Skip to content

Latest commit

 

History

History
55 lines (36 loc) · 2.93 KB

README.md

File metadata and controls

55 lines (36 loc) · 2.93 KB

Sahayak : Sahayak App

Untitled.video.-.Made.with.Clipchamp_compressed.mp4

  • Sahayak -a vigilant antidrug companion, is a dedicated chatbot committed to providing support and information.Main objective of this chatbot is to handle sensitive chats of drugs & alcohol, as it gives remedies & specialized responses.Therefore our main goal is to finetune the chatbot over such sensitive data, & drug addicted person can use & ask questions freely

  • With empathy as its core, Sahayak aims to guide users away from the clutches of substance abuse, offering resources, encouragement, and a listening ear.

  • Harnessing technology for a healthier tomorrow, Sahayak stands as a reliable ally in the fight against drugs, promoting a drug-free lifestyle through understanding and assistance using technologies like finetuning, small language model (SLM), PEFT, LoRA, Python, Anvil, Cloud, tensorflow, NLP, pytorch, Phi-2 by Microsoft,

  • The main purpose of the project is give the users companion which will eventually guide and educate its users about the bad effects of the drugs eventually leading to a drug-free society.

Key Highlights

  • Data Preparation

Prepare a dataset in the required format (JSONL) with input-output pairs or any suitable format.

  • Set Up GPU and Environment

Utilize Google Colab ,AWS Sagemaker or a similar platform to obtain GPU resources.Choose appropriate specifications (e.g., Python version, CUDA version) for the environment.Build and configure the environment using the provided badge.

  • Load Base Model (Phi-2)

Load the Phi-2 model using 8-bit quantization for efficient training.

  • Tokenization

Set up the tokenizer, considering padding, truncation, and max_length.Tokenize the dataset for training and validation.

  • Set Up LoRA

Apply preprocessing to the model for training using LoRA layers.Define LoRA configurations for the model.

  • Run Training

Train the model using the provided Trainer class.Monitor training metrics and adjust hyperparameters as needed.

  • Model Evaluation

Evaluate the trained model on a sample prompt to check performance.

  • User Interface (UI) Integration

Utilize Anvil to create a user-friendly interface for the chatbot.

Tech Stack:

  • Machine Learning Libraries: Hugging Face Transformers, Accelerate, PEFT, Datasets
  • GPU and Cloud Services: AWS Sagemaker & Google Colab
  • Data Processing and Visualization: Matplotlib
  • Experiment Tracking: Weights & Biases
  • User Interface: Anvil
  • Python Libraries: Torch, scipy, ipywidgets, einops
  • Version Control: Git

App Link :

Link : https://knowledgeable-strident-cold.anvil.app/

Screenshots of live demo:

Screenshot (175)

image