Views
No views yet
[!IMPORTANT] TEXT-ONLY OPTIMIZATION / テキスト専用最適化This Ollama release is a high-density text-only build. Vision support is excluded to prioritize logical throughput and weight efficiency in thefieldmouse-ainamespace.👁️ FOR VISION SUPPORT (Multimodal): Please visit our HuggingFace Forge to download the full GGUF weights and accompanying.mmprojfiles.
presence_penalty and repeat_penalty were applied to each parameter class to prevent cognitive drift.
階層型ペナルティ・スケーリング: 各パラメータクラスに応じたpresence_penaltyとrepeat_penaltyの独自設定により、認知的ドリフトを防止しています。imatrix process was specifically optimized for high-fidelity Japanese performance, ensuring cultural and linguistic nuance is preserved even at low bitrates.
日本語優先最適化: 日本語のパフォーマンスを最優先にimatrixプロセスを最適化。低ビットレートにおいても、文化的・言語的なニュアンスを高度に維持します。| Designation / 名称 | Parameters | Focus / 焦点 | Benchmark (TPS) | Size (Avg) / サイズ |
|---|---|---|---|---|
| GHOST | 0.8B | Legacy Reclamation | 6+ (CPU) / 130+ (GPU) | < 1 GB |
| VANGUARD | 2B | High-Density Logic | 115+ (GPU) | 1.3 GB |
| GUARDIAN | 4B | Technical Accuracy | 60+ (GPU) | 2.7 GB |
| TITAN | 9B | Deep Architecture | 40+ (GPU) | 5.6 GB |
| COLOSSUS | 27B | Frontier-class Intelligence | 15+ (GPU) | 16.0 GB |
1.4 (Ghost/Vanguard), 1.15 (Guardian/Titan/Colossus),0.50 (Ghost/Vanguard), 0.10 (Guardian), 0.00 (Titan/Colossus)4096