725fbd2ed0
[Inference] Remove unnecessary float4_ and rename float8_ to float8 ( #5679 )
2024-05-06 10:55:34 +08:00
ef8e4ffe31
[Inference/Feat] Add kvcache quant support for fused_rotary_embedding_cache_copy ( #5680 )
2024-04-30 18:33:53 +08:00
8ccb6714e7
[Inference/Feat] Add kvcache quantization support for FlashDecoding ( #5656 )
2024-04-26 19:40:37 +08:00
279300dc5f
[Inference/Refactor] Refactor compilation mechanism and unified multi hw ( #5613 )
...
* refactor compilation mechanism and unified multi hw
* fix file path bug
* add init.py to make pybind a module to avoid relative path error caused by softlink
* delete duplicated micros
* fix micros bug in gcc
2024-04-24 14:17:54 +08:00
a2878e39f4
[Inference] Add Reduce Utils ( #5537 )
...
* add reduce utils
* add using to delele namespace prefix
2024-04-01 15:34:25 +08:00
934e31afb2
The writing style of tail processing and the logic related to macro definitions have been optimized. ( #5519 )
2024-03-28 10:42:51 +08:00
87079cffe8
[Inference]Support FP16/BF16 Flash Attention 2 And Add high_precision Flag To Rotary Embedding ( #5461 )
...
* Support FP16/BF16 Flash Attention 2
* fix bugs in test_kv_cache_memcpy.py
* add context_kv_cache_memcpy_kernel.cu
* rm typename MT
* add tail process
* add high_precision
* add high_precision to config.py
* rm unused code
* change the comment for the high_precision parameter
* update test_rotary_embdding_unpad.py
* fix vector_copy_utils.h
* add comment for self.high_precision when using float32
2024-03-25 13:40:34 +08:00
7ff42cc06d
add vec_type_trait implementation ( #5473 )
2024-03-19 18:36:40 +08:00
48c4f29b27
refactor vector utils
2024-03-19 11:32:01 +08:00
388e043930
add implementatino for GetGPULaunchConfig1D
2024-03-14 11:13:40 +08:00
f366a5ea1f
[Inference/kernel]Add Fused Rotary Embedding and KVCache Memcopy CUDA Kernel ( #5418 )
...
* add rotary embedding kernel
* add rotary_embedding_kernel
* add fused rotary_emb and kvcache memcopy
* add fused_rotary_emb_and_cache_kernel.cu
* add fused_rotary_emb_and_memcopy
* fix bugs in fused_rotary_emb_and_cache_kernel.cu
* fix ci bugs
* use vec memcopy and opt the gloabl memory access
* fix code style
* fix test_rotary_embdding_unpad.py
* codes revised based on the review comments
* fix bugs about include path
* rm inline
2024-03-13 17:20:03 +08:00
b699f54007
optimize rmsnorm: add vectorized elementwise op, feat loop unrolling ( #5441 )
2024-03-12 17:48:02 +08:00
5eb5ff1464
refactor code
2024-03-08 15:41:14 +08:00
a46598ac59
add reusable utils for cuda
2024-03-08 14:53:29 +08:00
Steve Luo