Generated 2026-04-19 21:19:12 · auto-refreshes every 30 min · IDLE — no training job in queue
Flow-matching DiT TTS (779M) + trainable byte encoder (57M).
8 nodes × 8 MI250x GCDs, global batch 2052, AdamW + cosine LR, lr=1e-4.
Primary job 17574420, resume jobs 17598698, 17617259, 17617423, 17618275.
| job | name | state | elapsed | limit | nodes | reason / nodelist |
|---|---|---|---|---|---|---|
| 17626980 | tts_demo_emo | RUNNING | 1:15 | 30:00 | 1 | nid005017 |
| job | name | state | start | end | elapsed | exit |
|---|---|---|---|---|---|---|
| 17574420 | allocation | CANCELLED by 10035305 | 2026-04-17T20:43:07 | 2026-04-17T20:43:07 | 00:00:00 | 0:0 |
| 17598698 | tts_50k | FAILED | 2026-04-18T05:07:14 | 2026-04-18T05:11:10 | 00:03:56 | 15:0 |
| 17617259 | tts_50k | FAILED | 2026-04-18T22:02:20 | 2026-04-18T22:08:08 | 00:05:48 | 15:0 |
| 17617423 | tts_50k | FAILED | 2026-04-18T22:21:21 | 2026-04-18T22:30:23 | 00:09:02 | 15:0 |
| 17618275 | tts_50k | COMPLETED | 2026-04-19T00:09:43 | 2026-04-19T13:26:13 | 13:16:30 | 0:0 |
| step | mean WER | median WER | mean CER |
|---|---|---|---|
| 5,000 | 1.279 | 1.000 | 1.041 |
| 10,000 | 2.460 | 1.000 | 1.968 |
| 15,000 | 2.784 | 1.000 | 1.513 |
| 20,000 | 2.523 | 1.000 | 1.595 |
| 25,000 | 3.354 | 1.111 | 2.183 |
| 30,000 | 1.454 | 0.967 | 0.902 |
| 40,000 | 1.497 | 0.903 | 0.980 |
| 45,000 | 0.827 | 0.867 | 0.575 |
| 50,000 | 0.853 | 0.793 | 0.622 |
Eval at step 30K (median WER 0.97) was the first checkpoint below 1.0 median WER — intelligibility threshold crossed.