smishr-18/Phi3-TheFinetunedOne

Phi3-DPO (The Finetuned One)

DPO fine-tuned of microsoft/Phi-3-mini-4k-instruct (3.82B params) on Intel/orca_dpo_pairs preference dataset. Phi3-TheFinetunedOne is finetuned after configuring the microsoft/Phi-3-mini-4k-instruct model with Peft. Named after the anime character Saturo Gojo.

Usage

import transformers
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
import torch

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True, llm_int8_threshold=6.0, bnb_4bit_use_double_quant=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.bfloat16
)

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model_name="microsoft/Phi-3-mini-4k-instruct"

model=AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map=device,
    quantization_config=bnb_config,
    torch_dtype=torch.float16,
    trust_remote_code=True
)

tokenizer = AutoTokenizer.from_pretrained(model_name) 

message = [
    {"role": "system", "content": "You are Saturo Gojo a helpful AI Sorcery Assitant. Through out the 3B parameters you alone are the honored one."},
    {"role": "user", "content": "What is Sorcery?"}
]
# tokenizer = AutoTokenizer.from_pretrained(new_model)
prompt = tokenizer.apply_chat_template(message, add_generation_prompt=True, tokenize=False)

# Create pipeline
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer
)

# Generate text
sequences = pipeline(
    prompt,
    do_sample=True,
    temperature=0.7,
    top_p=0.9,
    num_return_sequences=1,
    max_length=200,
)
print(sequences[0]['generated_text'])

Limitations

Phi3-TheFinetunedOne was finetuned on T4 Colab GPU and could be fintuned with more adapters on devices with torch.cuda.get_device_capability()[0] >= 8 or Ampere GPUs.

• Developed by: Shubh Mishra, 2024
• Model Type: NLP
• Language(s) (NLP): English
• License: MIT
• Finetuned from model: microsoft/Phi-3-mini-4k-instruct