faizack/text-to-sql-dpo

Text-to-SQL DPO Model

A Direct Preference Optimization (DPO) fine-tuned LLaMA-3-8B model specialized for text-to-SQL generation tasks. This model has been trained using LoRA (Low-Rank Adaptation) for efficient parameter-efficient fine-tuning.

Model Details

Model Description

This model is a fine-tuned version of LLaMA-3-8B using Direct Preference Optimization (DPO) specifically for text-to-SQL tasks. It has been trained on preference pairs to generate accurate SQL queries from natural language descriptions.

• Developed by: faizack
• Model type: Causal Language Model with LoRA adapter
• Language(s) (NLP): English
• License: Apache 2.0 (inherited from base model)
• Finetuned from model: unsloth/llama-3-8B

Model Sources

• Repository: Text-to-SQL DPO Repository
• Base Model: unsloth/llama-3-8B

Uses

Direct Use

This model is designed for generating SQL queries from natural language descriptions. It can be used for:

• Converting natural language questions to SQL queries
• Database query generation
• Text-to-SQL applications
• Database interaction interfaces

Example Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

# Load the base model and tokenizer
base_model = "unsloth/llama-3-8B"
tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.float16)

# Load the LoRA adapter
model = PeftModel.from_pretrained(model, "faizack/text-to-sql-dpo")

# Generate SQL query
prompt = "Show me all users from the customers table"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Out-of-Scope Use

This model should not be used for:

• General-purpose text generation beyond SQL queries
• Generating malicious or harmful SQL queries
• Database operations without proper validation
• Production use without proper testing and validation

Bias, Risks, and Limitations

Limitations

• The model is specialized for SQL generation and may not perform well on other tasks
• Generated SQL queries should be validated before execution
• Performance may vary depending on database schema complexity
• The model may generate queries that are syntactically correct but logically incorrect

Recommendations

• Always validate generated SQL queries before execution
• Test the model on your specific database schema
• Use appropriate safety measures when executing generated queries
• Consider the model's limitations when integrating into production systems

How to Get Started with the Model

Installation

pip install transformers peft torch

Quick Start

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load model and adapter
base_model = "unsloth/llama-3-8B"
model = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(model, "faizack/text-to-sql-dpo")
tokenizer = AutoTokenizer.from_pretrained(base_model)

# Generate SQL
prompt = "Find all orders placed in the last 30 days"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=150, temperature=0.1)
sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(sql_query)

Training Details

Training Data

The model was trained on the zerolink/zsql-sqlite-dpo dataset, which contains preference pairs for text-to-SQL tasks.

Training Procedure

Training Hyperparameters

• Training regime: DPO (Direct Preference Optimization)
• Epochs: 6
• Batch size: 2
• Gradient accumulation: 32
• Learning rate: 5e-5
• LoRA rank: 16
• LoRA alpha: 16
• LoRA dropout: 0.05
• Target modules: q_proj, v_proj

Training Infrastructure

• Base model: unsloth/llama-3-8B
• Framework: PEFT (Parameter-Efficient Fine-Tuning)
• Training method: LoRA (Low-Rank Adaptation)
• Total steps: 120
• Steps per epoch: 3660

Technical Specifications

Model Architecture

• Base architecture: LLaMA-3-8B
• Adapter type: LoRA
• Trainable parameters: ~16M (LoRA adapter only)
• Total parameters: ~8B (base model + adapter)

Compute Infrastructure

• Hardware: GPU-based training
• Framework versions:
  • PEFT: 0.17.1
  • Transformers: 4.56.2
  • PyTorch: Compatible with CUDA

Citation

If you use this model in your research, please cite:

@misc{text-to-sql-dpo-2024,
  title={Text-to-SQL DPO Model},
  author={faizack},
  year={2024},
  url={https://huggingface.co/faizack/text-to-sql-dpo}
}

Model Card Contact

For questions or issues related to this model, please contact the model author or open an issue in the repository.

Framework versions

• PEFT 0.17.1
• Transformers 4.56.2