qwen38-mtp
One llama.cpp flag unlocks +33-39% decode speed for Qwen3.8-27B on consumer GPUs. The MTP head already ships inside your GGUF. Recipe, paired benchmarks, probe tool.
- 382
- Python
- Apache-2.0
- 8 wk ago
sudoingX
Infrastructure. Bangkok. @sudoingX
Built from sudoingX's public GitHub profile, Bangkok. Not on Take My Repo yet, so there are no posts or updates here.
One llama.cpp flag unlocks +33-39% decode speed for Qwen3.8-27B on consumer GPUs. The MTP head already ships inside your GGUF. Recipe, paired benchmarks, probe tool.
run a 27b local ai on your gaming gpu, rtx 3060 and up, radeon rx 6600 and up in beta, 8gb or more, linux and windows, one line to install
A pixel art space shooter built entirely by a 9B AI model on a single RTX 3060. Zero hand-written code.
DeepSeek-V4.1-Flash on two NVIDIA DGX Sparks: EXL3 weights from the official checkpoint, upstream vLLM with our patches and decode kernels
Time-boxed passwordless sudo for agents. Grants expire on a systemd timer, validated with visudo before install, audited with sudoreplay.
Serve the official Ling-3.0-flash INT4 on one NVIDIA DGX Spark, 38.7 tok/s, faster than the community GGUF. Recipe, watchdog, benchmarks.
Serve Poolside Laguna S 2.1 (NVFP4) on the NVIDIA DGX Spark (GB10) without hanging your box. Working stack, crash-safe configs, benchmarks, and the exact gotchas.
Floating Tree by sudoingx: a one-file prompt benchmark. Hand FLOATING_TREE.md to any coding agent and it builds a 3D floating island in Three.js, graded by 12 tests and a visual gate.