**Root Cause:** The requested context length (262,144 tokens / 256K) requires a KV cache that vastly exceeds the available GPU memory. Only 1,024 MiB (1 GB) of VRAM is free, which is insufficient for a 256K context on virtually any modern LLM. KV cache memory scales linearly with context length and model architecture; typical 256K contexts require tens to hundreds of GBs of VRAM.

**Next Two Actions:**
1. Reduce the context size (`--ctx-size`) to a value that fits your available VRAM (e.g., 4096, 8192, or 32768 depending on your model size and total GPU memory).
2. Decrease the number of layers offloaded to the GPU (`--gpu-layers` or `-ngl`) to free up VRAM for the KV cache, or switch to a GPU with significantly more memory.