TensorSharp v3.0.5.0

What's Changed

• optimize image edit by @zhongkaifu in #66
• docs: add Qwen-Image-Edit to model card indexes; remove stale model_cards_cn.md duplicate by @zhongkaifu in #68
• support vulkan backend by @zhongkaifu in #69
• optimize model loading time in vulkan backend by @zhongkaifu in #71
• docs(wiki): reflect auto-enabled GGML Vulkan backend and add ggml_vulkan examples by @zhongkaifu in #72
• feat(cuda): fused QK-Norm + RoPE kernel for Qwen3/Gemma prefill by @Manus-Vindicte in #75
• refactor(runtime): simplify scheduler — remove dead code, fix live-cache race by @Manus-Vindicte in #74
• Bug fix/qwen 9 b iq4 xs loading issue by @zhongkaifu in #76
• fix(runtime): cache TokenSampler per SequenceState by @Manus-Vindicte in #73
• fix broken build by @zhongkaifu in #77
• Update GitHub Pages actions to Node 24-native versions by @zhongkaifu in #78
• support PDF format document by @zhongkaifu in #79
• fix RoPE dimension mismatch in Qwen by @zhongkaifu in #80

New Contributors

• @Manus-Vindicte made their first contribution in #75

Full Changelog: v3.0.3.0...v3.0.5.0