Wiself commited on
Commit
d091bde
·
1 Parent(s): fa53502

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -10
README.md CHANGED
@@ -60,6 +60,16 @@ No rebuild. No extra setup at runtime.
60
 
61
  ---
62
 
 
 
 
 
 
 
 
 
 
 
63
  ## How it works
64
 
65
  Two benefits, no extras to learn:
@@ -117,16 +127,6 @@ voice cast ./model.gguf styletune-v2 --out ./voiced/
117
  > `delta-` is reserved. Voice names cannot start with it. Deltas always do. Full CLI reference: [`DOCS.md`](./DOCS.md) (generated from `voice --help`, 7 commands). Names are lowercased when saved.
118
 
119
 
120
- ## Not just Gemma — any compatible model
121
-
122
- <p align="center"><strong>Voice is not tied to Gemma.</strong></p>
123
-
124
- <p align="center">If the architecture and vocabulary match, the voice transfers — <strong>Qwen</strong>, <strong>Gemma</strong>, <strong>Llama</strong>, any family. Take the voice from a finetune and cast it onto the base you already have. The same file you run for reasoning keeps its reasoning; the voice just changes how it sounds.</p>
125
-
126
- <p align="center">For example, a voice from <a href="https://huggingface.co/Wiself/Holodeck-Lounge-MTP">Wiself/Holodeck-Lounge-MTP</a> casts onto any <code>qwen3.5-9b</code> base, <a href="https://huggingface.co/ReadyArt/For-Her-Darkside-12B-v1.4">ReadyArt/For-Her-Darkside-12B-v1.4</a> onto any <code>gemma4 12b</code> base — even mixed with <a href="https://huggingface.co/ReadyArt/Serenity-12B">ReadyArt/Serenity-12B</a> to see what happens. <code>Gemma 4 26B MoE</code> above is just one illustration. Mix and match within the same family: keep the more performant base, keep the better voice.</p>
127
-
128
- ---
129
-
130
  ## What StyleTune showed
131
 
132
  Gryphe reports for StyleTune-V2—on their private 200-prompt roleplay eval (private regex, certified cliché-free, 1–20 turns, greedy 0.0)—52% fewer clichés per 100 words (1.141→0.551) and 19.9% shared trigram vocabulary vs base instruct.[^1]
 
60
 
61
  ---
62
 
63
+ ## Not just Gemma — any compatible model
64
+
65
+ <p align="center"><strong>Voice is not tied to Gemma.</strong></p>
66
+
67
+ <p align="center">If the architecture and vocabulary match, the voice transfers — <strong>Qwen</strong>, <strong>Gemma</strong>, <strong>Llama</strong>, any family. Take the voice from a finetune and cast it onto the base you already have. The same file you run for reasoning keeps its reasoning; the voice just changes how it sounds.</p>
68
+
69
+ <p align="center">For example, a voice from <a href="https://huggingface.co/Wiself/Holodeck-Lounge-MTP">Wiself/Holodeck-Lounge-MTP</a> casts onto any <code>qwen3.5-9b</code> base, <a href="https://huggingface.co/ReadyArt/For-Her-Darkside-12B-v1.4">ReadyArt/For-Her-Darkside-12B-v1.4</a> onto any <code>gemma4 12b</code> base — even mixed with <a href="https://huggingface.co/ReadyArt/Serenity-12B">ReadyArt/Serenity-12B</a> to see what happens. <code>Gemma 4 26B MoE</code> above is just one illustration. Mix and match within the same family: keep the more performant base, keep the better voice.</p>
70
+
71
+ ---
72
+
73
  ## How it works
74
 
75
  Two benefits, no extras to learn:
 
127
  > `delta-` is reserved. Voice names cannot start with it. Deltas always do. Full CLI reference: [`DOCS.md`](./DOCS.md) (generated from `voice --help`, 7 commands). Names are lowercased when saved.
128
 
129
 
 
 
 
 
 
 
 
 
 
 
130
  ## What StyleTune showed
131
 
132
  Gryphe reports for StyleTune-V2—on their private 200-prompt roleplay eval (private regex, certified cliché-free, 1–20 turns, greedy 0.0)—52% fewer clichés per 100 words (1.141→0.551) and 19.9% shared trigram vocabulary vs base instruct.[^1]