# This is to set the path to save the model
from pathlib import Path

models_path = Path.home().joinpath('Question_Generation_model', 'UTeMGPT')
models_path.mkdir(parents=True, exist_ok=True)

# Download the model
from huggingface_hub import snapshot_download
my_model = snapshot_download(repo_id="KLimaLima/finetuned-Question-Generation-mistral-7b-instruct", local_dir=models_path)

To load the model that have been downloaded

python

max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = my_model,
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)
FastLanguageModel.for_inference(model) # Enable native 2x faster inference

This model uses alpaca prompt format such as below

python

alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

instruction = 'Write an inquisitive question about a specific text span in a given sentence such that the answer is not in the text.'
sentence = "I want to bake a cake during my free time. I need to know the ingredients that need to be use."

inputs = tokenizer(
[
    alpaca_prompt.format(
        instruction,
        sentence,
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

To generate output

python

outputs = model.generate(**inputs, max_new_tokens = 64, use_cache = True)
tokenizer.batch_decode(outputs)

Uploaded model

Developed by: KLimaLima
License: apache-2.0
Finetuned from model : unsloth/mistral-7b-instruct-v0.2-bnb-4bit

This mistral model was trained 2x faster with Unsloth and Huggingface's TRL library.