Zero Shot Inference with Transformers

This repository contains code that demonstrates the use of the Hugging Face Transformers library for zero-shot text classification and text generation. It includes examples with BART for classification and GPT-2 for text generation.

Contents

ZeroShotInference Class: A class to handle zero-shot text classification and text generation. main.py: Example usage of the ZeroShotInference class for text classification and text generation.

Installation

To run this code, you need to install the Hugging Face Transformers library and its dependencies. You can install them with:

pip install transformers torch

Usage

Zero-shot Text Classification

This example uses the Facebook BART large model (BART-Large-MNLI) to classify a text based on given candidate labels.

from transformers import pipeline
class ZeroShotInference:
    def __init__(self, task, pre_trained_model):
        self.model = pre_trained_model
        self.task = task
        self.pipe = pipeline(task=self.task, model=self.model)
    def classify_text(self, text, candidate_labels):
        result = self.pipe(text, candidate_labels)
        return result

Example usage:

text = "I love to play video games"
candidate_labels = ['hobby', 'habit', 'adventure', 'boredom', 'business']
task = "zero-shot-classification"
model = "facebook/bart-large-mnli"
zero_short_inference = ZeroShotInference(task, model)
result = zero_short_inference.classify_text(text, candidate_labels)
print(result)  # Output: {'sequence': 'I love to play video games', 'labels': ['hobby', 'habit', 'adventure', 'business', 'boredom'], 'scores': [0.8799885511398315, 0.09845343977212906, 0.016700521111488342, 0.0031407771166414022, 0.0017165272729471326]}

Text Generation

This example uses GPT-2 to generate text based on a given prompt.

def generate_text(self, prompt, max_length=100, temperature=0.7):
    output = self.pipe(prompt, max_length=max_length, do_sample=True, temperature=temperature, truncation=True)
    return output[0]['generated_text']

Example usage:

prompt = "I was doing coding last night"
model_2 = "gpt2"
task_2 = "text-generation"
zero_shot_infernece_2 = ZeroShotInference(task_2, model_2)
result_2 = zero_shot_infernece_2.generate_text(prompt)
print(result_2)

IN 2nd Task

Quatization

The model uses quantization techniques for efficient inference, reducing memory usage and speeding up response times using BitsAndBytesConfig

HF_TOKEN

for HF_token as mention i have generated the token from huggingFace and have added to secret of google colab to be used

IN 3rd Task

This task aims to fine-tune the Mistral model for generating instructional text using the 7B variant. It involves training the model on a specific dataset and then using it to generate instructional text based on given prompts.

FineTunning

The Fine-Tuning Mistral Instruct 7B project leverages the Mistral model, a large-scale language model, for generating instructional text. By fine-tuning the model on a dataset containing instructional text, the goal is to create a specialized model capable of generating coherent and informative instructions.

Contributions

Contributions to the project are welcome. Feel free to fork the repository, make your changes, and submit a pull request. You can also open issues to suggest improvements or report bugs.