ANLP Assignment 1 - C1 baseline

Custom encoder-decoder Transformer using sinusoidal positions, multi-head attention, pre-LayerNorm and from-scratch byte-level BPE subword tokenization.

Greedy test metrics

bit_accuracy: 0.633670
sequence_accuracy: 0.000000
levenshtein_distance: 457.302000
bleu: 0.149689
rouge1: 0.131599
rouge2: 0.015977
rougeL: 0.116996
parameters: 11214124.000000
best_validation_loss: 2.193132

Load best_model.pt with the architecture and tokenizers from this assignment code.

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. 🙋 Ask for provider support