15,017 issues in apache/tvm
Hi @yzh119 @Hzfengsy This is a minor PR and it aims to update the copyright year from 2020 to 2023. I would appreciate that if you can help to review/manage it, thanks. #15043
hexp is used in MLC-LLM but we didn t include it in cuda_half_t.h thus some old GPU architectures (sm_52 or earlier) users cannot compile their models. This PR fix the issue.
Fix activation parameters for operations according to the values in Vela. cc @neildhickey, @lhutton1, @ekalda, @leandron
This PR improves the optimization of clip in simplify expr pass. Previously, there is a SimplifyCastClip class, it deals with the pattern cast- clip : it checks if the a_min/a_max values of clip equals ...
ci_arm was building llvm from source but this should no longer be required now we ve update to Ubuntu 20.04
When I load a TFlite model with relay.frontend.from_tflite API, I get an AttributeError, as show below: Actual behavior Traceback (most recent call last): File tflite_main.py , line 59, in module mod, ...
This reverts commit 7dc1d6a99f7984855edf1a5c42b66215addeb13f, since CI ARM has been broken on main and PR jobs since that commit. See https://github.com/apache/tvm/pull/14983#issuecomment-1583221170 for ...
This would be useful when creating multiple kv caches with the same shape. On A10G, compared to creating 64 kv caches separately in LLaMA from mlc-llm, doing a single allocation can save about 35 ms. ...
No arguments were passed to CopyComputeReordering pass and the same parameters were used for all targets. This fix takes arguments for CopyComputeReordering pass from Vela. In networks tests, the amount ...
This is needed for https://github.com/apache/tvm/pull/15046. It has been split from the PR since changing the CI config seems to require sending a PR from apache.

