capy-flash
An inference engine developed in C++/CUDA, likely intended for local large-model inference or related computational tasks.
Timeline (last 90 days)
The tester selected 9 real regression bugs from the private commit history of the capy-flash C++/CUDA inference engine for comparative testing.
The test has clear limitations: all 9 tasks come from a single codebase (the GPU inference domain), the bug reports were written in Italian, Claude and the local model use different harnesses, and the judge itself is a Claude model, so the sample confidence interval is about ±0.2.
All Facts (2)
The tester selected 9 real regression bugs from the private commit history of the capy-flash C++/CUDA inference engine for comparative testing.
50%UnverifiedThe test has clear limitations: all 9 tasks come from a single codebase (the GPU inference domain), the bug reports were written in Italian, Claude and the local model use different harnesses, and the judge itself is a Claude model, so the sample confidence interval is about ±0.2.
50%