THE COLLECTION
EVERY SIGNAL, SORTED.
Search the whole shelf by topic, tool, term, or plain old curiosity.
China, Global Open-Source Inference, Tencent YouTu Lab, Shanghai Jiao Tong University, Xiamen University, DSpine, Qwen3.8-27B, Speculative Decoding, SGLang, Draft Models, Throughput, Latency, Benchmarking and Production Reliability
Tencent made the draft model guess farther. Benchmark the whole serving stack
Tencent released a two-billion-parameter DSpine draft model for Qwen3.8-27B. Its selected tests show longer accepted drafts and higher throughput, but the useful number is not a headline multiplier. It is the gain your target model, draft, hardware, runtime and traffic shape can reproduce together.
Germany, Austria, Autonomous Driving, CLAP-AD, End-to-End Models, Imitation Learning, Speed Control, Collision Avoidance, Traffic Rules, Closed-Loop Evaluation, Benchmarking and Road Safety
German driving model separated the path from the speed. Test the safety score in traffic
CLAP-AD learns where to go from expert driving, then scores several possible speeds against collision, traffic-light and stop-sign criteria. The split is easier to inspect. It is not a road-safety certificate.
AI Code Review, Software Quality, Open Benchmarks, Evaluation Design, Developer Tools and Human Oversight
GitHub built a code-review benchmark that measures useful findings, not comment confetti
ReviewBench asks whether an AI reviewer finds real problems without burying developers in noise. Its public dataset is unusually inspectable, but a benchmark that anyone can study is also a benchmark agents can study for.