tepirale/Ornith-Agents-A1-3.6-35B-A3B-GGUF overview
Note: I haven't used it. It's an experiment combining the model with MTP. I don't know if it will load or take forever to load. I don't know what response it w…
Runs locally from ~866.8 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-Q4_K_M.gguf | GGUF | Q4_K_M | 20.55 GB | Download |
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-Q5_K_M.gguf | GGUF | Q5_K_M | 23.87 GB | Download |
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-Q6_K.gguf | GGUF | Q6_K | 27.39 GB | Download |
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-Q8_0.gguf | GGUF | Q8_0 | 35.21 GB | Download |
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-f16.gguf | GGUF | F16 | 64.61 GB | Download |
| Ornith-Agents-A1-3.6-35B-A3B-dare_ties-mtp-sidecar.gguf | GGUF | GGUF | 866.8 MB | Download |
Model Details
Model README
---
base_model: tepirale/Ornith-Agents-A1-3.6-35B-A3B-dare_ties
tags: [gguf, mtp, qwen3.6, moe, dare-ties, ornith]
---
Note:
- I haven't used it.
- It's an experiment combining the model with MTP.
- I don't know if it will load or take forever to load.
- I don't know what response it will give.
- I haven't evaluated it.
- This model will be removed at some point.
Ornith-Agents-A1-3.6-35B-A3B-GGUF
GGUF multi-quant experimento de tepirale/Ornith-Agents-A1-3.6-35B-A3B-dare_ties con bloque MTP (Multi-Token Prediction) injertado desde el GGUF oficial de Qwen3.6-35B-A3B, para speculative decoding nativo en llama.cpp.
Uso (llama.cpp con MTP)
llama-server -m Ornith-Agents-A1-3.6-35B-A3B-dare_ties-Q4_K_M.gguf \
--spec-type draft-mtp --spec-draft-n-max 2 \
-ngl 99 -fa on -c 65536 --jinja \
--cache-type-k q8_0 --cache-type-v q8_0
Detalles técnicos
- Bloque MTP (
blk.40, 20 tensores) injertado después de cuantizar: pesos en Q8_0, routing gates en BF16, norms en F32 en todos los quants (sin degradación de las draft heads). block_count=41,nextn_predict_layers=1.- Quants generados directo desde F16 (sin doble cuantización).
Ornith-Agents-A1-3.6-35B-A3B-dare_ties-mtp-sidecar.gguf: sidecar con el bloque MTP crudo, por si quieres injertarlo sobre otros quants congraft_mtp.py.
Quants
| Quant | Tamaño |
|-------|--------|
| Q4_K_M | 20.5 GiB |
| Q5_K_M | 23.9 GiB |
| Q6_K | 27.4 GiB |
| Q8_0 | 35.2 GiB |
| f16 | 69.4 GiB |
Run tepirale/Ornith-Agents-A1-3.6-35B-A3B-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models