Nibmeister/voice-bench
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Frame epoch 2 vs epoch 3 as a live-testing decision, not a settled pick

Browse Source 
Metrics lean toward ep3 but aren't decisive enough to call it outright.
README now presents both checkpoints as candidates and walks through
running them side-by-side (different ports) for direct comparison in
voice-bench.html, rather than naming one as current production.
This commit is contained in:
Joey Grasty
2026-08-30 19:23:12 -05:00
parent 0bec3bb3f0
commit 7ac0b36ae6
1 changed files with 22 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
README.md
View File

@@ -24,17 +24,16 @@ The 208 training/holdout packages themselves are already embedded inside
### 1. Get the model file
Download `Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep3-Q8_0.gguf` (~31GB) from
Google Drive — **this is the current production checkpoint**, switched from
ep2 on 2026-08-30 based on separability and length-calibration evidence
(see `JOINT_TRAINING_CLOSEOUT_REPORT.md`):
Two candidate checkpoints are available. Separability and length-calibration
metrics lean slightly toward epoch 3 (see `JOINT_TRAINING_CLOSEOUT_REPORT.md`
for the numbers), but not decisively enough to settle it on metrics alone —
**live testing through `voice-bench.html` is what should decide between
them.** Download both, run one at a time (or both at once on different
ports — see step 3), and compare generations on real held-out packages
before picking one.
https://drive.google.com/open?id=1QZT488Uu1b9Pxoie9Hs2lnxtOhMD4Pwr
The previous production checkpoint, `Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep2-Q8_0.gguf`,
remains available for comparison:
https://drive.google.com/open?id=1q1w_XMp6oRmfx8xqXDV-HvDx_krVxQtD
- `Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep3-Q8_0.gguf` (~31GB): https://drive.google.com/open?id=1QZT488Uu1b9Pxoie9Hs2lnxtOhMD4Pwr
- `Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep2-Q8_0.gguf` (~31GB): https://drive.google.com/open?id=1q1w_XMp6oRmfx8xqXDV-HvDx_krVxQtD
Memory needed: ~31GB for the weights plus ~3GB of KV cache at the full
32768-token context, so ~35GB total. Confirmed working on a DGX Spark;
@@ -57,6 +56,8 @@ This produces `llama.cpp/build/bin/llama-server`, which is what
### 3. Start the model server
Point `MODEL_PATH` at whichever checkpoint you're testing:
```
MODEL_PATH=/path/to/Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep3-Q8_0.gguf \
LLAMA_SERVER=/path/to/llama.cpp/build/bin/llama-server \
@@ -67,6 +68,17 @@ Leave this running in a terminal — it's the process `voice-bench.html`
talks to. If you'd rather not set environment variables every time, edit the
defaults at the top of `start-model.sh` instead.
**To compare epoch 2 and epoch 3 directly**, run both at once on different
ports (memory permitting — see below):
```
MODEL_PATH=/path/to/Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep2-Q8_0.gguf PORT=8200 ./start-model.sh
MODEL_PATH=/path/to/Qwen3-30B-A3B-VoxDay-Nuttall-SFT-ep3-Q8_0.gguf PORT=8201 ./start-model.sh
```
then switch `voice-bench.html`'s "Server" field between `:8200` and `:8201`
on the same packet to compare their generations side by side.
### 4. Open voice-bench.html
Open the file in a browser. The "Server" field in the top-right corner