Yaman Umuroglu
Yaman Umuroglu
We are seeing some FINN activation export tests randomly failing, first mentioned here on the FINN repo: https://github.com/Xilinx/finn/pull/134#issuecomment-634714886 Copy pasting one of the failed cases we are observing here: ```...
Dependencies: https://github.com/Xilinx/finn-hlslib/pull/109 This PR introduces a new `HLSCustomOp` called `StreamingEltwise`, which corresponds to the proposed finn-hlslib component with the same name. This layer is intended to generalize parameter-less, elementwise operations...
Update QONNX commit hash to get the latest changes: * https://github.com/fastmachinelearning/qonnx/pull/21 * https://github.com/fastmachinelearning/qonnx/pull/20 * https://github.com/fastmachinelearning/qonnx/pull/19 * https://github.com/fastmachinelearning/qonnx/pull/18 * https://github.com/fastmachinelearning/qonnx/pull/17
Dependencies: https://github.com/Xilinx/finn-hlslib/pull/110 Adds 1D support to the existing `Downsample` layer, including the necessary changes to the corresponding `convert_to_hls.InferConvInpGen` transformation, and a new testbench for both 1D and 2D downsampling including...
Dependency: https://github.com/Xilinx/finn-hlslib/pull/100 This PR introduces a `mem_mode` attribute for the `VectorVectorActivation` layer and implements `mem_mode="decoupled"` for `memstream`-generated weight streams, very similar to how this is implemented in `MatrixVectorActivation`. This enables...
Prior to generating stitched IP, FINN should check that the stream interfaces that are being stitched together are compatible. This compatibility should at least be the number of bits, but...
To speed up the compilation process for large models or large layers, it would make sense to have a caching mechanism for long-running transformations. The cached outputs would be persistent...
In QNNs we sometimes get "dead neurons" (for FC layers) or "dead channels" (for conv layers). Sometimes the output from the dot product is always guaranteed to fall within a...
Currently we extract data from .vcd traces for performance debugging purposes. This can get very slow for large networks as the traces can be huge, and the cycle-by-cycle behavior has...
Sometimes it's necessary to iterate over a design multiple times by changing some parameters to observe how the behavior changes. To speed up the dev process for these use cases,...