vh
a6c1d3c454
A/B of the live STT seat (fv-ml1 GPU 0, sherpa-onnx int8 v3) against nvidia/parakeet-unified-en-0.6b, measured on GPU 3 with the seat's own image, k2-fsa's published unified int8 export, fp32/fp16 exports made with k2-fsa's recipe, v2 int8, and NeMo 3.0.0 (fp32, bf16 autocast, bf16 weights). - Seat int8 graph runs on one CPU thread (cpu/wall 1.00, GPU 2-9%). - unified-en under NeMo: -121/-234/-530 ms vs the seat at 1-3/3-8/8-20 s (paired, n=120/bin; floor <=6 ms; +50 ms positive control reads +52-54). - unified-en WER lower in every runtime: -0.7 pp clean, -1.5 pp other, -3.2 to -4.4 pp AMI (paired CIs exclude 0). - Seat defects found: hard 400 s input ceiling (HTTP 500), truncation after a quiet 1.5 s pause, and severe long-window dropouts (int8 v3 only). - B-bf16w needs +0.8 to +1.5 GB over the seat's 1,690 MiB on GPU 0. Raw requests, hypotheses, manifests and the full harness under services/parakeet-ab-2026-09-30/. No deploy; live seat untouched apart from 240 light test requests.
17 lines
2.7 KiB
JSON
17 lines
2.7 KiB
JSON
{"file": "scotus", "arm": "ab-a1", "k": 1, "status": 500, "err": "Internal Server Error"}
|
|
{"file": "scotus", "arm": "ab-a1", "k": 2, "status": 500, "err": "Internal Server Error"}
|
|
{"file": "scotus", "arm": "ab-b16w", "k": 1, "status": -1, "err": "ConnectionResetError(104, 'Connection reset by peer')"}
|
|
{"file": "scotus", "arm": "ab-b16w", "k": 2, "status": -1, "err": "ConnectionResetError(104, 'Connection reset by peer')"}
|
|
{"file": "scotus", "arm": "ab-b16wl", "k": 1, "e2e_s": 2.6, "wer": 4.99, "S": 70, "D": 209, "I": 9, "ref_words": 5772, "hyp_words": 5572, "dropouts": 4, "dropout_words": 54, "clean_dropouts": 1, "clean_dropout_words": 13, "insertion_runs": 0, "insertion_words": 0, "drop_spans_s": [[596.1, 597.7, 13, "crosstalk"], [608.6, 610.8, 11, "crosstalk"], [861.2, 864.4, 17, "crosstalk"], [1650.2, 1654.4, 13, "clean"]]}
|
|
{"file": "scotus", "arm": "ab-b16wl", "k": 2, "e2e_s": 2.3, "wer": 5.02, "S": 72, "D": 206, "I": 12, "ref_words": 5772, "hyp_words": 5578, "dropouts": 4, "dropout_words": 54, "clean_dropouts": 1, "clean_dropout_words": 13, "insertion_runs": 0, "insertion_words": 0, "drop_spans_s": [[375.4, 377.8, 14, "crosstalk"], [596.1, 598.3, 16, "crosstalk"], [608.6, 610.8, 11, "crosstalk"], [1650.2, 1654.4, 13, "clean"]]}
|
|
{"file": "scotus", "arm": "ab-cf32", "k": 1, "status": -1, "err": "ConnectionResetError(104, 'Connection reset by peer')"}
|
|
{"file": "scotus", "arm": "ab-cf32", "k": 2, "status": -1, "err": "ConnectionResetError(104, 'Connection reset by peer')"}
|
|
{"file": "wilde", "arm": "ab-a1", "k": 1, "status": 500, "err": "Internal Server Error"}
|
|
{"file": "wilde", "arm": "ab-a1", "k": 2, "status": 500, "err": "Internal Server Error"}
|
|
{"file": "wilde", "arm": "ab-b16w", "k": 1, "status": -1, "err": "RemoteDisconnected('Remote end closed connection without response')"}
|
|
{"file": "wilde", "arm": "ab-b16w", "k": 2, "status": -1, "err": "ConnectionResetError(104, 'Connection reset by peer')"}
|
|
{"file": "wilde", "arm": "ab-b16wl", "k": 1, "e2e_s": 2.0, "wer": 3.85, "S": 54, "D": 81, "I": 3, "ref_words": 3580, "hyp_words": 3502, "dropouts": 1, "dropout_words": 57, "clean_dropouts": 1, "clean_dropout_words": 57, "insertion_runs": 0, "insertion_words": 0, "drop_spans_s": [[12.8, 40.1, 57, "clean"]]}
|
|
{"file": "wilde", "arm": "ab-b16wl", "k": 2, "e2e_s": 1.6, "wer": 3.77, "S": 52, "D": 80, "I": 3, "ref_words": 3580, "hyp_words": 3503, "dropouts": 1, "dropout_words": 57, "clean_dropouts": 1, "clean_dropout_words": 57, "insertion_runs": 0, "insertion_words": 0, "drop_spans_s": [[12.8, 40.1, 57, "clean"]]}
|
|
{"file": "wilde", "arm": "ab-cf32", "k": 1, "status": 500, "err": "Internal Server Error"}
|
|
{"file": "wilde", "arm": "ab-cf32", "k": 2, "status": -1, "err": "RemoteDisconnected('Remote end closed connection without response')"}
|