Highlights
- Pro
Pinned Loading
-
pth-flash-attn
pth-flash-attn PublicFlashAttention-style CUDA kernel that reads vLLM's int8 per-token-head paged KV cache in place (sm80/86/89), with a vLLM attention-backend plugin.
-
syv-ai/HyperQwen
syv-ai/HyperQwen PublicServe large Qwen models fast on the GPUs you actually own. Qwen3.8-27B on a single 24 GB card with vLLM: 127 tok/s single-user (381 when the answer quotes the prompt), ~1,035 tok/s at 64 concurrent…
-
Niko1221/Strata
Niko1221/Strata PublicQwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
-
geforce-sriov
geforce-sriov PublicEnable SR-IOV vGPU for consumer Ampere and Ada GeForce cards.
-
geforce-bar1-resize
geforce-bar1-resize PublicGive GeForce cards (Turing and later) a BAR1 at least as large as their VRAM on Linux, without flashing.
-
geforce-dynbar1-p2p
geforce-dynbar1-p2p PublicGPU peer-to-peer over dynamic BAR1 for GeForce cards: patched NVIDIA open kernel modules.
If the problem persists, check the GitHub status page or contact support.