周周周 comments

Results 7 comments of


                                            周周周

[Paddle-TRT] constnt-folding

我关掉它

why 7234 和 8406 diffs in layer's output dimensions when constructing network

> Yes, that was a bug already fixed in TRT 8.4. I think is was not a bug. It seems like this: in 7234, shape infer is carried out when...

why 7234 和 8406 diffs in layer's output dimensions when constructing network

> > Yes, that was a bug already fixed in TRT 8.4. > > I think is was not a bug. It seems like this: in 7234, shape infer is...

Scale: scale (Output: tmp_131142): only activation types allowed as input to this layer

oh! I know , Scale's input dtype is int , should be int32!

how to build a whl install package?

python -m pip wheel ./python --wheel-dir=/zhoukangkang/triton/built_wheel_by_zkk --no-deps

Can Triton support split-k matrix multiplication

> The provided triton.ops.matmul appears to do so: https://github.com/openai/triton/blob/main/python/triton/ops/matmul.py#L96 3Q

BF16 is slower than fp16 of TensorRT 9.1 when running my R50 model on A800 GPU

> ## Description > ## Environment > **TensorRT Version**: > > **NVIDIA GPU**: > > **NVIDIA Driver Version**: > > **CUDA Version**: > > **CUDNN Version**: > > Operating System:...