Delete finetune_qlora.py

finetune_qlora.py

@@ -1,158 +0,0 @@
-import os
-import torch
-from datasets import load_dataset
-from transformers import (
-    AutoModelForCausalLM,
-    AutoTokenizer,
-    BitsAndBytesConfig,
-    TrainingArguments,
-)
-from peft import LoraConfig, get_peft_model, PeftModel
-from trl import SFTTrainer
-
-# 1. Model and Tokenizer Configuration
-model_name = "./qwen2.5-coder-3b-instruct"  # Path to the local model directory
-dataset_name = "corrected_syntax_dataset.jsonl" # Path to your dataset
-new_model = "qwen2.5-coder-3b-instruct-syntax-finetuned" # Name for the fine-tuned model
-
-# 2. Q-LoRA Configuration
-lora_r = 64
-lora_alpha = 16
-lora_dropout = 0.1
-
-# 3. BitsAndBytes Configuration for 4-bit Quantization
-use_4bit = True
-bnb_4bit_compute_dtype = "float16"
-bnb_4bit_quant_type = "nf4"
-use_nested_quant = False
-
-# 4. Training Arguments
-output_dir = "./results"
-num_train_epochs = 1
-fp16 = False
-bf16 = True # Use bf16 for better performance on modern GPUs
-per_device_train_batch_size = 4
-per_device_eval_batch_size = 4
-gradient_accumulation_steps = 1
-gradient_checkpointing = True
-max_grad_norm = 0.3
-learning_rate = 2e-4
-weight_decay = 0.001
-optim = "paged_adamw_32bit"
-lr_scheduler_type = "cosine"
-max_steps = -1
-warmup_ratio = 0.03
-group_by_length = True
-save_steps = 25
-logging_steps = 5
-
-# 5. SFTTrainer Configuration
-max_seq_length = None
-packing = False
-device_map = {"": 0}
-
-# --- Script Execution ---
-
-# Load dataset
-# The dataset is in JSONL format, with each line having a "messages" field.
-dataset = load_dataset('json', data_files=dataset_name, split="train")
-
-# Load tokenizer and model with Q-LoRA configuration
-compute_dtype = getattr(torch, bnb_4bit_compute_dtype)
-
-bnb_config = BitsAndBytesConfig(
-    load_in_4bit=use_4bit,
-    bnb_4bit_quant_type=bnb_4bit_quant_type,
-    bnb_4bit_compute_dtype=compute_dtype,
-    bnb_4bit_use_double_quant=use_nested_quant,
-)
-
-# Load base model
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    quantization_config=bnb_config,
-    device_map=device_map
-)
-model.config.use_cache = False
-model.config.pretraining_tp = 1
-
-# Load tokenizer
-tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-tokenizer.pad_token = tokenizer.eos_token
-tokenizer.padding_side = "right"
-
-# PEFT Configuration
-peft_config = LoraConfig(
-    lora_alpha=lora_alpha,
-    lora_dropout=lora_dropout,
-    r=lora_r,
-    bias="none",
-    task_type="CAUSAL_LM",
-)
-
-# Set training parameters
-training_arguments = TrainingArguments(
-    output_dir=output_dir,
-    num_train_epochs=num_train_epochs,
-    per_device_train_batch_size=per_device_train_batch_size,
-    gradient_accumulation_steps=gradient_accumulation_steps,
-    optim=optim,
-    save_steps=save_steps,
-    logging_steps=logging_steps,
-    learning_rate=learning_rate,
-    weight_decay=weight_decay,
-    fp16=fp16,
-    bf16=bf16,
-    max_grad_norm=max_grad_norm,
-    max_steps=max_steps,
-    warmup_ratio=warmup_ratio,
-    group_by_length=group_by_length,
-    lr_scheduler_type=lr_scheduler_type,
-    report_to="tensorboard"
-)
-
-# Initialize the SFTTrainer
-trainer = SFTTrainer(
-    model=model,
-    train_dataset=dataset,
-    peft_config=peft_config,
-    dataset_text_field="messages",  # The field in your dataset that contains the text
-    max_seq_length=max_seq_length,
-    tokenizer=tokenizer,
-    args=training_arguments,
-    packing=packing,
-)
-
-# Train the model
-print("Starting model training...")
-trainer.train()
-print("Training complete.")
-
-# Save the fine-tuned model
-print(f"Saving fine-tuned model to {new_model}...")
-trainer.model.save_pretrained(new_model)
-print("Model saved.")
-
-# Merge and save the final model (optional)
-# Reload model in FP16 and merge it with LoRA weights
-base_model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    low_cpu_mem_usage=True,
-    return_dict=True,
-    torch_dtype=torch.float16,
-    device_map=device_map,
-)
-model = PeftModel.from_pretrained(base_model, new_model)
-model = model.merge_and_unload()
-
-# Reload tokenizer to save it
-tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-tokenizer.pad_token = tokenizer.eos_token
-tokenizer.padding_side = "right"
-
-# Save the merged model
-merged_model_dir = os.path.join(output_dir, "final_merged_model")
-model.save_pretrained(merged_model_dir, safe_serialization=True)
-tokenizer.save_pretrained(merged_model_dir)
-
-print(f"Final merged model saved to {merged_model_dir}")