---
language: en
license: apache-2.0
library_name: peft
tags:
- llama
- llama-3
- construction
- building-regulations
- lora
- custom construction industry dataset
---

# LLAMA3.1-8B-Instruct-Construction

This is a fine-tuned version of LLAMA3.1-8B-Instruct optimized for construction industry and building regulations knowledge.

## Model Details

- **Base Model:** meta-llama/Llama-3.1-8B-Instruct
- **Fine-tuning Method:** LoRA (Low-Rank Adaptation)
- **Training Data:** Custom dataset focusing on construction industry standards, building regulations, and safety requirements
- **Usage:** This model is designed to answer questions about building codes, construction best practices, and regulatory compliance

## Example Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel, PeftConfig
import torch
import re

# Load the adapter configuration
config = PeftConfig.from_pretrained("SamuelJaja/llama-3.1-8b-instruct-construction-lora-a100")

# Load base model with quantization
bnb_config = BitsAndBytesConfig(load_in_8bit=True)
model = AutoModelForCausalLM.from_pretrained(
    config.base_model_name_or_path,
    quantization_config=bnb_config,
    device_map="auto"
)

# Load LoRA adapter
model = PeftModel.from_pretrained(model, "SamuelJaja/llama-3.1-8b-instruct-construction-lora-a100")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
tokenizer.pad_token = tokenizer.eos_token

# Clean response function
def clean_response(text):
    return re.sub(r'\[/?INST\]', '', text).strip()

# Generate text
def generate_response(prompt, temperature=0.1, max_tokens=256):
    # Format properly
    if not prompt.startswith("[INST]"):
        formatted_prompt = f"[INST] {prompt} [/INST]"
    else:
        formatted_prompt = prompt
        
    inputs = tokenizer(formatted_prompt, return_tensors="pt").to("cuda")
    
    outputs = model.generate(
        input_ids=inputs.input_ids,
        attention_mask=inputs.attention_mask,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=0.9,
        do_sample=False
    )
    
    full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Remove prompt from output
    if formatted_prompt in full_response:
        response = full_response.replace(formatted_prompt, "").strip()
    else:
        response = full_response
        
    # Clean any remaining instruction tags
    response = clean_response(response)
    
    return response
    
# Example use
question = "What are the main requirements for fire safety in commercial buildings?"
answer = generate_response(question)
print(answer)
```