Phi-nut-Butter-Codebagel-v1

Model Details

Model Name: Phi-nut-Butter-Codebagel-v1 Base Model: microsoft/Phi-3-mini-128k-instruct
Fine-tuning Method: Supervised Fine-Tuning (SFT)
Dataset: Code Bagel
Training Data: 75,000 randomly selected rows from Code Bagel dataset
Training Duration: 23 hours
Hardware: Nvidia RTX A4500
Epochs: 3

Training Procedure

This model was fine-tuned to provide better instructions on code.

The training was conducted using PEFT and SFTTrainer on the Code Bagel dataset. Training was completed in 3 epochs over a span of 23 hours on an Nvidia A4500 GPU.

Intended Use

This model is designed to improve instruction-following capabilities, particularly for code-related tasks.

Getting Started

Instruct Template

<|system|>
{system_message} <|end|>
<|user|>
{Prompt) <|end|>
<|assistant|>

Transfromers

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_name_or_path = "thesven/Phi-nut-Butter-Codebagel-v1"

# BitsAndBytesConfig for loading the model in 4-bit precision
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype="float16",
)

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name_or_path,
    device_map="auto",
    trust_remote_code=False,
    revision="main",
    quantization_config=bnb_config
)
model.pad_token = model.config.eos_token_id

prompt_template = '''
<|system|>
You are an expert developer. Please help me with any coding questions.<|end|>
<|user|>
Create a function to get the total sum from an array of ints.<|end|>
<|assistant|>
'''

input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
output = model.generate(inputs=input_ids, temperature=0.1, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=256)

generated_text = tokenizer.decode(output[0, len(input_ids[0]):], skip_special_tokens=True)
print(generated_text)