| --- |
| tags: |
| - model_hub_mixin |
| - pytorch_model_hub_mixin |
| license: wtfpl |
| datasets: |
| - bobox/OpenbookQA-4ST |
| language: |
| - en |
| pipeline_tag: feature-extraction |
| --- |
| |
| # PurpleGloVe |
|
|
| Simple implementation of GloVe using PyTorch and modern (2026) HuggingFace devkit such as Datasets, Tokenizers, and Safetensors. |
| This repository is suitable for any beginners trying to understand how to incorporate HuggingFace into their workflow with PyTorch. |
|
|
| We will be demonstrating the use of saved tokenizer and config by calling the `from_pretrained` method of a model or a tokenizer class. |
| A very first-principled use of HF devkit made easy to understand. |
|
|
| ## Usage |
|
|
| ```sh |
| uv run app.py |
| ``` |
|
|
| See `train.ipynb` for training recipe. |
|
|
| ## Notes on Variable Misnaming |
|
|
| I initially written W_i, W_j, B_i, and B_j like so. |
| But in the paper, W_j and B_j is actually also subscript `i` but with tidle above W and B to denote the context word vector. |
|
|
| The `app.py` script already has a note for this, the script works just fine. |
