Popular repositories Loading

1. SWITCHLESS-NCCL-RING-4-GB10 SWITCHLESS-NCCL-RING-4-GB10 Public

   NCCL over a switchless 4-node DGX Spark / GB10 ring using both PCIe halves of every QSFP cable: ~193 Gb/s per cable instead of ~112, up to +34% vLLM prefill. One patch on top of switchless-nccl.

   Cuda 3

2. vllm-engram-nvme-4gb10 vllm-engram-nvme-4gb10 Public

   Python 1

3. deepseek-v4_1-1017-4-gb10 deepseek-v4_1-1017-4-gb10 Public

   Serve DeepSeek-V4.1-Flash-1017 on 4× DGX Spark / GB10 in a switchless QSFP ring — Engram from NVMe, 67 tok/s single-stream, 312 tok/s at 16 streams

   Python 1