---
library_name: pytorch
tags:
  - sequence-to-sequence
  - transformer
  - byte-latent-transformer
---

# ANLP Assignment 1 checkpoints

Repository: https://huggingface.co/BManav/ANLP_A1

These are the best-validation checkpoints for the controlled C1-C5 custom
Transformer ablation. Core Transformer components and the learned BPE tokenizer
were implemented from scratch for the assignment.

## Checkpoints

- `C1/best.pt`
- `C2/best.pt`
- `C3/best.pt`
- `C4/best.pt`
- `C5/best.pt`

## Weights & Biases runs

- [C1 WandB run](https://wandb.ai/manavberiwal006-iiit-hyderabad/anlp-ass1/runs/hccc1li3)
- [C2 WandB run](https://wandb.ai/manavberiwal006-iiit-hyderabad/anlp-ass1/runs/6hl9la7v)
- [C3 WandB run](https://wandb.ai/manavberiwal006-iiit-hyderabad/anlp-ass1/runs/5av2z9tt)
- [C4 WandB run](https://wandb.ai/manavberiwal006-iiit-hyderabad/anlp-ass1/runs/f5mnhh4i)
- [C5 WandB run](https://wandb.ai/manavberiwal006-iiit-hyderabad/anlp-ass1/runs/q6hxhm8r)

The accompanying `artifact_links.json`, configuration metadata, tokenizer merge
files, metrics, plots, and split fingerprint make the results auditable.