File size: 5,417 Bytes
cd6952b |
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 |
# -*- coding: utf-8 -*-
"""PreFinetuningForRunPod.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1LtsUCcWfL2VpWLJXVkE5076XX5k3PTyg
"""
# IMPORTS
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import datasets
from datasets import load_dataset
from trl import SFTTrainer
from peft import PeftConfig, PeftModel
from multiprocessing import cpu_count
from peft import prepare_model_for_kbit_training
from peft import LoraConfig, get_peft_model
import bitsandbytes as bnb
import transformers
# LOGIN TO HUB FOR MODEL DEPLOYMENT
# from huggingface_hub import notebook_login
# notebook_login()
# LOADING THE TOKENIZER
model_id = "mistralai/Mistral-7B-v0.1"
print("-----------------------------loading tokenizer-----------------------------------------------------------")
tokenizer = AutoTokenizer.from_pretrained(model_id, add_eos_token=True)
print("-----------------------------loading dataset-----------------------------------------------------------")
# LOAD DATA FROM HUGGINFACE
data = load_dataset("gbharti/finance-alpaca", split='train')
# PREPARE DATA FOR FINE-TUNING
def generate_prompt(data_point):
"""Gen. input text based on a prompt, task instruction, (context info.), and answer
:param data_point: dict: Data point
:return: dict: tokenzed prompt
"""
# Samples with additional context into.
if data_point['input']:
text = 'Below is an instruction that describes a task, paired with an input that provides' \
' further context. Write a response that appropriately completes the request.\n\n'
text += f'### Instruction:\n{data_point["instruction"]}\n\n'
text += f'### Input:\n{data_point["input"]}\n\n'
text += f'### Response:\n{data_point["output"]}'
# Without context
else:
text = 'Below is an instruction that describes a task. Write a response that ' \
'appropriately completes the request.\n\n'
text += f'### Instruction:\n{data_point["instruction"]}\n\n'
text += f'### Response:\n{data_point["output"]}'
return text
print("-----------------------------Preparing dataset for fine-tuning-----------------------------------------------------------")
prompt = [generate_prompt(data_point) for data_point in data]
data = data.add_column("prompt", prompt);
data = data.map(lambda sample: tokenizer(sample["prompt"]),num_proc=cpu_count(), batched=True)
# data = data.remove_columns(['Context', 'Response'])
data = data.shuffle(seed=1234)
data = data.train_test_split(test_size=0.1)
train_data = data["train"]
test_data = data["test"]
# LOADING MODEL IN N(4, 8.....) BIT
bnb_config = BitsAndBytesConfig(
load_in_4bit = True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
d_map = {"": torch.cuda.current_device()} if torch.cuda.is_available() else None
print("-----------------------------loading model-----------------------------------------------------------")
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype="auto",
use_cache=False, # set to False as we're going to use gradient checkpointing
quantization_config=bnb_config,
device_map=d_map
)
def find_all_linear_names(model):
cls = bnb.nn.Linear4bit #if args.bits == 4 else (bnb.nn.Linear8bitLt if args.bits == 8 else torch.nn.Linear)
lora_module_names = set()
for name, module in model.named_modules():
if isinstance(module, cls):
names = name.split('.')
lora_module_names.add(names[0] if len(names) == 1 else names[-1])
if 'lm_head' in lora_module_names: # needed for 16-bit
lora_module_names.remove('lm_head')
return list(lora_module_names)
modules = find_all_linear_names(model)
lora_config = LoraConfig(
r=8, # Number of quantization levels
lora_alpha=32, # Hyperparameter for LoRA
target_modules = modules, # Modules to apply LoRA to
lora_dropout=0.05, # Dropout probability
bias="none", # Type of bias
task_type="CAUSAL_LM" # Task type (in this case, Causal Language Modeling)
)
model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, lora_config)
# trainable, total = model.get_nb_trainable_parameters()
# print(f"Trainable: {trainable} | total: {total} | Percentage: {trainable/total*100:.4f}%")
tokenizer.pad_token = tokenizer.eos_token
torch.cuda.empty_cache()
trainer = SFTTrainer(
model=model,
train_dataset=train_data,
eval_dataset=test_data,
dataset_text_field="prompt",
peft_config=lora_config,
args=transformers.TrainingArguments(
per_device_train_batch_size=1,
gradient_accumulation_steps=4,
gradient_checkpointing=True,
fp16=True,
warmup_steps=0.03,
max_steps=10,
learning_rate=2e-4,
logging_steps=1,
output_dir="outputs",
optim="paged_adamw_8bit",
save_strategy="epoch",
),
data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
print("-----------------------------finetuning starts------------------------------------------------------------")
model.config.use_cache = False
trainer.train()
|