Projects with this topic
-
Self-hosted LLM serving where the model store IS the registry: drop a GGUF in the directory and it is servable, no registration step.
Updated -
Multimodal Vision Demo: On-device Android app running Qwen3-VL-2B offline. Full tokenization and inference pipeline within mobile memory limits. Vulkan GPU acceleration via custom llama.cpp bridges. By Muhammad Assad Ullah.
Updated -
SATE AI: Fault Injection & Reliability Engineering Framework for On-Device AI. 11 fault injectors, 8 model adapters (ONNX, TFLite, llama.cpp, MediaPipe), 246-case test matrix. pub.dev score 160/160. 9 releases. 393 commits by Muhammad Assad Ullah. Zenodo: doi:10.5281/zenodo.23250418. Preparing for JOSS/arXiv submission.
Updated -
Exact 2026 LLM VRAM calculator & GPU compatibility matrix -> https://sheerai.net/can-it-run
Updated -
Benchmark harness and measured results: GGUF vs NVFP4 (llama.cpp and vLLM) and DFlash speculative decoding on a 12 GB Blackwell (sm120) GPU
Updated -
Make small local LLMs reliable at multi-step agent workflows. Constrained decoding + verify-before-commit: 96.5% end-to-end vs 2.0% unguarded on a 50-step run. Zero dependencies, runs on-prem.
Updated -
Native macOS app for Shapearator: extract, name and export every icon from a single sheet, entirely on your Mac. Bundles its own engine, so there is nothing to install.
Updated -
Turn one crowded sheet of drawn icons into a clean, individually named, ready-to-use asset library. Local-first desktop app and CLI, with optional AI naming by a vision model running on your own machine.
Updated