ANLP_C4

This repository hosts model checkpoints, logs, and evaluation predictions for ANLP Assignment 1: Neural Decryption & Sequence-to-Sequence Modeling.

---

Benchmark Performance

Evaluated on the held-out test split:

Model Configuration	Sequence Accuracy	Bit Accuracy	BLEU	ROUGE-L	Mean Levenshtein Distance
C4_run_baseline_eng_tokens	16.00%	83.51%	91.57%	96.01%	15.80

---

How to Load Checkpoints

Download and inspect checkpoints directly using huggingface_hub and torch:

import torch
from huggingface_hub import hf_hub_download

ckpt_path = hf_hub_download(
    repo_id="anuml/ANLP_C4",
    filename="C4_run_baseline_eng_tokens_best.pt"
)

checkpoint = torch.load(ckpt_path, map_location="cpu")
print("Loaded checkpoint from epoch:", checkpoint.get("epoch"))
print("Model configuration:", checkpoint.get("config"))

---

Files in this Repository

• C4_run_baseline_eng_tokens_best.pt
• C4_run_baseline_eng_tokens_history.json
• C4_run_baseline_eng_tokens_loss.png
• C4_run_baseline_eng_tokens_metrics.log
• C4_run_baseline_eng_tokens_test_predictions.json
• C4_run_baseline_eng_tokens_throughput.png