Sharded, differentiable quantum simulation and training in PyTorch, reaching domestic accelerators through FlagOS.
-
Updated
Nov 21, 2026 - Python
Sharded, differentiable quantum simulation and training in PyTorch, reaching domestic accelerators through FlagOS.
Personal fork for vLLM inference optimization on Iluvatar BI-V150: attention launch policies, profiling, numerical validation and reproducible benchmarks.
1st place solution for FlagOS Open Computing Global Challenge — Track 3: Long-Context LLM Automatic Data Annotation (Score: 87.24)
FlagOS Track 1 - 20 Triton GPU operators (PyTorch + Triton, Apache-2.0). FlagOS Open Computing Global Challenge, Season 1.
High-throughput MiniCPM5-2B inference optimization on FlagOS and FlagGems.
FlagOmni is part of FlagOS. It is a high-performance operator library for omni-modality generation models (video, image, audio), implemented in Triton and designed for multiple hardware backends.
To associate your repository with the flagos topic, visit your repo's landing page and select "manage topics."