transformers

1ddc53ac - Reduce memory for speech recognition tests via --max_duration_in_seconds 3.0

Reduce memory for speech recognition tests via --max_duration_in_seconds 3.0 librispeech_asr_dummy has 73 validation samples ranging 1.6–29.4s (mean 6.6s). Filtering to <= 3s keeps 10 samples, cutting audio data and sequence lengths. - test_run_speech_recognition_ctc_adapter: PSS ~2,589 MB → ~1,250 MB (-52%) - test_run_speech_recognition_seq2seq: PSS ~2,617 MB → ~1,227 MB (-53%) Both still pass eval_loss < train_loss. Runtime: ~24s → ~11s each (-55%). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>