295 lines
10 KiB
Python
295 lines
10 KiB
Python
# Adopted 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 dataclasses import dataclass, field
|
|
import json
|
|
import pathlib
|
|
from typing import Dict, Optional
|
|
import os
|
|
import numpy as np
|
|
import torch
|
|
from torch.utils.data import Dataset
|
|
import transformers
|
|
from transformers import Trainer
|
|
from transformers.trainer_pt_utils import LabelSmoother
|
|
|
|
from toolbench.tool_conversation import SeparatorStyle
|
|
from toolbench.model.model_adapter import get_conversation_template
|
|
from toolbench.train.llama_condense_monkey_patch import replace_llama_with_condense
|
|
|
|
|
|
IGNORE_TOKEN_ID = LabelSmoother.ignore_index
|
|
torch.set_printoptions(profile="full")
|
|
|
|
@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."}
|
|
)
|
|
eval_data_path: str = field(
|
|
default=None, metadata={"help": "Path to the training data."}
|
|
)
|
|
conv_template: str = field(
|
|
default=None, metadata={"help": "Template used to format the training data."}
|
|
)
|
|
lazy_preprocess: bool = False
|
|
|
|
|
|
@dataclass
|
|
class TrainingArguments(transformers.TrainingArguments):
|
|
cache_dir: Optional[str] = field(default=None)
|
|
optim: str = field(default="adamw_torch")
|
|
source_model_max_length: int = field(
|
|
default=2048,
|
|
metadata={
|
|
"help": "Original maximum sequence length. Sequences will be right padded (and possibly truncated)."
|
|
},
|
|
)
|
|
model_max_length: int = field(
|
|
default=8192,
|
|
metadata={
|
|
"help": "Expanded maximum sequence length. Sequences will be right padded (and possibly truncated)."
|
|
},
|
|
)
|
|
|
|
|
|
local_rank = None
|
|
|
|
|
|
def rank0_print(*args):
|
|
if local_rank == 0:
|
|
print(*args)
|
|
|
|
|
|
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)
|
|
|
|
|
|
def preprocess(
|
|
sources,
|
|
tokenizer: transformers.PreTrainedTokenizer,
|
|
template: str="tool-llama"
|
|
) -> Dict:
|
|
conv = get_conversation_template(template)
|
|
if template == "tool-llama":
|
|
roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
|
|
elif template == "tool-llama-single-round" or template == "tool-llama-multi-rounds":
|
|
roles = {"system": conv.roles[0], "user": conv.roles[1], "function": conv.roles[2], "assistant": conv.roles[3]}
|
|
|
|
# Apply prompt templates
|
|
conversations = []
|
|
for i, source in enumerate(sources):
|
|
conv.messages = []
|
|
for j, sentence in enumerate(source):
|
|
role = roles[sentence["from"]]
|
|
conv.append_message(role, sentence["value"])
|
|
conversations.append(conv.get_prompt())
|
|
|
|
# Tokenize conversations
|
|
input_ids = tokenizer(
|
|
conversations,
|
|
return_tensors="pt",
|
|
padding="max_length",
|
|
max_length=tokenizer.model_max_length,
|
|
truncation=True,
|
|
).input_ids
|
|
targets = input_ids.clone()
|
|
|
|
# Mask targets. Only compute loss on the assistant outputs.
|
|
sep = conv.sep + conv.roles[-1] + ": "
|
|
for conversation, target in zip(conversations, targets):
|
|
total_len = int(target.ne(tokenizer.pad_token_id).sum())
|
|
turns = conversation.split(conv.sep2)
|
|
cur_len = 1
|
|
target[:cur_len] = IGNORE_TOKEN_ID
|
|
for i, turn in enumerate(turns):
|
|
if turn == "":
|
|
continue
|
|
turn_len = len(tokenizer(turn).input_ids)
|
|
|
|
parts = turn.split(sep)
|
|
|
|
# only train on the last assistant reply, treat the history chat as instruction
|
|
prefix = parts[:-1]
|
|
instruction = ""
|
|
for part in prefix:
|
|
instruction += part
|
|
instruction += sep
|
|
|
|
# "-2" is hardcoded for the LLaMA tokenizer to make the offset correct.
|
|
instruction_len = len(tokenizer(instruction).input_ids) - 2
|
|
|
|
# Ignore the user instructions
|
|
target[cur_len : cur_len + instruction_len] = IGNORE_TOKEN_ID
|
|
cur_len += turn_len
|
|
|
|
target[cur_len:] = IGNORE_TOKEN_ID
|
|
|
|
if False: # Inspect and check the correctness of masking
|
|
z = target.clone()
|
|
z = torch.where(z == IGNORE_TOKEN_ID, tokenizer.unk_token_id, z)
|
|
rank0_print(tokenizer.decode(z))
|
|
|
|
if cur_len < tokenizer.model_max_length:
|
|
if cur_len != total_len:
|
|
target[:] = IGNORE_TOKEN_ID
|
|
rank0_print(
|
|
f"WARNING: tokenization mismatch: {cur_len} vs. {total_len}."
|
|
f" (ignored)"
|
|
)
|
|
return dict(
|
|
input_ids=input_ids,
|
|
labels=targets,
|
|
attention_mask=input_ids.ne(tokenizer.pad_token_id),
|
|
)
|
|
|
|
|
|
class SupervisedDataset(Dataset):
|
|
"""Dataset for supervised fine-tuning."""
|
|
|
|
def __init__(self, raw_data, tokenizer: transformers.PreTrainedTokenizer, template="tool-llama"):
|
|
super(SupervisedDataset, self).__init__()
|
|
|
|
rank0_print("Formatting inputs...")
|
|
sources = [example["conversations"] for example in raw_data]
|
|
self.template = template
|
|
data_dict = preprocess(sources, tokenizer, self.template)
|
|
self.input_ids = data_dict["input_ids"]
|
|
self.labels = data_dict["labels"]
|
|
self.attention_mask = data_dict["attention_mask"]
|
|
|
|
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],
|
|
attention_mask=self.attention_mask[i],
|
|
)
|
|
|
|
|
|
class LazySupervisedDataset(Dataset):
|
|
"""Dataset for supervised fine-tuning."""
|
|
|
|
def __init__(self, raw_data, tokenizer: transformers.PreTrainedTokenizer, template="tool-llama"):
|
|
super(LazySupervisedDataset, self).__init__()
|
|
self.tokenizer = tokenizer
|
|
|
|
rank0_print("Formatting inputs...Skip in lazy mode")
|
|
self.tokenizer = tokenizer
|
|
self.raw_data = raw_data
|
|
self.cached_data_dict = {}
|
|
self.template = template
|
|
|
|
def __len__(self):
|
|
return len(self.raw_data)
|
|
|
|
def __getitem__(self, i) -> Dict[str, torch.Tensor]:
|
|
if i in self.cached_data_dict:
|
|
return self.cached_data_dict[i]
|
|
|
|
ret = preprocess([self.raw_data[i]["conversations"]], self.tokenizer, self.template)
|
|
ret = dict(
|
|
input_ids=ret["input_ids"][0],
|
|
labels=ret["labels"][0],
|
|
attention_mask=ret["attention_mask"][0],
|
|
)
|
|
self.cached_data_dict[i] = ret
|
|
|
|
return ret
|
|
|
|
|
|
def make_supervised_data_module(
|
|
tokenizer: transformers.PreTrainedTokenizer, data_args
|
|
) -> Dict:
|
|
"""Make dataset and collator for supervised fine-tuning."""
|
|
dataset_cls = (
|
|
LazySupervisedDataset if data_args.lazy_preprocess else SupervisedDataset
|
|
)
|
|
rank0_print("Loading data...")
|
|
raw_data = json.load(open(data_args.data_path, "r"))
|
|
if data_args.eval_data_path is not None:
|
|
train_raw_data = raw_data
|
|
eval_raw_data = json.load(open(data_args.eval_data_path, "r"))
|
|
else:
|
|
# Split train/test
|
|
perm = np.random.permutation(len(raw_data))
|
|
split = int(len(perm) * 0.98)
|
|
train_indices = perm[:split]
|
|
eval_indices = perm[split:]
|
|
train_raw_data = [raw_data[i] for i in train_indices]
|
|
eval_raw_data = [raw_data[i] for i in eval_indices]
|
|
rank0_print(f"#train {len(train_raw_data)}, #eval {len(eval_raw_data)}")
|
|
train_dataset = dataset_cls(train_raw_data, tokenizer=tokenizer, template=data_args.conv_template)
|
|
eval_dataset = dataset_cls(eval_raw_data, tokenizer=tokenizer, template=data_args.conv_template)
|
|
return dict(train_dataset=train_dataset, eval_dataset=eval_dataset)
|
|
|
|
|
|
def train():
|
|
global local_rank
|
|
|
|
parser = transformers.HfArgumentParser(
|
|
(ModelArguments, DataArguments, TrainingArguments)
|
|
)
|
|
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
|
if training_args.source_model_max_length < training_args.model_max_length:
|
|
condense_ratio = int(training_args.model_max_length/training_args.source_model_max_length)
|
|
# ratio = N means the sequence length is expanded by N, remember to change the model_max_length to 8192 (2048 * ratio) for ratio = 4
|
|
replace_llama_with_condense(ratio=condense_ratio)
|
|
local_rank = training_args.local_rank
|
|
tokenizer = transformers.AutoTokenizer.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,
|
|
)
|
|
tokenizer.pad_token = tokenizer.unk_token
|
|
|
|
data_module = make_supervised_data_module(tokenizer=tokenizer, data_args=data_args)
|
|
world_size = int(os.environ.get("WORLD_SIZE", 1))
|
|
ddp = world_size != 1
|
|
device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)} if ddp else None
|
|
model = transformers.AutoModelForCausalLM.from_pretrained(
|
|
model_args.model_name_or_path,
|
|
cache_dir=training_args.cache_dir,
|
|
device_map=device_map
|
|
)
|
|
model.config.use_cache = False
|
|
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()
|