Boulesis-26B-A4B-Voice
Boulesis is judged the best RP model in its class. Try its voice on your GGUF, see how it sounds — nothing crazy.
This is the voice of SubMaroon/Boulesis-26B-A4B in a single portable tensor (~1.5 GB). If you already have a Gemma 4 26B A4B GGUF, cast it on with the Voice tool and hear how it reads — your base underneath, Boulesis's writing on top.
What it is
Boulesis is a composite RP build: QK task arithmetic plus a fused LoRA over an abliterated Heretic body. The part that matters for this voice is documented in the author's own experiment log — of every operation in the build, the most measurable roleplay gain came from a single transplanted tensor: the lm_head from Gryphe's StyleTune-V2 (one tensor out of 659). Style, decisiveness, lore pulled from the character card instead of mirrored from the user.
We lift exactly that tensor and skip the rest, so the download is one tensor instead of a full model.
The source's numbers, linked not repeated
We have not probed this voice ourselves yet — everything below is the source author's published result, not ours:
- Best RP / ERP / Dark RP model in the 24–26B class per CaliperBench (thinking mode, September 2026).
- Zero catastrophic forgetting of base intelligence, per the author's testing — the build was designed to keep core knowledge while diversifying prose.
- Recommended settings from the source card: thinking ON, temperature 1.0, top-K 64, top-P 0.95, repetition penalty 1.05–1.1.
Two steps
# 1. Get the voice tool (one-time): https://huggingface.co/Wiself/voice
python3 voice.py path # optional, puts `voice` on your PATH
# 2. Cast it onto any Gemma 4 26B A4B GGUF you already have
voice cast ./gemma-4-26b-a4b-it-Q4_K_M.gguf voice.safetensors --out ./voiced/gemma-4-26b-a4b-boulesis.gguf
Append --speak to force the voice through output.weight (invented if the target lacks it and geometry survives; GGUF only).
Run it:
llama serve -m ./voiced/gemma-4-26b-a4b-boulesis.gguf --jinja
One file out. No adapter at runtime, no LoRA, no second model.
Works on more than the base model
Because the voice is just a tensor, it casts onto any compatible Gemma 4 26B A4B GGUF:
- Different finetunes — keep the stronger base, gain Boulesis's voice
- Quantized GGUFs — any quant; your tensors stay untouched, only the head is replaced
On heavily modified targets a direct cast can loop. The delta path fixes it — it carries only the difference from the base instruct, so MoE routing stays calm:
voice delta voice.safetensors --base google/gemma-4-26B-A4B-it
voice cast ./model.gguf delta-voice.safetensors --out ./voiced/model.gguf
Note: Google's instruct checkpoint is tied (no separate head tensor), so this needs a current Voice tool — it deltas the head against the tied embedding automatically.
What's inside
voice.safetensors— thelm_head.weighttensor, source dtype preserved, arch-derived shape[262144, 2816], ~1.5 GBvoice.json— metadata: source, dtype, shape, architecture
Confirm after download: voice info voice.safetensors → lm_head.weight · [262144, 2816].
Compatibility
| Target | Works? |
|---|---|
| Any Gemma 4 26B A4B GGUF (any quant) | ✅ |
| QAT-derived GGUFs | ✅ works, but the QAT voice is the exact match |
| Gemma 4 other sizes (9B, 12B, 31B) | ❌ shape mismatch — use the matching voice |
| Non-Gemma architectures | ❌ |
Notes
- Sampler tips from the source card: thinking ON, temp 1.0, top-K 64, top-P 0.95, repetition penalty 1.05–1.1.
- Usage terms carry over: the source is Gemma-licensed — see the source card before sharing voiced models.
- Casting quantizes only the head to Q8_0 (near-lossless); all other tensors byte-copied.
Credits
SubMaroon — the Boulesis composite (QK arithmetic + fused LoRA; fine-tunes by coder3101 and Gryphe, reasoning graft from Pantheon). The Voice tool lifts and casts it. Base model is Google's, under Gemma terms.