This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct optimized for laser cleaning customer service interactions. It was developed for RustBustersHSV, a laser cleaning and resurfacing company in Huntsville, Alabama.
Model Details
Model type: Fine-tuned Llama-3.2-3B-Instruct with LoRA
Finetuning approach: Parameter-efficient fine-tuning with Low-Rank Adaptation (LoRA)
Intended Uses & Limitations
Intended Uses
This model is designed to:
Answer customer inquiries about laser cleaning services
Provide detailed information about RustBustersHSV's services
Help customers understand the laser cleaning process
Address common concerns and objections
Guide customers toward requesting a free quote
Limitations
This model:
Is not designed to provide specific pricing information
Should not be used for non-laser cleaning domains without further adaptation
Is limited to English language responses
May not have expertise in very technical aspects beyond its training data
Should be monitored when deployed in a customer-facing environment
Training Procedure
Training Data
The model was fine-tuned on 3,000 synthetic QA pairs categorized into:
General inquiries about laser cleaning
Service-specific questions
Logistics and location information
Process details
Concerns and objections
Customer experience
Technical aspects
All QA pairs were generated using templates and variations designed to mimic real customer service interactions for a laser cleaning business.
Training Hyperparameters
LoRA Configuration:
r: 8
lora_alpha: 16
lora_dropout: 0.1
bias: "none"
target_modules: ["q_proj", "v_proj"]
task_type: "CAUSAL_LM"
Training Hyperparameters:
Batch size: 1
Learning rate: 2e-5
Optimizer: AdamW
Sequence length: 128
Epochs: 3
Warmup ratio: 0.1
Early stopping patience: 3
Framework Versions
Transformers 4.38.0+
PyTorch 2.0+
PEFT for LoRA fine-tuning
Uses
This model is intended to be used as a customer service assistant for a laser cleaning business. It can be integrated into:
Live chat on a company website
Customer inquiry response systems
Internal knowledge base for employees
Training materials for new customer service representatives
Bias, Risks, and Limitations
The model is specialized for laser cleaning customer service and may:
Emphasize the benefits of laser cleaning over alternative methods
Always attempt to guide customers toward requesting quotes
Have limited knowledge outside the laser cleaning domain
Not understand or respond accurately to highly technical queries outside its training
Training Performance
The model was trained using the AdamW optimizer with a linear learning rate scheduler and warmup. Early stopping was used to prevent overfitting.
Environmental Impact
The model was fine-tuned using parameter-efficient LoRA techniques to minimize computational resources
Training was performed on TPU to maximize efficiency
How to Use
You can use this model with the Transformers pipeline:
python
1from peft import PeftModel, PeftConfig
2from transformers import AutoModelForCausalLM, AutoTokenizer
34# Load base model5model_name ="meta-llama/Llama-3.2-3B-Instruct"6tokenizer = AutoTokenizer.from_pretrained(model_name)7model = AutoModelForCausalLM.from_pretrained(model_name)89# Load adapter10adapter_path ="RustBustersHSV/Llama-3.2-3B-Instruct-RustBusters"11model = PeftModel.from_pretrained(model, adapter_path)1213# Format your prompt appropriately14system_prompt ="""You are Lloyd, the first point of contact for customers of Rustbusters. Please be warm and friendly and offer actionable information. Rustbusters is a laser cleaning company that specializes in removing rust, paint, and other contaminants using advanced laser technology. Our services include industrial cleaning, restoration, paint removal, and surface preparation."""15user_prompt ="What is laser cleaning and how does it work?"1617prompt =f"<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{user_prompt}<|im_end|>\n<|im_start|>assistant\n"1819# Generate response20inputs = tokenizer(prompt, return_tensors="pt")21outputs = model.generate(**inputs, max_length=512, temperature=0.7, top_p=0.9)22response = tokenizer.decode(outputs[0], skip_special_tokens=True)23print(response)
Community and Contributions
This model is maintained by RustBustersHSV. For questions or issues, please contact [contact information].
Citation
If you use this model in research, please cite:
@misc{rustbustersllama32,
author = {RustBustersHSV},
title = {RustBustersHSV-Llama-3.2-3B-Instruct-LoRA},
year = {2025},
publisher = {Hugging Face},
journal = {Hugging Face model repository},
howpublished = {\url{https://huggingface.co/RustBustersHSV/Llama-3.2-3B-Instruct-RustBusters}}
}
