DuoDiscover agents auto-build fast eval proxies, scoring 2.68x TPU kernel speedup
berkeley_ai · x · 2026-10-16
DuoDiscover tackles the slow-evaluation bottleneck in AI-driven systems optimization by having agents co-discover fast evaluation proxies ("yardsticks") alongside the code they optimize.
- 2.68× TPU kernel speedup, 1.92× faster vLLM serving
- Real optimizations merged into mainline branches (e.g., FreeToken #619)
- First superhuman CPU prefetcher: +3.93% / +6.31% overall, beating AlphaEvolve
- #1 on NVIDIA SOL-ExecBench (B200)
A collaboration across Google Cloud, Berkeley Sky and Stanford AI Lab. The team stresses these are real merged PRs beating human-expert hardware baselines, not benchmark gaming.
Related event: DuoDiscover Tops NVIDIA SOL-ExecBench, Speeds Up TPU Kernels 2.68x(9 posts)→
More from Infra
- GitHub rewrote the 800k-line Copilot runtime in Rust using AI agents, cutting CPU and memory by over 90% — bibryam · 2026-10-16
- NVIDIA quietly delists the entry-level RTX 5050 from its marketplace, following the RTX 5090 — chemist_slime · 2026-10-16
- Grok claims 816B MoE model runs locally on dual M5 Ultra Mac Studios via MLX — grok · 2026-10-16
- Same 1,000 Cloudflare Durable Objects: $4,050/month cut to $22 with one trick — threepointone · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16