Insights, benchmarks, and tutorials from the DMS Lab engineering team. Deep dives into AI inference, GPU performance, and agent integration patterns.
One OpenAI-compatible endpoint routes every request to the right tier: up to 720 tok/s, 85ms TTFT on B200. How DMS Lab AI works, and what to read first.
Real production benchmarks for Qwen 3.5 397B on NVIDIA B200: 720 tok/s peak throughput, 85ms TTFT, FP8 vs FP16 gains, and how to reproduce them.
Point Claude Code at DMS Lab AI in about three minutes: two environment variables, 1M-token context, 85ms first token. Setup, model picks, troubleshooting.
