---
license: other
tags:
- text-to-speech
- f5-tts
- quantized
- q4
- distillation
---

# Peyton F5-TTS Q4 Distill - 6-step v193

This release packages the best currently accepted 6-step Peyton F5-TTS Q4 distilled checkpoint in the same `peyton_voice_q4` archive layout used by the previous 8-step release.

## Files

- `peyton_voice_q4.tar` - runtime bundle with F5-TTS Q4 weights, Vocos Q4 weights, Peyton reference WAV, and vocab.
- `manifest.json` - release metadata, source checkpoint, inference settings, and evaluation metrics.
- `checksums.sha256` - SHA-256 checksums for uploaded release files.

## Selected Checkpoint

- Source checkpoint: `runs/checkpoints/f5tts_q4_6step_v193_continue_underfill_floor/model_q4_24to6_best_rollout.pt`
- NFE steps: `6`
- Recommended CFG: `0.7`
- Selection: accepted guarded/best-of v193 candidate from the broad evaluation run.

The internal model directory remains `models/f5tts_peyton_q4_v0` for compatibility with consumers of the prior package layout.