize_token_embeddings(len(tokenizer))
if num_new_tokens > 0:
input_embeddings = model.get_input_embeddings().weight.data
output_embeddings = model.get_output_embeddings().weight.data
input_embeddings_avg = input_embeddings[:-num_new_tokens].mean(
dim=0, keepdim=True
)
output_embeddings_avg = output_embeddings[:-num_new_tokens].mean(
dim=0, keepdim=True
)
input_embeddings[-num_new_tokens:] = input_embeddings_avg
output_embeddings[-num_new_tokens:] = output_embeddings_avg
def _tokenize_fn(
strings: Sequence[str], tokenizer: transformers.PreTrainedTokenizer
) -> Dict:
"""Tokenize a list of strings."""
tokenized_list = [
tokenizer(
text,
return_tensors="pt",
padding="longest",
max_length=tokenizer.model_max_length,
truncation=True,
)
for text in strings
]
input_ids = labels = [tokenized.input_ids[0] for tokenized in tokenized_list]
input_ids_lens = labels_lens = [
tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item()
for tokenized in tokenized_list
]
return dict(
input_ids=input_ids,
labels=labels,
input_ids_lens=input_ids_lens,
labels_lens=labels_lens,
)
def _form_qa(
q_list,
a_list,
tokenized_conversation,
tokenized_lens,
speakers,
header_len,
max_length,
eos_id,
):
cur_idx = header_len
conv_len = len(tokenized_conversation)
for tokenized_len, speaker in zip(tokenized_lens, speakers):
if cur_idx >= conv_len:
break
if speaker == "gpt":
# truncate answer if it is too long
content_a = None
if tokenized_len > max_length:
content_a = tokenized_conversation[cur_idx : cur_idx + max_length]
else:
content_a = tokenized_conversation[cur_idx : cur_idx + tokenized_len]
content_a.append(eos_id)
a_list.append(content_a)
content_q = None
if cur_idx >= max_length:
content_q = tokenized_conversation[cur_idx - max_length : cur_idx]
else:
content_q = tokenized_conversation[:cur_idx]
content_q.append(eos_id)
q_list.append(content_q)
# asser the last token is actually a EOS for an answer
assert a_list[-1][-1] == eos_id, "Last Token is not EOS!"
cur_idx += tokenized_len
def _add_speaker_and_signal(header, source, get_conversation=True):
"""Add speaker and start/end signal on each round."""
BEGIN_SIGNAL = "### "
END_SIGNAL = "\n"
conversation = header
unknown_role = "unknown" # use default unknown role
roles = {
"human": default_conversation.roles[0], # human role
"gpt": default_conversation.roles[1], # gpt role
}
for i in range(len(source)):
sentence = source[i]
sentence_from = sentence["from"].lower()
# TODO(Dacheng): verify this is a good way to split sentences
if sentence_from == "human":
# if this is not the last sentence
if i != len(source) - 1:
next_sentence = source[i + 1]
sentence["value"] = (
BEGIN_SIGNAL
+ roles.get(sentence_from, unknown_role)
+ ": "
+ sentence["value"]
+ END_SIGNAL
+ BEGIN_SIGNAL
+ roles.get(next_sentence["from"].lower(), unknown_role)
+ ": "
)
else:
# if human is the last speaker, it does not contribute to an answer
pass
else:
sentence["value"] = sentence["value"] + END_SIGNAL
if get_conversation:
conversation += sentence["value"]
return conversation
def preprocess(
sources: Sequence[str],
tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
"""
Given a list of sources, each is a conversation list. This transform:
1. Add signal '### ' at the beginning each sentence, with end signal '\n';
2. Concatenate conversations together;
3. Tokenize the concatenated conversation;
4. Make a deepcopy as the target. Mask human words with IGNORE_INDEX.
"""
# add end signal and concatenate together
conversations = []
header = f"{default_conversation.system_message}\n\n"
for source in sources:
conversation = _add_speaker_and_signal(header, source, tokenizer)
conversations.append(conversation)
# TODO(Dacheng): This is related to whether the dataset has been truncated..
# Assume we get long conversations, don't pad, don't return tensor
tokenized_conversations = tokenizer(conversations, max_length=None)["input_ids"]
q_list = []
a_list = []
# count for EOS length
header_len = _tokenize_fn([header], tokenizer)["input_ids_lens"][0] - 1
from tqdm import tqdm
for tokenized_conversation, source in tqdm(zip(tokenized_conversations, sources)):
tokenized_sentence = _tokenize_fn([s["value"] for s in source], tokenizer)
tokenized_lens = tokenized_sentence["input_ids_lens"]
tokenized_lens = [l - 1 for l in tokenized_lens]
speakers = [sentence["from"] for sentence in source]
ids = tokenized_sentence["input_ids"]
_form_qa(
q_list,
a_list,
tokenized_conversation,
tokenized_lens,
speakers,
header_len,
tokenizer.model_max_length,
tokenizer.eos_token_id,
)
return dict(input_ids=q_list, labels=a_list)
class SupervisedDataset(Dataset):
"""Dataset for supervised fine-tuning."""
def __init__(
self,
data_path: str,
tokenizer: transformers.PreTrainedTokenizer,
preprocessed_path,
num_data,
):
super(SupervisedDataset, self).__init__()
# save to file
# Make sure only the first process is processing the dataset
if dist.get_rank() != 0:
dist.barrier()
self.preprocessed_path = preprocessed_path
if os.path.exists(self.preprocessed_path):
logging.warning("loading from preprocessed data")
with open(self.preprocessed_path, "r") as f:
data_dict = json.load(f)
if dist.get_rank() == 0:
dist.barrier()
else:
if not os.path.exists("preprocessed_data"):
os.mkdir("preprocessed_data")
assert dist.get_rank() == 0, "Only the first process should process"
logging.warning("Loading data...")
list_data_dict = json.load(open(data_path, "r"))
logging.warning("Formatting inputs...")
sources = []
sources = [example["conversations"] for example in list_data_dict]
data_dict = preprocess(sources, tokenizer)
json_data_dict = json.dumps(data_dict)
# Remember to close file to avoid concurrent r/w
with open(self.preprocessed_path, "w") as f:
f.write(json_data_dict)
# Release barrier
dist.barrier()
if num_data != -1:
data_dict["input_ids"] = data_dict["input_ids"][:num_data]
data_dict["labels"] = data_dict["labels"][:num_data]
# Shuffle data to see more conversations, if only train on partial data
temp = list(zip(data_dict["input_ids"], data_dict["labels"]))
random.shuffle(temp)
res1, res2 = zip(*temp)
data_dict["input_ids"], data_dict["labels"] = list(res1), list(res2)
# Dacheng: Get rid of short QA pair
self.input_ids = copy.deepcopy(data_dict["input_ids"])
self.labels = copy.deepcopy(data_dict["labels"])
length_arr = defaultdict(int)
for idx, (input, label) in enumerate(
zip(data_dict["input_ids"], data_dict["labels"])
):
length_arr[str(len(label) // 100)] += 1
if len(input) <= 5:
del_idx = self.input_ids.index(input)
self.input_ids.pop(del_idx)
self.labels.pop(del_idx)
if len(label) <= 5:
del_idx = self.labels.index(label)
self.input_ids.pop(del_idx)
self.labels.pop(del_idx)
for input, label in zip(self.input_ids, self.labels):
assert len(input) >= 5
assert len(label) >= 5
def __len__(self):
return len(self.input_ids)
def __getitem__(self, i) -> Dict[str, torch.Tensor]:
return dict(input_ids=self.input_ids[i], labels=self.labels[i])
@dataclass
class DataCollatorForSupervisedDataset(object):
"""Collate examples for supervised fine-tuning."""
tokenizer: transformers.PreTrainedTokenizer
def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
input_ids, labels = tuple(
[
torch.as_tensor(instance[key], dtype=torch.int64)
for instance in instances
]
for key in ("input_ids", "labels")
)
input_ids = torch.nn.utils.rnn.pad_sequence(
input_ids, batch_first=True, padding_value=self.tokenizer.pad_token_id
)
labels = torch.nn.utils.rnn.pad_sequence(
labels, batch_first=True, padding_value=IGNORE_INDEX
)
ret = dict(
input_ids=input_ids,
labels=labels,
attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
)
torch.set_printoptions(profile="full")
return ret
def make_supervised_data_module(
tokenizer: transformers.PreTrainedTokenizer, data_args
) -> Dict:
"""Make dataset and collator for supervised fine-tuning."""
dataset_cls = SupervisedDataset
train_dataset = dataset_cls(
tokenizer=tokenizer,
data_path=data_args.data_path,
preprocessed_path=data_args.preprocessed_path,
num_data=data_args.num_data,
)
data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
return dict(
train_dataset=train_dataset, eval_dataset=None, data_collator=data_collator
)
def train():
parser = transformers.HfArgumentParser(
(ModelArguments, DataArguments, TrainingArguments)
)
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
model = transformers.AutoModelForSeq2SeqLM.from_pretrained(
model_args.model_name_or_path,
cache_dir=training_args.cache_dir,
)
# Dacheng: Note we can only use T5Tokenizer, otherwise it will prepend
# a space before special tokens.
tokenizer = transformers.T5Tokenizer.from_pretrained(
model_args.model_name_or_path,
cache_dir=training_args.cache_dir,
model_max_length=training_args.model_max_length,
padding_side="right",
use_fast=False,
)
smart_tokenizer_and_embedding_resize(
special_tokens_dict=dict(pad_token=DEFAULT_PAD_TOKEN),
other_tokens=["<", "{", "\n", "}", "`", " ", "\\", "^", "\t"],
tokenizer=tokenizer,
model=model,
)
data_module = make_supervised_data_module(tokenizer=tokenizer, data_args=data_args)
trainer = Trainer(
model=model, tokenizer=tokenizer, args=training_args, **data_module
)
if list(pathlib.Path(training_args.output_dir).glob("checkpoint-*")):
trainer.train(resume_from_checkpoint=True)
else:
trainer.train()
trainer.save_state()
safe_save_model_for_hf_trainer(trainer=trainer, output_dir=training_args.output_dir)
if __name__ == "__main__":
train()