This repository provides from-scratch implementations of various text decoding strategies, offering users insights into the inner workings and nuances of each method. You can read the full in-depth blog here: Blog Link

• Text Decoding Strategies in LLMS
  • Table of Contents
  • Strategies Implemented
  • Prerequisites
  • Usage
  • Contributing
  • License

• Greedy Search: Picks the most probable next word at each step.
• Beam Search: Expands the most promising nodes in a breadth-first manner to maintain a beam of likely sequences.
• Random Sampling with Temperature: Selects the next word based on its probability distribution, with temperature controlling the randomness.
• TOP-K Sampling: Limits the next word candidates to the top K probable words.
• TOP-p (nucleus) Sampling: Chooses the smallest set of words whose cumulative probability exceeds a threshold p.

plotly==5.16.1
torch==2.0.1
transformers==4.33.1

For instance:

from greedy import GreedySampler
from beam import BeamSampler
from tempreature import RandomTempSampler
from topk import TOPKsampler
from topp import NucleusSampler

For a complete use case, see the usage.ipynb file.

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Distributed under the MIT License. See LICENSE for more information.