Atlas Inference
community
AI & ML interests
We built Atlas, a pure Rust LLM inference engine with custom CUDA kernels for the NVIDIA DGX Spark GB10, and we are gearing up for an open source release. 102 tok/s on Qwen3.5-35B 3x over vLLM on the same hardware, 2GB binary, 2 minute cold start (at LEAST 10x smaller and faster)
Recent Activity
Atlas-Inference 's datasets
None public yet