philipwan

Fix IR gap for shape inference: update input from initializer (#2901)

1

* Update input from initializer if missing * add a test and comments * remove useless modification * fix flake8 * do not update shape if exisit in input *...

build tf v2.5.0 falied

5

According to your method, I compiled tensorflow_aarch64 on tf2.5, but it reported an error: `ERROR: /root/tensorflow/tensorflow/python/BUILD:2531:29: Linking of rule '//tensorflow/python:gen_candidate_sampling_ops_py_wrappers_cc' failed (Exit 1): gcc failed: error executing command /usr/bin/gcc @bazel-out/host/bin/tensorflow/python/gen_candidate_sampling_ops_py_wrappers_cc-2.params...

efficient_attention_forward_cutlass op is incompatible with Torch JIT

2

# 🐛 Bug RuntimeError: unsupported output type: int, from operator: xformers::efficient_attention_forward_cutlass ## Command ## To Reproduce StableDiffusionXL model use with torch.jit and enable_xformers_memory_efficient_attention function ``` m.enable_xformers_memory_efficient_attention torch.jit.trace(m) ``` Steps to...

According to examples file and implementation of **pipeline.py**, every time using new prompt, you need to warm up at least **denoising_steps_num** times. Am I understanding it wrong? If different prompts...

Can Customers use FM module for inference without training？