aNLP Assignment 3 (2024102060)

Checkpoints, configs, results, outputs and logs for ANLP Assignment 3. The code is in the course repository; scripts/part1/test_assn.sh and scripts/part2/test_gnn.sh download the checkpoints from this repository.

Layout

part1/                        cipher decryption (Part A)
  checkpoints/{transformer,mamba,deltanet,hybrid}.pt   best-validation checkpoint of each model
  configs/cipher.yaml                                  architectures, training and inference settings
  results/{validation,test}_metrics.json, *_tables.md  full-document metrics of the four models
  results/efficiency.json                              FLOPs, latency and peak memory against sequence length
  results/train/{model}.json                           training summary and validation curve
  results/eval/{split}_{model}.json                    metrics plus accuracy by position and by distance to noise
  outputs/{split}_decrypt_{model}.jsonl                decrypted validation and test documents ({"id", "text"})
  outputs/{split}_decrypt.jsonl                        decryptions of the default model (hybrid)
  logs/                                                training, evaluation and profiling logs
part2/                        fake news detection on UPFD GossipCop (Part B)
  checkpoints/{model}_{feature}_seed{s}.pt             best-validation-F1 checkpoint of each run
  configs/                                             config and the hyperparameters chosen by the sweep
  results/                                             results.json, per-run records (runs/) and the sweep (sweep/)
  logs/                                                sweep, training and evaluation logs

Usage

bash scripts/part1/test_assn.sh                          # all four Part A models on the test split
bash scripts/part1/test_assn.sh --heldout                # output/part1/heldout_result_2024102060.jsonl (hybrid)
bash scripts/part1/test_assn.sh --heldout --model mamba  # transformer | mamba | deltanet | combo
bash scripts/part2/test_gnn.sh --ta                      # Part B evaluation
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support