**Root Cause:** Insufficient GPU VRAM to allocate the Key-Value (KV) cache for the requested 262,144-token context length. KV cache memory scales linearly with context size, model depth, and hidden dimensions; 1 GB free is drastically below what's required for a 256K context.

**Next Two Actions:**
1. Reduce the context size via `--ctx-size` (or `-c`) to a smaller value (e.g., 4096–32768) that fits within your available VRAM.
2. Lower GPU offload layers using `--gpu-layers` (or `-ngl`) or set it to `0` to shift model weights and KV cache to system RAM, freeing VRAM. Alternatively, terminate other GPU-consuming processes to reclaim memory.