YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

skimo2022/anlp-a2-p1-dense

Advanced NLP Assignment 2, Part 1: dense.

Decoder-only Transformer translating Vietnamese / Japanese -> English, trained from scratch on belumind/en-vi-ja-curated-500k-triplets.

Prompt format: <bos> <vi|ja> {source} <2en> -> model generates English until <eos>.

eval_results.json

{
 "variant": "dense",
 "params": {
  "total": 35265024,
  "active": 35265024,
  "non_embedding_total": 18881024,
  "non_embedding_active": 18881024
 },
 "test_ppl": {
  "vi": 4.276593038997194,
  "ja": 6.008487370690277,
  "all": 5.069107935761143
 },
 "bleu_vi": 42.22442660458943,
 "chrf_vi": 61.837753083140036,
 "bleu_ja": 31.45992023328144,
 "chrf_ja": 55.048373339384796,
 "bleu_all": 36.85744077459305,
 "n_bleu_examples": 10000,
 "decode_time_s": 76.35277032852173,
 "train": {
  "tokens_trained": 110009907,
  "steps": 10456,
  "epochs": 2.999093074141973,
  "train_time_s": 2151.8958716392517,
  "tokens_per_s": 51122.318904863016,
  "params": {
   "total": 35265024,
   "active": 35265024,
   "non_embedding_total": 18881024,
   "non_embedding_active": 18881024
  }
 }
}

train_summary.json

{
 "tokens_trained": 110009907,
 "steps": 10456,
 "epochs": 2.999093074141973,
 "train_time_s": 2151.8958716392517,
 "tokens_per_s": 51122.318904863016,
 "params": {
  "total": 35265024,
  "active": 35265024,
  "non_embedding_total": 18881024,
  "non_embedding_active": 18881024
 }
}

Load with Transformer.load(<local snapshot dir>) from src/common/model.py of the course repo.

Downloads last month
10
Safetensors
Model size
35.3M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support