Popular repositories Loading

1. vllm-pcie-allreduce vllm-pcie-allreduce Public

   Fast tensor-parallel all-reduce for V100 GPUs connected only by PCIe. Patches 1Cat-vLLM so its SM70 push kernel runs across PCIe peer-to-peer instead of NCCL: 4-GPU decode 76 → 91 tok/s, within 4% …

   Python

2. turnstone_TTS turnstone_TTS Public

   Forked from turnstonelabs/turnstone

   Personal fork of Turnstone with voice (TTS, STT, voice mode). LLM-generated, unofficial; see README.

3. desktop desktop Public

   Forked from turnstonelabs/desktop

   Desktop client for the Turnstone AI orchestration platform

   JavaScript