Hugging Face's logo

xuan-luo
/
DiffSkip-Llama-3-8B-Instruct

Text Generation
Transformers
Safetensors
English
ddllama
conversational
custom_code
Model card Files
xet
Community
DiffSkip-Llama-3-8B-Instruct
File size: 4,043 Bytes 
ae99ccd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import sys
import logging

import datasets
from datasets import load_dataset
import torch
import transformers
from trl import SFTTrainer
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, BitsAndBytesConfig
from typing import Dict, List

logger = logging.getLogger(__name__)

"""
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 accelerate launch  --gradient_clipping=1.0  --multi_gpu   --num_processes=8   --num_machines=1   --mixed_precision=bf16   --zero_stage=3   sft.py
"""
###################
# Hyper-parameters
###################

training_config = {
    "bf16": True,
    "do_eval": False,
    "learning_rate": 1e-04,
    "log_level": "info",
    "logging_steps": 20,
    "logging_strategy": "steps",
    "lr_scheduler_type": "cosine",
    "num_train_epochs": 1,
    "max_steps": -1,
    "output_dir": "./ckpts",
    "overwrite_output_dir": True,
    "per_device_eval_batch_size": 8,
    "per_device_train_batch_size": 8,
    "remove_unused_columns": True,
    "save_steps": 1000,
    "save_total_limit": 1,
    "seed": 0,
    "gradient_checkpointing": True,
    "gradient_checkpointing_kwargs":{"use_reentrant": False},
    "gradient_accumulation_steps": 1,
    "warmup_ratio": 0.03,
    }
train_conf = TrainingArguments(**training_config)


###############
# Setup logging
###############
logging.basicConfig(
    format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
    datefmt="%Y-%m-%d %H:%M:%S",
    handlers=[logging.StreamHandler(sys.stdout)],
)
log_level = train_conf.get_process_log_level()
logger.setLevel(log_level)
datasets.utils.logging.set_verbosity(log_level)
transformers.utils.logging.set_verbosity(log_level)
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()

# Log on each process a small summary
logger.warning(
    f"Process rank: {train_conf.local_rank}, device: {train_conf.device}, n_gpu: {train_conf.n_gpu}"
    + f" distributed training: {bool(train_conf.local_rank != -1)}, 16-bits training: {train_conf.fp16}"
)
logger.info(f"Training/evaluation parameters {train_conf}")


################
# Model Loading
################

checkpoint_path = "./"
model_kwargs = dict(
    use_cache=False,
    trust_remote_code=True,
    attn_implementation="flash_attention_2",
    torch_dtype=torch.bfloat16,
    device_map=None
)
model = AutoModelForCausalLM.from_pretrained(checkpoint_path, **model_kwargs)
tokenizer = AutoTokenizer.from_pretrained(checkpoint_path)
tokenizer.model_max_length = 2048
tokenizer.pad_token = tokenizer.eos_token
tokenizer.pad_token_id = tokenizer.convert_tokens_to_ids(tokenizer.eos_token)
tokenizer.padding_side = 'right'


##################
# Data Processing
##################
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("allenai/tulu-v2-sft-mixture")
train_dataset = raw_dataset["train"]
column_names = list(train_dataset.features)

processed_dataset = train_dataset.map(
    apply_chat_template,
    fn_kwargs={"tokenizer": tokenizer},
    num_proc=64,
    remove_columns=column_names,
    desc="Applying chat template to train_sft",
)


###########
# Freeze Transformer
###########
for param in model.parameters():
    param.requires_grad = False

for name, param in model.named_parameters():
    if 'router' in name.lower():
        param.requires_grad = True

###########
# Training
###########
trainer = SFTTrainer(
    model=model,
    args=train_conf,
    peft_config=None,
    train_dataset=processed_dataset,
    eval_dataset=None,
    max_seq_length=2048,
    dataset_text_field="text",
    tokenizer=tokenizer,
    packing=False
)

train_result = trainer.train()
metrics = train_result.metrics
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()

# ############
# # Save model
# ############
trainer.save_model(train_conf.output_dir)