File size: 4,571 Bytes
c941f24
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import sys
import logging

import datasets
from datasets import load_dataset
from peft import LoraConfig
import torch
import transformers
from trl import SFTTrainer
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, BitsAndBytesConfig

training_config = {
    "bf16": True,
    "do_eval": False,
    "learning_rate": 5.0e-06,
    "log_level": "info",
    "logging_steps": 20,
    "logging_strategy": "steps",
    "lr_scheduler_type": "cosine",
    "num_train_epochs": 1,
    "max_steps": -1,
    "output_dir": "./instruct_chk_dir",
    "overwrite_output_dir": True,
    "per_device_eval_batch_size": 4,
    "per_device_train_batch_size": 4,
    "remove_unused_columns": True,
    "save_steps": 100,
    "save_total_limit": 1,
    "seed": 0,
    "gradient_checkpointing": True,
    "gradient_checkpointing_kwargs":{"use_reentrant": False},
    "gradient_accumulation_steps": 1,
    "warmup_ratio": 0.2,
    }

peft_config = {
    "r": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.05,
    "bias": "none",
    "task_type": "CAUSAL_LM",
    "target_modules": "all-linear",
    "modules_to_save": None,
}

config = {
    "max_len": 4096,
}

train_conf = TrainingArguments(**training_config)
peft_conf = LoraConfig(**peft_config)


# Model Init
checkpoint_path = "microsoft/Phi-3-mini-128k-instruct"
model_kwargs = dict(
    use_cache=False,
    trust_remote_code=True,
    attn_implementation="flash_attention_2",  # loading the model with flash-attenstion support
    torch_dtype=torch.bfloat16,
    #device_map=None
    device_map="sequential"
)
model = AutoModelForCausalLM.from_pretrained(checkpoint_path, **model_kwargs)
tokenizer = AutoTokenizer.from_pretrained(checkpoint_path)
tokenizer.model_max_length = config['max_len']
tokenizer.pad_token = tokenizer.unk_token  # use unk rather than eos token to prevent endless generation
tokenizer.pad_token_id = tokenizer.convert_tokens_to_ids(tokenizer.pad_token)
tokenizer.padding_side = 'right'

dataset_id = "BAAI/Infinity-Instruct"
raw_dataset = load_dataset(dataset_id, "0625", split="train")
dataset = raw_dataset.select(range(10000))


# Preproc dataset
def preproc(example, tokenizer):
    convo = example['conversations']
    for i, dic in enumerate(convo):
        dic['role'] = dic.pop('from')
        dic['content'] = dic.pop('value')
        if dic['role'] == 'gpt':
            dic['role'] = 'assistant'
        elif dic['role'] == 'human':
            dic['role'] = 'user'

    example['text'] = tokenizer.apply_chat_template(convo, tokenize=False, add_generation_prompt=False)
    return example

column_names = list(dataset.features)
train_dataset = dataset.map(
    preproc,
    fn_kwargs={"tokenizer": tokenizer},
    num_proc=10,
    remove_columns=column_names
)

# eval_dataset = dataset[9000:]
# eval_dataset = eval_dataset.map(
#     preproc,
#     fn_kwargs={"tokenizer": tokenizer},
#     num_proc=10,
#     remove_columns=column_names
# )


# Train Model
trainer = SFTTrainer(
    model=model,
    args=train_conf,
    peft_config=peft_conf,
    train_dataset=train_dataset,
    #eval_dataset=eval_dataset,
    max_seq_length=config['max_len'],
    dataset_text_field="text",
    tokenizer=tokenizer,
    packing=True
)
train_result = trainer.train()
metrics = train_result.metrics
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()


# Eval Model
tokenizer.padding_side = 'left'
metrics = trainer.evaluate()
metrics["eval_samples"] = len(processed_test_dataset)
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)


# Save model
trainer.save_model(train_conf.output_dir)

# def apply_chat_template(
#     example,
#     tokenizer,
# ):
#     messages = example["messages"]
#     example["text"] = tokenizer.apply_chat_template(
#         messages, tokenize=False, add_generation_prompt=False)
#     return example

# raw_dataset = load_dataset("HuggingFaceH4/ultrachat_200k")
# train_dataset = raw_dataset["train_sft"].select(range(10000))
# test_dataset = raw_dataset["test_sft"].select(range(1000))
# column_names = list(train_dataset.features)

# processed_train_dataset = train_dataset.map(
#     apply_chat_template,
#     fn_kwargs={"tokenizer": tokenizer},
#     num_proc=10,
#     remove_columns=column_names,
#     desc="Applying chat template to train_sft",
# )

# processed_test_dataset = test_dataset.map(
#     apply_chat_template,
#     fn_kwargs={"tokenizer": tokenizer},
#     num_proc=10,
#     remove_columns=column_names,
#     desc="Applying chat template to test_sft",
# )