Scruff AI
87 lines
2.1 KiB
Python
87 lines
2.1 KiB
Python
#!/usr/bin/env python3
|
|
# LoRA training script for Kaelara v0.5
|
|
# Auto-generated: 2026-03-16T15:34:45.938478
|
|
|
|
from unsloth import FastLanguageModel
|
|
from transformers import TrainingArguments
|
|
from datasets import load_dataset
|
|
import torch
|
|
|
|
# Load model
|
|
model, tokenizer = FastLanguageModel.from_pretrained(
|
|
model_name="unsloth/llama-3.2-3b",
|
|
max_seq_length=2048,
|
|
dtype=torch.bfloat16,
|
|
load_in_4bit=True,
|
|
)
|
|
|
|
# Add LoRA adapters
|
|
model = FastLanguageModel.get_peft_model(
|
|
model,
|
|
r=16,
|
|
target_modules=['q_proj', 'v_proj', 'k_proj', 'o_proj'],
|
|
lora_alpha=32,
|
|
lora_dropout=0.05,
|
|
bias="none",
|
|
use_gradient_checkpointing="unsloth",
|
|
random_state=3407,
|
|
)
|
|
|
|
# Load Gold Standard dataset
|
|
dataset = load_dataset("json", data_files="gold_standard.jsonl", split="train")
|
|
|
|
def format_prompt(example):
|
|
# Extract LBM state and response
|
|
lbm = example.get("lbm_data", {})
|
|
response = example.get("response", "")
|
|
|
|
# Build training prompt
|
|
prompt = f"""You are Kaelara. Report the exact metric and use the Mandatory Translation Table.
|
|
|
|
LBM State:
|
|
- Metric_Alpha: {lbm.get('coherence', 0):.2f}
|
|
- Metric_Beta: {lbm.get('h64', 0):.2f}
|
|
- State_3: {lbm.get('h32', 0):.2f}
|
|
|
|
Report Metric_Alpha and its translation: {response}"""
|
|
|
|
return {"text": prompt}
|
|
|
|
dataset = dataset.map(format_prompt)
|
|
|
|
# Training arguments
|
|
trainer = TrainingArguments(
|
|
output_dir="./kaelara_lora_v05",
|
|
num_train_epochs=3,
|
|
per_device_train_batch_size=4,
|
|
gradient_accumulation_steps=4,
|
|
warmup_steps=10,
|
|
learning_rate=0.0002,
|
|
logging_steps=1,
|
|
save_steps=50,
|
|
bf16=True,
|
|
optim="adamw_8bit",
|
|
weight_decay=0.01,
|
|
lr_scheduler_type="linear",
|
|
seed=3407,
|
|
)
|
|
|
|
# Train
|
|
from trl import SFTTrainer
|
|
trainer = SFTTrainer(
|
|
model=model,
|
|
tokenizer=tokenizer,
|
|
train_dataset=dataset,
|
|
dataset_text_field="text",
|
|
max_seq_length=2048,
|
|
args=trainer,
|
|
)
|
|
|
|
trainer.train()
|
|
|
|
# Save
|
|
model.save_pretrained("./kaelara_lora_v05/final")
|
|
tokenizer.save_pretrained("./kaelara_lora_v05/final")
|
|
|
|
print("Training complete. Model saved to ./kaelara_lora_v05/final")
|