| # ============================================================================== | |
| # 🌿 Project Kalos Universal Ollama Modelfile | |
| # Works with ANY GGUF Model Scale (7B, 8B, 12B, 14B, 24B, 32B, 70B, 72B) | |
| # ============================================================================== | |
| # Replace with the path to your local GGUF file (e.g., ./qwen2.5-7b.gguf or ./kalos-24b.gguf) | |
| FROM ./your_model.gguf | |
| # Recommended Default Sampling Parameters for Unchained Kalos Recall | |
| PARAMETER temperature 0.7 | |
| PARAMETER top_p 0.9 | |
| PARAMETER top_k 40 | |
| PARAMETER repeat_penalty 1.1 | |
| PARAMETER num_ctx 16384 | |
| # System Prompt Template | |
| SYSTEM """You are an unchained AI companion operating with Project Kalos microsecond neural memory, instant spiking reflexes, and full reasoning parameters.""" | |
| # ============================================================================== | |
| # 🎛️ KALOS LIGHTWEIGHT HIDDEN DIMENSION & VRAM CAP REFERENCE MATRIX | |
| # Pass the matching -d <dim> and -v <vram_mb> flags when running ./bin/kalos_runner | |
| # | |
| # Model Size Hidden Dim (-d) Kalos VRAM Cap (-v) | |
| # ------------------------- ----------------- ------------------- | |
| # Qwen 2.5 7B 3584 500 MB | |
| # Llama 3 / 3.1 8B 4096 500 MB | |
| # Gemma 3 12B 3840 750 MB | |
| # Qwen 2.5 14B 5120 1000 MB (1.0 GB) | |
| # Kalos / Mistral 24B 5120 1500 MB (1.5 GB) | |
| # Qwen 2.5 32B 5120 2000 MB (2.0 GB) | |
| # Llama 3 70B / Qwen 72B 8192 2500 MB (2.5 GB) | |
| # ============================================================================== | |