**Root Cause:**
The requested context window (262,144 tokens) requires significantly more GPU VRAM than the 1,024 MiB currently available. Even with aggressive quantization, a 256k context window typically requires several gigabytes of VRAM just for the KV cache.

**Next Actions:**
1. **Reduce Context Size:** Lower the `--ctx-size` parameter to a manageable value (e.g., 4096 or 8192) to fit within your available memory.
2. **Optimize Model Loading:** Use a smaller model or a more compressed quantization (e.g., Q4_K_S) to reduce the base memory footprint, or use the `--tensor-split` / `--offload` flags to move parts of the model to system RAM.