IC-1237Llama-2-7b's pre-existing translation knowledge is diluted by large amounts of parallel data, causing COMET to decline after 100k examples

Haoran Xu, Young Jin Kim, Amr Sharaf, Hany Hassan Awadalla

SourceA Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models

Fine-tuning Llama-2-7b on en-to-ru parallel data at five scales (10k, 100k, 1m, 5m, 20m) reveals that its COMET score peaks at 10k-100k and then declines, while MPT-7b shows continuous improvement. A randomly initialized 7B model trained from scratch on 20m data reaches similar performance, supporting the hypothesis that large parallel data causes catastrophic forgetting of the LLM's pre-trained knowledge. BLEU does not capture this decline because the WMT training data is domain-matched to the test set.

Evidence
correlational
Key metric
Llama-2-7b COMET peaks at 10k/100k then declines at 5m/20m; MPT-7b COMET improves monotonically; random-init 7B at 20m reaches similar BLEU and COMET to fine-tuned LLMs
Caveat
Single language pair (en-to-ru) only; the decline is visible in COMET but not BLEU; the authors hypothesize catastrophic forgetting but do not directly measure it.
Model
Llama 2 / Llama 2 base Llama 2 7B, MPT MPT-7B
Concepts
Failure mode
Methods
COMET [eval], BLEU / BLEU@4 [eval]
Related work
NLLB-54b [compared-to]
Related findings
IC-1236, IC-1238, IC-1239
Extraction
automatic-extraction