https://stondo.github.io/tags/long-context/Recent content in Long-Context on Bits-EntangledHugo -- 0.162.1en-usThu, 27 Aug 2026 22:00:00 +0000https://stondo.github.io/posts/qwen38-flash-next-512k-two-rtx-pro-6000/Thu, 27 Aug 2026 22:00:00 +0000https://stondo.github.io/posts/qwen38-flash-next-512k-two-rtx-pro-6000/The official vLLM recipe for Qwen3.8-Flash-Next trusts GB300 at TP4. I wanted it on two workstation-class RTX PRO 6000 cards: TP2, the PLE n-gram table offloaded to pinned host RAM, YaRN stretched to 512K context, MTP speculative decoding. That stack delivers a 1.42M-token KV pool, 146.5 tok/s decode and working vision. It also taught me that fp8 KV is architecturally dead on QSA (two engines, two failure modes), that systemd strips quotes out of JSON args, and that 1M context dies at compile, not at KV budgeting.https://stondo.github.io/posts/qwen38-27b-nvfp4kv-262k-single-rtx5090/Thu, 27 Aug 2026 13:00:00 +0000https://stondo.github.io/posts/qwen38-27b-nvfp4kv-262k-single-rtx5090/How I got the full native 256K context of Qwen3.8-27B serving on a single 32 GB RTX 5090 — NVFP4 everything checkpoint, a patched vLLM nightly that unlocks NVFP4 KV cache on consumer Blackwell, 380K tokens of resident KV pool, and every landmine I stepped on: silent download stalls, a tokenizer that cost me an hour, an MTP setting that benched 18% faster and then exploded in production.