Research preview for reproducible MoE routing and memory oversubscription on consumer GPUs. Not a production inference engine.
-
Updated
Aug 26, 2026 - Python
#
model-offloading
Here are 5 public repositories matching this topic...
DualDeadline adds separate gate/up and down-projection transfer deadlines to exact MoE offloading, with H200 validation and Triton-optimized predictors.
reproducible-research gpu cuda pytorch triton moe inference-optimization mixture-of-experts llm-inference systems-research expert-prefetching model-offloading
-
Updated
Jul 30, 2026 - Python
Fits BF16 models too large for your GPU's VRAM by streaming losslessly compressed weights layer by layer, with speculative decoding, and an experimental benchmark suite showing it beats AirLLM and Hugging Face Accelerate at matched memory.
gpu model-compression vram local-llm llm-inference speculative-decoding llm-benchmarking openai-compatible-api model-offloading
-
Updated
Aug 28, 2026 - Python
Research artifact for Memory-Sovereign Inference: Output-Exact Execution Beyond Full Residency
reproducibility nvme mixture-of-experts llm-inference systems-research research-artifact model-offloading storage-backed-inference
-
Updated
Aug 24, 2026 - C++
Quality-first local MiniMax H3 video-series generation and loopback API for dual RTX 4090 workstations, with native audio, references, P8/P9 continuity, and preserved artifacts.
ffmpeg storyboard webapp video-series multi-gpu video-api video-generation image-to-video dual-gpu text-to-video audio-generation ai-video comfyui local-ai rtx-4090 reference-to-video world-travel lazying-art minimax-h3 model-offloading
-
Updated
Aug 28, 2026 - Python
Improve this page
Add a description, image, and links to the model-offloading topic page so that developers can more easily learn about it.
Add this topic to your repo
To associate your repository with the model-offloading topic, visit your repo's landing page and select "manage topics."