Fine-Tune Gemma 2 2b باستخدام Transformers و qLoRA (الجزء الثاني)

!nvidia-smi

from kaggle_secrets import UserSecretsClient
from huggingface_hub import login

user_secrets = UserSecretsClient()
HF_Read = user_secrets.get_secret("HF-Read")
login(token=HF_Read)

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

model_id = "google/gemma-2b-it"
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map={"": 0})
tokenizer = AutoTokenizer.from_pretrained(model_id, add_eos_token=True)

from datasets import load_dataset
dataset = load_dataset("arbml/CIDAR", split="train")

def generate_prompt(data_point):
    prefix_text = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n'
    text = f"<start_of_turn>user {prefix_text} {data_point['instruction']} <end_of_turn>\n<start_of_turn>model {data_point['output']} <end_of_turn>"
    return text

# إضافة عمود 'prompt' إلى مجموعة البيانات
text_column = [generate_prompt(data_point) for data_point in dataset]
dataset = dataset.add_column("prompt", text_column)

# تقسيم مجموعة البيانات إلى تدريب واختبار
dataset = dataset.shuffle(seed=1234)
dataset = dataset.map(lambda samples: tokenizer(samples["prompt"]), batched=True)

dataset = dataset.train_test_split(test_size=0.2)
train_data = dataset["train"]
test_data = dataset["test"]

from peft import LoraConfig, PeftModel, prepare_model_for_kbit_training, get_peft_model

model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)

modules = find_all_linear_names(model)  # احصل على الوحدات لتطبيق LoRA عليها

lora_config = LoraConfig(
    r=64,
    lora_alpha=32,
    target_modules=modules,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM"
)

model = get_peft_model(model, lora_config)

import transformers
from trl import SFTTrainer

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,
        max_steps=10,
        learning_rate=2e-4,
        output_dir="outputs",
        optim="paged_adamw_8bit",
        save_strategy="epoch",
    ),
    data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False),
)

trainer.train()

base_model = AutoModelForCausalLM.from_pretrained(
    model_id,
    low_cpu_mem_usage=True,
    return_dict=True,
    torch_dtype=torch.float16,
    device_map={"": 0},
)
merged_model = PeftModel.from_pretrained(base_model, model)
merged_model = merged_model.merge_and_unload()

# حفظ النموذج المدمج
merged_model.save_pretrained("merged_model", safe_serialization=True)
tokenizer.save_pretrained("merged_model")

HF_write = user_secrets.get_secret("HF_WRITE")
login(token=HF_write)

merged_model.push_to_hub("Ruqiya/gemma2_2b_fine_tuned_arabic_dataset", use_auth_token=True)
tokenizer.push_to_hub("Ruqiya/gemma2_2b_fine_tuned_arabic_dataset", use_auth_token=True)