Skip to content

v1.0.0 - Qwen3.8-9B Recurrent-Loop (1..6 Loops) Windows CUDA

Latest

Choose a tag to compare

@CympUser CympUser released this 20 Nov 22:41
· 9 commits to qwen35-loop since this release

1-Click Quickstart (Windows x64 + NVIDIA CUDA GPU, 8 GB+ VRAM)

  1. Download llama-cpp-qwen-loop-v1.0.0-win-cuda-x64.zip below and extract it.
  2. Double-click run.bat - it automatically downloads the base model (Qwen3.8-9B-Distill-uncensored-heretic.Q4_K_S.gguf) and the 45 MB Loop-6 adapter (qwen3.8-9b-heretic-loop6-lora.gguf), starts llama-server.exe, and opens http://localhost:8080.
  3. Use the model selector dropdown in the Web UI to compare 1-Loop, 2-Loop, 4-Loop, and 6-Loop live!

Requirement: NVIDIA GPU with up-to-date CUDA runtime installed on Windows x64 (tested on RTX 3070 Ti 8 GB VRAM - 6244 MiB peak at K=6).

Included Binaries

  • llama-server.exe, llama-cli.exe, llama-bench.exe, test-qwen35-loop.exe + CUDA/GGML DLLs
  • run.bat, run.ps1, models.example.ini, README_QWEN_LOOP.md