Reproducible RTX 5090 dual-runtime 128K agent recipe for Qwen3.8-27B
-
Updated
Sep 18, 2026 - PowerShell
#
nvfp4
Here are 3 public repositories matching this topic...
Self-hosted llama.cpp station serving Claude Code from local weights on an RTX 5090, with the measurement behind every flag
windows powershell cuda self-hosted llama-cpp local-llm llm-inference speculative-decoding claude-code nvfp4
-
Updated
Sep 27, 2026 - PowerShell
GLM-5.3-Flash on one RTX 5090 with crow-nest and Crow: the 3.05-bit EXL3 MUL1 container with experts over VRAM, pinned RAM and NVMe, the arm4.env operating set, commands and measured numbers. No engine code, no model files. Windows.
windows cuda moe glm quantization zai blackwell openai-api llm local-llm local-ai rtx-5090 nvfp4 glm-5 exllamav3 exl3
-
Updated
Oct 12, 2026 - PowerShell
Add this topic to your repo
To associate your repository with the nvfp4 topic, visit your repo's landing page and select "manage topics."