EmbBERT Bundle

This folder is a standalone EmbBERT bundle centered on the verified checkpoint-616000 pretraining artifact.

It includes:

• checkpoints/pretraining/checkpoint-616000/
• tokenizers/bpe_book_corpus_8192.json
• configs/EmbBERT_config.json
• manifest.json and loaders.py
• lib/
• runnable scripts such as pretrain.py, finetune.py, quantize.py, evaluate_embbert_bundle.py, and embbert_semantic_search_test.py
• pyproject.toml and uv.lock
• datasets/embbert_semantic_search_benchmark.json

Visualization utilities are intentionally excluded from this bundle. In particular, plotting helpers such as graphing.py and loss_plotter.py are not included.

Example

Run from inside this directory:

from loaders import load_latest_pretraining_checkpoint

model, tokenizer = load_latest_pretraining_checkpoint()