tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
To optimize i4_to_f16 decoding, we can use some advanced hardware instructions to do fast type conversion to alleviate the cost of decoding, we can do that by tensorize in tvm....
This pr is to address this issue:https://github.com/apache/tvm/issues/16490
This is a pull request for MSC(Multi-System Compile) RFC: https://discuss.tvm.apache.org/t/rfc-unity-msc-introduction-to-multi-system-compiler/15251/5 Tracking issue: https://github.com/apache/tvm/issues/15233 This is the Milestone 4 for MSC: Add plugin builder, enable plugin wrap in different frameworks. To...
Currently, target features are determined by a set of fixed checks on the target string. This works well for checking support of a small number of simple features, but it...
I used this code to compile an onnx network model on python with the suffix so. ```python onnx_model = onnx.load('mobilenetv2.onnx') input_name = 'input.1' shape_dict = {input_name: (1,3,224,224)} sym, params =...
Hi， tvm: This PR fixes a bug related to different opsets in the clip and unsqueeze operations. My model's opset default version is 10, and there is no implementation for...
Bumps [cryptography](https://github.com/pyca/cryptography) from 41.0.6 to 42.0.2. Changelog Sourced from cryptography's changelog. 42.0.2 - 2024-01-30 * Updated Windows, macOS, and Linux wheels to be compiled with OpenSSL 3.2.1. * Fixed an...
![dependabot[bot] avatar](https://avatars.githubusercontent.com/in/29110?v=4 "dependabot[bot] avatar"){kind=avatar}
* Handle non-bundled parameters in LazyTransformParams. * Check for `"num_input"` attribute * Handle relax.Const in LazyTransformParams Prior to this commit, `LazyTransformParams` would only output a call to the `fset_item` function...
Prior to this commit, `VerifyWellFormed` would state that a dataflow block or pure function contained an impure call, but identifying which call was impure was left to the user. This...
Prior to this commit, if the weight of a matmul a dynamic shape, and that matmul is being combined with the `CombineParallelMatmul` pass, it could cause a segfault when `dim.as()`...