R9V
A shape-specialized inference setup that runs the Qwen3.8 Flash Next model across two AMD Radeon AI PRO R9700 GPUs, combining a pinned vLLM fork, GGUF loading and speculative decoding.
R9V runs a specific large language model, Qwen3.8 Flash Next, across two AMD Radeon AI PRO R9700 GPUs by combining a pinned fork of vLLM, GGUF model loading, kernels specialized for the RDNA4 architecture, expert offloading and four-token speculative decoding. It exposes an OpenAI-compatible API for text, tool calls and single-image requests.
Each configuration profile binds together a specific model package, runtime, hardware layout and expert placement, and downloads are checked against pinned revisions and file hashes before use; the first start of a profile qualifies the workload and available memory headroom before reporting the setup ready. The project tracks its profiles' status through a versioned changelog and reference-host benchmarks, including how much free VRAM headroom each profile leaves at minimum.