schema = "m64/capsule@1"description = "Qwen3.8-Flash-Next on Basalt (RTX 5090), for opencode. A local OpenAI-compatible chat endpoint for Qwen3.8-Flash-Next (IQ3_S pack) on the Basalt engine, with a 384k context (YaRN x1.5), for opencode running on your own machine: opencode and its file access stay there, only the model calls come here. Streams, tool calls, images. About 300 tok/s on one RTX 5090 with 64 GB of RAM. Needs no network."kinds = ["chat"]authors = ["dh7"]license = "mit"[hardware]platform = "linux-x86_64"gpu = "NVIDIA GeForce RTX 5090"gpu_count = 1vram_gb = 30# The pack's experts (46.8 GiB) sit in system RAM next to what VRAM caches: a 64 GB# machine (which reports ~60.5 GB).ram_gb = 60# CUDA 13; tested on driver 595.driver = ">=580"disk_gb = 90[runtime]sandbox = "container"# No image: the Dockerfile builds Basalt at a pinned commit on nvidia/cuda 13.0.0, pinned by digest.gpu = trueshm_gb = 8internal_port = 9000health = "/health"# ~30 s when the pack is in the page cache; a cold read of 87 GB off disk takes longer.expected_load_seconds = 240# The front rewrites every request's "model" to this; start.sh gives Basalt the same alias.model_name = "qwen3.8-flash-next-iq3_s-basalt-384k"user = "65534:65534"# The Basalt pack is the whole model (weights, tokenizer, chat template, vision, drafter).[[artifacts]]type = "hf"repo = "jesdga/Qwen3.8-Flash-Next-Basalt"revision = "2a6a758dc12f1d027d2d35b8964110df7e80f70b"include = ["LICENSE", "Qwen3.8-FN-GSQ-RCO-IQ3_S.basalt"]mount = "/weights/basalt"