Diogo
Diogo
Adds support for 16 more onnx ops. Also replaced numpy.triu usage in llama and stable diffusion. Might conflict with #935, will update this pr if it gets merged first Closes...
When using the where operation with a tensor that contains infinity nan is returned. Seems the issue is in the false condition as float("inf")*0 = NaN This behavior can be...
When using cstyle runtimes like METAL an intermediate cast op has no effect leading to differing behaviors compared to other runtimes like torch. The kernels generated will use the initial...
Maintain input tensor dtypes on CMPEQ op for pytorch and in the tri functions
Adds float64 support to the different runtimes, also included an excluded field on dtypes for unsupported platforms From my testing float64 is not supported on metal and OpenCL on M2...
When trying to run the CLANG runtime on windows I get an error when building the lib. The issues stems from [here](https://github.com/geohot/tinygrad/blob/2d567ef688473734cd21f31ae9344d06d01d6ec4/tinygrad/runtime/ops_clang.py#LL13C11-L13C11) as this is building a shared library for...
Adds floor/ceil ops to tensor and Onnx ops Onnx ops after: 190 failed, 654 passed, 1790 skipped, 2 warnings in 15.17s Tested on: - METAL - CUDA - GPU -...
Floor OP
Adds floor op as per the suggestion in discord discussion, avoiding cast to int Added tests for edge cases when values don't have a decimal Tested on: - CPU -...
In collab with @[martinloretzzz](https://github.com/geohot/tinygrad/commits?author=martinloretzzz) Closes #1030
Int32 dtype test coverage is missing, add int32 tests to `test/test_dtype.py`. In part should also resolve the issues with `ops_cpu`, as numpy will upcast an op between int32 and float32...