CodeLlama-7b-Instruct-SQL

Model creator: Codellama
Original model: CodeLlama-7b-Instruct-hf

Description

This repo contains LoRA finetuned model files for CodeLlama-7b-Instruct-hf.

Prompt template: CodeLlama

<s>[INST] <<SYS>> {system_msg} <</SYS>> \n {prompt} [/INST]

Instruction format

In order to leverage instruction fine-tuning, your prompt should be surrounded by [INST] and [/INST] tokens. The system message should be surrounded by <<SYS>> and <</SYS>> tokens. The very first instruction should begin with a sentence id. The next instructions should not. The assistant generation will be ended by the end-of-sentence token id.

For example:

from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch

model = "machinists/CodeLlama-7b-Instruct-SQL"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
    device_map="auto",
)

table_schema = "CREATE TABLE head (age INTEGER)"

question = "How many heads of the departments are older than 56 ?"

system_msg = f"<<SYS>> Generate a correct SQL query from the following database schema. \n {table_schema} <</SYS>>"

prompt = f"<s>[INST] {system_msg} \n \n {question} [/INST]"

sequences = pipeline(
    prompt,
    max_length=1000,
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
)
for seq in sequences:
    print(f"Result: {seq['generated_text']}")

Finetuning

Finetuning technique involving rank of matrices to reduce computational and memory requirements. Read more: LoRA Hugging Face article

Epoch - 05

Dataset Name - b-mc2/sql-create-context

No. of records - 78.6k

Model Loading - bf16

Finetuning Technique - Using LoRA

MaxSeqLength - 1024

Mixed Precision Training - tf32

Hardware and Software

Training Hardware: 1 X Nvidia A100 80GB GPU

The Machinists Team

Manish Kumar, Aakash Sarin

ReadMe References

https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf

https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-AWQ