Repository Issue Activity (beta)

#252 [Feature Request] Support compressed-tensors per-channel FP8 checkpoints for Qwen3.5 MoE

#373 Congratulations for the project, finally able to run something useful locally using this project

#141 Add support for all major 2026 KV cache compression techniques (TurboQuant, RotorQuant, KIVI, KVQuant, eviction methods)

#82 ROCm RDNA4 gfx1201 Ported. Windows only for now. Test please

#410 [Model request] KAT-Coder-V2.5-Dev NVFP4 community exports (compressed-tensors MoE): dense-branch misroute + GDN granularity

#97 Difficult to load model despite more than enough vram/ram available

#15 Support for dual GPU

#248 Two blocking bugs in Windows Desktop on consumer Blackwell (RTX 50): FP8 row-wise gate crash + Mistral tokenizer empty Chinese output

#227 [Bug]: RuntimeError: Rowwise scaling is not currently supported on your device

#143 [Bug]: Desktop UI window close leaves orphaned backend engine process locking port 8000 and GPU VRAM

#255 Feature request: create custom per-model cache config presets and/or remember last settings

#100 FP4 models are shown on 30XX Hardware

#35 FreeToken Engine Install Fails in Windows with RTX4060

#148 Ram size detection issue for loading models, and gpu auto detect issue

#207 [Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts)

#155 [Bug] App crashes on startup under Wayland/Arch Linux (WebKit2GTK NetworkProcessCrashed)

#158 Updater/downloader should honor HTTPS_PROXY / all_proxy so users behind MITM firewalls or tunnels can download the engine

#362 [Feature Request]: Support EXL3 Format

#60 Feature request: AMD GPU support

#365 Qwen3.5-122B-A10B-NVFP4

#409 Qwen3.8-Flash-Next-NVFP4 cannot start on a 12 GB RTX 3060: ~9.9 GiB of unquantized (BF16) attention/embedding weights leave no room for MoE cache + KV

#407 Qwen3.6‑35B‑A3B‑Uncensored‑HauhauCS‑Aggressive GGUF 加载报错 qwen35moe architecture not supported

#403 prompt is too long: 16535 tokens > 8221 maximum

#301 fix(server): allow overriding the distributed port

#302 feat(server): add a server-side thinking override