Spaces:
Paused
Paused
File size: 3,508 Bytes
1804a7a |
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 |
import torch
from unsloth import FastLanguageModel
from trl import SFTTrainer
from transformers import TrainingArguments
from datasets import load_dataset
import os
import psutil
import builtins
# --- PATCH FOR UNSLOTH BUG ---
builtins.psutil = psutil
# --- CONFIG ---
PROJECT_ROOT = "/content/drive/MyDrive/ProjectA_Backup"
DATASET_PATH = os.path.join(PROJECT_ROOT, "src/data/final_finetune_dataset.jsonl")
SAVE_PATH = os.path.join(PROJECT_ROOT, "models/project_a_14b_finetuned")
OUTPUT_DIR = "outputs"
max_seq_length = 2048
load_in_4bit = True
def train_model():
print(f"🚀 Initializing Training...")
print(f"📂 Data: {DATASET_PATH}")
if not os.path.exists(DATASET_PATH):
raise FileNotFoundError("❌ Dataset not found! Run 'src/tools/merge_datasets.py' first.")
# 1. Load Base Model
model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "Qwen/Qwen2.5-Coder-14B-Instruct",
max_seq_length = max_seq_length,
dtype = None,
load_in_4bit = load_in_4bit,
)
# 2. Add LoRA
model = FastLanguageModel.get_peft_model(
model,
r = 16,
target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
"gate_proj", "up_proj", "down_proj",],
lora_alpha = 16,
lora_dropout = 0,
bias = "none",
use_gradient_checkpointing = "unsloth",
random_state = 3407,
use_rslora = False,
loftq_config = None,
)
# 3. Load & Clean Data
dataset = load_dataset("json", data_files=DATASET_PATH, split="train")
def formatting_prompts_func(examples):
conversations = examples["messages"]
texts = []
for conv in conversations:
if not conv or not isinstance(conv, list):
texts.append("")
continue
try:
# Apply Chat Template
text = tokenizer.apply_chat_template(conv, tokenize=False, add_generation_prompt=False)
texts.append(text)
except:
texts.append("")
return { "text" : texts }
dataset = dataset.map(formatting_prompts_func, batched = True)
dataset = dataset.filter(lambda x: x["text"] != "")
print(f"📊 Training on {len(dataset)} samples.")
# 4. Train
print("🏋️♂️ Starting Training Cycle...")
trainer = SFTTrainer(
model = model,
tokenizer = tokenizer,
train_dataset = dataset,
dataset_text_field = "text",
max_seq_length = max_seq_length,
dataset_num_proc = 2,
packing = False,
args = TrainingArguments(
per_device_train_batch_size = 2,
gradient_accumulation_steps = 4,
warmup_steps = 5,
num_train_epochs = 3, # Train for 3 full loops
learning_rate = 2e-4,
fp16 = not torch.cuda.is_bf16_supported(),
bf16 = torch.cuda.is_bf16_supported(),
logging_steps = 1,
optim = "adamw_8bit",
weight_decay = 0.01,
lr_scheduler_type = "linear",
seed = 3407,
output_dir = OUTPUT_DIR,
report_to = "none",
),
)
trainer.train()
# 5. Save Adapters Locally
print(f"💾 Saving to local folder 'lora_model'...")
model.save_pretrained("lora_model")
tokenizer.save_pretrained("lora_model")
print("✅ Training Complete.")
if __name__ == "__main__":
train_model() |