i-robot
f785fabcf2
!66157 add feature dvm
...
Merge pull request !66157 from looop5/feature_dvm_3_7
2024-03-08 12:07:00 +00:00
e1b550ce8b
回退 'Pull Request !65980 : pijit one stage'
2024-03-08 02:38:04 +00:00
eb7b129656
add feature dvm
...
fix conflict
2024-03-07 22:35:05 +08:00
i-robot
447e604edf
!65121 No graph Automatic differentiation
...
Merge pull request !65121 from luochao60/pyboost-dev
2024-03-07 13:58:18 +00:00
i-robot
f97bc69373
!65980 pijit one stage
...
Merge pull request !65980 from r1chardf1d0/r2.3
2024-03-07 13:20:02 +00:00
luochao
21a422c2fe
fix bug
2024-03-07 12:07:23 +08:00
4cc6add445
clean code
2024-03-07 11:00:39 +08:00
2530f311c8
AKG-MLIR: fuse dynamic-shape operators in graphkernel
...
* add SymbolEngine to frontend for dynamic shape nodes. save the symbolic_shape and symbolic_value to Abstract.
* add symbol_ops_impl to core/ops, which implement the infering symbolic shape and/or symbolic value functions.
* use "enable_dynamic_shape_fusion" to fuse dynamic-shape nodes in graphkernel.
* add dynamic_akg_cpu_kernel_mod and dynamic_akg_gpu_kernel_mod for akg-mlir kernels.
* use symbol_engine_jit to optimize the infershape of dynamic akg kernels.
* optimize the frontend graph with symbol engine, in the pass "symbol_engine_optimizer"
* fuse the shape calculation nodes to a KernelPacket node, to optimize the runtime pipeline.
2024-03-06 22:54:32 +08:00
be7d16423a
Implicit type conversion by signatures.
2024-03-04 23:14:02 +08:00
lilinjie
e76bb29d23
host api adapt: ReLU
2024-03-01 16:47:49 +08:00
i-robot
27ef7b4fcd
!63956 refactor cell call func
...
Merge pull request !63956 from chujinjin/refactor_cell_call_func
2024-02-29 07:07:43 +00:00
i-robot
0837b1d299
!65168 Add testcases for dataset_sink_mode
...
Merge pull request !65168 from maning202007/r2.3
2024-02-29 03:50:11 +00:00
eb90c6cb37
Add testcases for dataset_sink_mode
...
Add testcase for hccl dump
fix random failure for ge dump testcases
2024-02-27 19:26:51 +08:00
panzhihui
b6722754f2
CpuKernel use cann pakcage library and header files
2024-02-24 14:36:39 +08:00
cb72044510
refactor cell call func
2024-02-22 16:50:05 +08:00
panzhihui
78a5f2b6e7
Remove cust_aicpu flag in ir
2024-02-21 10:11:28 +08:00
panzhihui
887810fd39
Add op ge support
2024-02-19 15:29:31 +08:00
0b0f24c7aa
Revert "回退 'Pull Request !65105 : merge pijit project'"
...
This reverts commit 9480bc3f7f
2024-02-17 21:19:34 +08:00
yanghaoran
9480bc3f7f
回退 'Pull Request !65105 : merge pijit project'
2024-02-06 11:20:59 +00:00
i-robot
d299bcc422
!65103 kbk dump
...
Merge pull request !65103 from maning202007/feature-2.3-dump
2024-02-06 07:30:55 +00:00
1a5e0da93b
json acl
2024-02-05 16:06:26 +08:00
yanghaoran
0cd72b78c3
upgrade ascend 20231101 r2.3
...
fix ms adapter bug
Revert "upgrade ascend 20231101 r2.3"
This reverts commit fc74ed6fc88bf3799dd26e59f6f2443b7f225976.
2024-02-05 11:04:19 +08:00
952c017b6c
add eval cnode function
...
add ut test for example
2024-02-05 11:04:19 +08:00
38e7654e8a
Add fft/ifft/idct/correlate kernel and func api
2024-01-26 17:16:03 +08:00
fengyixing
97c4a6e8b0
回退 'Pull Request !64600 : add idct/fft/ifft at cpu/aicpu backend & correlate at cpu/aicpu/gpu backend'
2024-01-26 03:47:09 +00:00
ffb8171623
add idct at cpu/aicpu backend
2024-01-25 10:23:28 +08:00
e34475c4d9
use cann inc api
2024-01-23 19:38:40 +08:00
zjun
07e001ae6d
Opt grad execute by pyboost
...
Signed-off-by: zjun <zhangjun0@huawei.com>
2024-01-17 11:14:50 +08:00
i-robot
8598d2c510
!62796 aclnn ops: silu/silu_grad and sigmoid/sigmoid_grad
...
Merge pull request !62796 from jiaxueyu/pyboost
2024-01-16 02:25:01 +00:00
34c7048859
pyboost ops: silu/silu_grad and sigmoid/sigmoid_grad
2024-01-15 21:00:24 +08:00
lizhenyu
915fc71879
[Refactor] Delete deprecated interface in KernelTensor
2024-01-13 19:15:03 +08:00
hw_hz
414d867e0c
1. softmax/softmax_backward pyboost
...
2. replace origin softmax grad with SoftmaxBackward.
3. Softmax aclnnKernelMod
4. unify testcase of softmax
5. Format::ND no need to transpose to device shape, modify class Kernel
6. GetValue() in CreateInputAddress
7. SoftmaxCPU register fp16 and fp64
8. Softmax cpu implementation with parallel
9. Tensor.softmax
del softmax_
2024-01-09 15:03:17 +08:00
4d633c5483
sync python cal_tuple_slice_mask to C++
2024-01-05 16:22:17 +08:00
b9738861f2
fix some bugs of to_enum in arg_handler
2024-01-01 23:10:41 +08:00
ec3d9f0855
linear sum assignment op support all dtypes
2023-12-22 10:31:32 +08:00
i-robot
116bd7e29d
!62962 add large model inference test case
...
Merge pull request !62962 from zhangminli/llama2-2.3
2023-12-19 09:28:34 +00:00
i-robot
370abc06b2
!63201 optimize trace in Cell
...
Merge pull request !63201 from NaCN/optimize_trace
2023-12-19 06:29:15 +00:00
019e879e82
add large model inference test case
2023-12-19 09:33:33 +08:00
huangchengnuo
64a01fe2c4
optimize trace in Cell
2023-12-15 09:57:30 +08:00
dba0ec3fc6
optimize setitem with tuple contains slice
2023-12-14 15:01:35 +08:00
i-robot
d0406cb6dc
!63045 [LITE] sync r2.2.10 ascendc code
...
Merge pull request !63045 from mengyuanli/r2.3
2023-12-13 06:26:49 +00:00
58a648eaba
add PromptKVCache and DecoderKVCache ascend op
...
inplace update for ge
fix bug of prompt kv cache
fix buf of kernel sync problem
prompt kvcache add multi batch index
prompt kvcache and decoder kvcache support multi data type
decoder kvcache jump -1 seq_len
prompt kvcache jump -1 batch index
fix bug of decoder_kv_cache not support dig batch size
prompt_kv_cache support valid_seq_len
prompt_kv_cache and decoder_kv_cache support BSD format
make prompt_kv_cache and decoder_kv_cache support BSD format
2023-12-13 10:04:58 +08:00
17742cd102
add runtime committer
2023-12-12 08:45:18 +08:00
7e698a5c9a
support override getattr
2023-12-08 09:15:57 +08:00
bb6ac254c7
add runtime committer
2023-11-21 10:39:47 +08:00
i-robot
8b75c2270c
!61642 Fix range problem and a in b problem
...
Merge pull request !61642 from LiangZhibo/range
2023-11-13 03:13:07 +00:00
i-robot
1da59bf782
!61455 Support starred_expression in graph mode.
...
Merge pull request !61455 from Margaret_wangrui/starred_expression_unpack_r2.3
2023-11-11 02:16:15 +00:00
liangzhibo
1484a1a608
Fix range with empty output
2023-11-10 23:10:17 +08:00
cacd854fed
Support starred_expression in graph mode.
2023-11-09 11:28:21 +08:00
ligan
2d7808f8ac
fix ut bug.
2023-11-08 22:35:08 +08:00
i-robot
zhunaipan
luochao
lilinjie
panzhihui
yanghaoran
fengyixing
zjun
lizhenyu
hw_hz
huangchengnuo
liangzhibo
ligan