This is the repo to run CI jobs with Ascend backend.
The ONNXRuntime source code is from main
branch of microsoft/onnxruntime
and will be run and tested daily with Ascend related.
Key | Value |
---|---|
CPU | Intel Xeon x86 |
NPU | Ascend310 |
OS | openEuler |
Period | UTC 1200 daily |
Branch | main |
Status | |
Recheck By Hand | comment 'recheck' in any issue |
Update on 2023.03.26
New provider option: flag of dump_om_model
.
When building an onnx model with CANN EP, the intermediate OM(offline model for Ascend NPU) is automatically saved. There are some users don't want to dump OM when resources are limited.
PR #20075 has resovled this situation with dump_om_model=False
.
PR #20138 for the doc update.
Update on 2024.03.16
PR #17365 avioded using patchelf but lost cann_dependencies
, PR #19929 adds cann_dependencies
to avoid require cann libraries when repairing wheel.
Update on 2023.07.18
PR #16506 changed the public constructor function MLFloat16(uint16_t x)
to private, and added a public function MLFloat16::FromBits(uint16_t x)
in the file include/onnxruntime/core/framework/float16.h
, which broke the CANN CI. This has been fixed by PR #16733 by replacing the constructor function MLFloat16()
with the public member function FromBits()
in the file onnxruntime/core/providers/cann/cann_common.cc
, but is waiting for upstream merge and this PR has been merged into upstream.
Update on 2023.07.06
This PR refactored the ExecutionProvider API for memory management, and broke the CANN EP build. This has been fixed by this PR, but is waiting for upstream merge.
Update on 2023.06.08
This PR introduced a missing registration of CANN Identity operator for version greater than 14. It has been fixed in this PR.
The Deepspeed source code is from main
branch of microsoft/deepspeed
and will be run and tested daily with Ascend related.
Key | Value |
---|---|
CPU | Arrch64 |
NPU | Ascend910B |
OS | Ubantu |
Period | UTC 1200 daily |
Branch | main |
Status | |
Recheck By Hand | comment 'recheck' in any issue |
TBD