llama.cpp PR cuts Qwen Flash Next VRAM use

A llama.cpp pull request (#29825) halves the indexer score memory, reducing VRAM usage for Qwen Flash Next.

• Open-source Mac app DigUp runs EmbeddingGemma 2 locally for file search2026-10-10
• llama.cpp merges MTP support for Qwen Flash Next2026-10-01
• Qwen 3.8 Flash runs at ~21 tok/s on RTX 30602026-10-10
• Four open decision models benchmarked on an RTX 40902026-10-08
• User builds 256GB VRAM rig with 8 Radeon Pro V620s2026-10-08