tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
The FuseReductionEpilogue primitive currently supports fusing bias addition epilogues into reduction blocks. This commit extends the primitive to also support ReLU activation functions in epilogue blocks, enabling fusion of patterns...
### Description When converting a PyTorch model containing `squeeze` operation on a dimension that is not 1, TVM fails with an InternalError. PyTorch's `squeeze` operation silently ignores dimensions that are...
This commit extends the make_fused_bias_activation_pattern function to support PyTorch frontend's specific IR generation pattern for convolution operations with bias. When PyTorch models with bias=True are converted to Relax IR, the...
### summary The latest doc about installing tvm from source code cannot give enough guidance to install tvm successfully. 1. It seems that tvm depends on the os system libraries...
Bumps [pip](https://github.com/pypa/pip) from 22.1.1 to 25.3. Changelog Sourced from pip's changelog. 25.3 (2025-10-24) Deprecations and Removals Remove support for the legacy setup.py develop editable method in setuptools editable installs; setuptools...
![dependabot[bot] avatar](https://avatars.githubusercontent.com/in/29110?v=4 "dependabot[bot] avatar"){kind=avatar}
When running meta_schedule.tune_tir on a valid TIR module involving multi-dimensional access patterns, TVM crashes during the schedule rewriting phase (RewriteParallelVectorizeUnroll). ### Actual behavior ``` Traceback (most recent call last): File...
Running meta_schedule.tune_tir on a valid TIR module triggers a crash inside RewriteFuseSplitParallelVectorize -> Parallel with a ScheduleError. This seems to happen during initial population generation (SampleInitPopulation) in the evolutionary search,...
When attempting to build a TIR module containing `pragma_unroll_explicit` annotations with `None` values, TVM encounters a segmentation fault during the FlattenBuffer pass execution. `pragma_unroll_explicit=None` should represent using the compiler's default...
This PR introduces a temporary runtime patch to add missing non-persistent buffers (`position_ids`, `token_type_ids`) into the exported program’s state_dict. This resolves the KeyError that occurs during the Hugging Face to...
This change just keep stride terms order the same with fused loop order in `fuse` primitive. In symbolic circumstances, previous form suffer from simplification issues and would make the expression...
