Ray Smith f10ac41a20 Added flash attention, with both a single-q function, and a register-tiled function. ... The register-tiled version achieves a speed-up by a factor of about 9.7 over the previous attention function on an AVX3-enabled machine. PiperOrigin-RevId: 804913784		2025-09-09 08:05:26 -07:00
..
bindings	Replace mt19937 with new generator to enable parallel sampling	2025-09-04 23:49:10 -07:00
evals	Add MMLU eval to github	2024-05-20 10:20:53 -07:00
activations.h	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
api_client.cc	feature: add API server and client with Google protocol	2025-08-21 11:32:48 +09:00
api_server.cc	Replace mt19937 with new generator to enable parallel sampling	2025-09-04 23:49:10 -07:00
attention.cc	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
attention.h	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
configs.cc	Remove Griffin support	2025-09-05 02:35:40 -07:00
configs.h	Remove Griffin support	2025-09-05 02:35:40 -07:00
configs_test.cc	Minor: rename compression/shared -> types.h	2025-05-13 06:53:21 -07:00
flash_attention.cc	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
flash_attention.h	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
flash_attention_test.cc	Added flash attention, with both a single-q function, and a register-tiled function.	2025-09-09 08:05:26 -07:00
gemma-inl.h	MatMul simplification, threading strategy improvements	2025-09-03 21:45:07 -07:00
gemma.cc	1.15x speedup: parallel sampling, enabled by new RNG	2025-09-05 07:24:02 -07:00
gemma.h	Memory use reduction: smaller/single MMStorage	2025-09-09 05:32:46 -07:00
gemma_args.h	Matmul refactoring towards fusion	2025-09-09 07:13:38 -07:00
kv_cache.cc	Remove Griffin support	2025-09-05 02:35:40 -07:00
kv_cache.h	Remove Griffin support	2025-09-05 02:35:40 -07:00
model_store.cc	Remove Griffin support	2025-09-05 02:35:40 -07:00
model_store.h	Write SBS progressively.	2025-07-31 06:05:38 -07:00
run.cc	Replace mt19937 with new generator to enable parallel sampling	2025-09-04 23:49:10 -07:00
tensor_info.cc	Remove Griffin support	2025-09-05 02:35:40 -07:00
tensor_info.h	Remove Griffin support	2025-09-05 02:35:40 -07:00
tensor_info_test.cc	Minor: ModelWeightsPtrs -> WeightsPtrs	2025-07-11 06:11:51 -07:00
tokenizer.cc	(Resubmit) Prepare profiler annotations for new API	2025-08-13 01:38:24 -07:00
tokenizer.h	6x large-batch, short-prompt prefill speedup	2025-06-10 09:56:20 -07:00
vit.cc	Replace mt19937 with new generator to enable parallel sampling	2025-09-04 23:49:10 -07:00
vit.h	Minor: ModelWeightsPtrs -> WeightsPtrs	2025-07-11 06:11:51 -07:00
weights.cc	Matmul refactoring towards fusion	2025-09-09 07:13:38 -07:00
weights.h	Remove Griffin support	2025-09-05 02:35:40 -07:00