18 Commits

Author	SHA1	Message	Date
Tianxing Wu	5ee7e2cf97	fixed var name	2025-09-11 10:43:00 +00:00
Tianxing Wu	aebafe7f55	xcd remap	2025-09-11 10:32:12 +00:00
coderfeli	6a600f0cff	fix unittest	2025-09-10 10:14:22 +00:00
Feng Shijie	4b49839c63	fix the tensor_view shape of C in gemm1-gate-up	2025-09-09 07:41:32 +00:00
Feng Shijie	08bae35abd	fix backward compatibility	2025-09-09 04:27:33 +00:00
Feng Shijie	b09b6cdce9	fix scale_m gather load for a8w8 moe	2025-09-09 04:18:48 +00:00
Feng Shijie	027f5311c6	support persist-mode for moe-gemm	2025-09-09 03:52:03 +00:00
Feng Shijie	a725bc08a7	fix gate-up mismatch when OutputNRepeat > 1	2025-09-09 03:52:03 +00:00
Feng Shijie	9efde19691	Refactor scale and bias function, encapsulate scale/bias pointer to a tensor view and load data by tilewise operation	2025-09-09 03:52:03 +00:00
coderfeli	946be930dd	update	2025-09-07 08:22:07 +00:00
lalala-sh	adc8db73e0	only padding N in epilogue	2025-09-05 02:50:47 +00:00
Feng Shijie	5c484a5672	Add bias for f16xf4 moe_flatmm	2025-08-28 08:02:50 +00:00
Feng Shijie	65b702454c	support swiglu activaion and use rcpf to accelerate silu	2025-08-26 12:32:29 +00:00
Feng Shijie	85976b0b87	use int64_t as expert stride to avoid overflow	2025-08-21 06:58:55 +00:00
Feng Shijie	f7f0306eea	fix gate-up when GU_NRepeat > 1	2025-08-18 18:43:37 +00:00
Feng Shijie	be55c0f9cb	add fp16xf4 moe	2025-08-18 17:28:11 +00:00
Feng Shijie	cc9c7b9e58	optimize gemm2 atomic_add pattern	2025-08-11 08:38:23 +00:00
Feng Shijie	6d3cbc7c0e	add moe_flatmm	2025-08-06 08:33:33 +00:00