Ray Smith 7b55d41f46 Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention. ... PiperOrigin-RevId: 868146247		2026-02-13 01:58:48 -08:00
..
testdata	Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention.	2026-02-13 01:58:48 -08:00
benchmark.cc	Abort if args are unrecognized, refactor argument passing	2025-12-15 03:18:45 -08:00
benchmark_helper.cc	Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention.	2026-02-13 01:58:48 -08:00
benchmark_helper.h	Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention.	2026-02-13 01:58:48 -08:00
benchmarks.cc	Abort if args are unrecognized, refactor argument passing	2025-12-15 03:18:45 -08:00
cross_entropy.cc	Major cleanup of profiler zones, add Caller annotation for all pool.Run	2025-10-23 01:54:24 -07:00
cross_entropy.h	Move MatMulEnv out of Gemma to enable concurrent calls	2025-06-23 01:20:09 -07:00
debug_prompt.cc	Abort if args are unrecognized, refactor argument passing	2025-12-15 03:18:45 -08:00
gemma_batch_bench.cc	Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention.	2026-02-13 01:58:48 -08:00
gemma_test.cc	Abort if args are unrecognized, refactor argument passing	2025-12-15 03:18:45 -08:00
prompts.h	Benchmark gemma.cpp with different length inputs.	2024-10-10 15:59:26 -07:00
run_mmlu.cc	Abort if args are unrecognized, refactor argument passing	2025-12-15 03:18:45 -08:00
wheat_from_chaff_test.cc	Rewrote flash attention to use BF16, transpose k and v, rewrote the task distribution, increase parallelism on decode, and use double the registers for the core of flash attention.	2026-02-13 01:58:48 -08:00