tts_e1_repro_c
Fine-tune of IndicF5 on Vāgdhenu chant, trained on A100 40GB (Google Cloud, spot). Started 9 Oct 2026, ended 9 Oct 2026.
Hypothesis
Third launch of TTS-E1, with the summary bug fixed.
What we learned
Stopped while preparing data: a library version on the cloud machine did not match the one used at home. The versions were pinned and the run relaunched as TTS-E1.
Scores
No scores published yet.
Curves
No training curves were published for this run.Its scores above are what it published.
Model
- Preset
- IndicF5
- Parameters
- —
- Layers · heads · width
- 22 · 16 · 1,024
Data
- Recordings
- Vāgdhenu chant
- Audio published
- None: these voices stay internal
Compute
- GPU
- A100 40GB
- Where
- Google Cloud (spot)
- Steps
- — / 15,000
- GPU hours
- 0.07
- Cost
- $0.10
- Spot restarts
- 0
Lineage
gcpa100spotindicf5