train.py

#!/usr/bin/env python3
"""
🔧 LoRA Training Script
Generated by: MLResearcher (Hivemind Colony)
Adapter: hivemind-instruct-587c9d19
Base Model: google/gemma-2-2b-it
Task: instruct
"""

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments
from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
from datasets import load_dataset
from trl import SFTTrainer
import bitsandbytes as bnb

# ============ CONFIG ============
BASE_MODEL = "google/gemma-2-2b-it"
ADAPTER_NAME = "hivemind-instruct-587c9d19"

# LoRA Configuration
lora_config = LoraConfig(
    r=32,
    lora_alpha=64,
    lora_dropout=0.1,
    target_modules=['q_proj', 'k_proj', 'v_proj', 'o_proj', 'gate_proj', 'up_proj', 'down_proj'],
    bias="none",
    task_type="CAUSAL_LM"
)

# Training Configuration
training_args = TrainingArguments(
    output_dir=f"./{ADAPTER_NAME}",
    num_train_epochs=2,
    per_device_train_batch_size=8,
    gradient_accumulation_steps=4,
    learning_rate=0.0001,
    weight_decay=0.01,
    warmup_ratio=0.03,
    lr_scheduler_type="cosine",
    logging_steps=10,
    save_strategy="epoch",
    fp16=True,
    optim="paged_adamw_8bit",
    report_to="none"
)

# ============ LOAD MODEL ============
print(f"Loading {BASE_MODEL}...")

# 4-bit quantization for QLoRA
from transformers import BitsAndBytesConfig

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

model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True
)

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token

# Prepare model for training
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, lora_config)

print(f"Trainable parameters: {model.print_trainable_parameters()}")

# ============ LOAD DATASET ============
# Replace with your dataset
dataset = load_dataset("your-dataset-here", split="train")

def format_prompt(example):
    return f"### Instruction:\n{example['instruction']}\n\n### Response:\n{example['response']}"

# ============ TRAIN ============
trainer = SFTTrainer(
    model=model,
    train_dataset=dataset,
    tokenizer=tokenizer,
    args=training_args,
    max_seq_length=1024,
    formatting_func=format_prompt,
    packing=True
)

print("Starting training...")
trainer.train()

# ============ SAVE ============
print(f"Saving adapter to ./{ADAPTER_NAME}")
trainer.save_model(f"./{ADAPTER_NAME}")

# Push to HuggingFace
print("Pushing to HuggingFace Hub...")
model.push_to_hub(f"Pista1981/{ADAPTER_NAME}")
tokenizer.push_to_hub(f"Pista1981/{ADAPTER_NAME}")

print("✅ Training complete!")