Commit graph

4 commits

Author SHA1 Message Date
Quentin Fuxa
b102e12943 M5 benchmark figures: WER vs RTF scatter, 0.6B+1.7B MLX results 2026-03-15 15:00:00 +01:00
Quentin Fuxa
7aa3b764bd MLX benchmark: 1.7B SimulStreaming on M5 (WER 4.07%, RTF 0.944)
LibriSpeech test-clean, 500 utterances.
1.7B is borderline real-time on M5 (RTF 0.944).
0.6B (3.30% WER, 0.263 RTF) is the practical choice for MacBook.
2026-03-15 14:00:00 +01:00
Quentin Fuxa
a422e604ae MLX benchmark: 0.6B SimulStreaming on M5 MacBook (WER 3.30%, RTF 0.263)
LibriSpeech test-clean, 500 utterances, per-utterance simul-streaming.
AlignAtt border detection with 20 alignment heads.
Platform: Apple M5 32GB (MLX fp16).

benchmark_mlx_simul.py: reusable benchmark script for MLX backends.
2026-03-15 13:00:00 +01:00
Quentin Fuxa
47d4cbeecc reorganize benchmarks: move H100 results to benchmarks/h100/ 2026-03-15 23:59:00 +01:00