rajatsainju2025/code-explainer
Add tests for seq2seq tokenization and causal masking
オープン
#2 opened on 2025/08/08
good first issuehelp wanted
Repository metrics
- Stars
- (1 個のスター)
- PR merge metrics
- (PR metrics pending)
説明
Write unit tests covering:\n- Seq2seq preprocessing: source/target tokenization, -100 label masking for pad tokens.\n- Causal preprocessing: prompt masking (-100) and padding masking.\nUse tiny configs to avoid large downloads. Ensure tests run in CI.