Pinned Loading

1. llama-cpp-pascal-cuda-windows llama-cpp-pascal-cuda-windows Public

   CUDA compile guides for llama-cpp-python on legacy NVIDIA Pascal GPUs (GTX 1050/1060/1070/1080) — Windows, VS2019/CUDA 12.1 and VS2022/CUDA 12.6 paths

2. SovereignKernel SovereignKernel Public

   Custom C++/CUDA LLM inference runtime built from scratch. Standalone GGUF loading, RoPE, KV cache management, and custom kernels without cuBLAS or external high-level frameworks.

   Cuda

3. turbovec-quantization-benchmark turbovec-quantization-benchmark Public

   Does TurboVec quantization hold up on exact-fact retrieval, or only on semantic similarity? Tests recall@k across bit_width 4/2 vs full precision on a synthetic report with 104 ground-truth facts. …

   Python

4. qwen-local-code-eval qwen-local-code-eval Public

   Empirical benchmark of Qwen quantization tiers (IQ1, Q6, IQ4) for local code generation under strict 8GB VRAM limits.

5. local-whisper-transcription-pipeline local-whisper-transcription-pipeline Public

   Local video/audio to subtitles pipeline. Zero cloud, VRAM-aware model selection, runs locally on any NVIDIA GPU. Built on faster-whisper and ffmpeg. Designed for readable orchestration logic, not d…

   Python

6. sdxl-lora-gguf-blueprint sdxl-lora-gguf-blueprint Public

   Blueprint for training custom SDXL LoRAs and running local GGUF inference on 8GB VRAM — no cloud, no ComfyUI