Enabling onednn Graph API from framework level

[akote123]

Enabling onednn Graph API from framework level

[TaoLv]

Hi @akote123, it's enabled in Intel Extension for TensorFlow (https://github.com/intel/intel-extension-for-tensorflow/tree/main/itex/core/graph/onednn_graph), but not in TensorFlow upstream yet.

[akote123]

@TaoLv ,Thank you. I have tried export ITEX_ONEDNN_GRAPH=1 flag in intel extension for tensorflow for hugging face bert and transformer model, But when I check onednn log, graph apis are not seen. Do I need to change anything in code?

[gyhintel]

@TaoLv ,Thank you. I have tried export ITEX_ONEDNN_GRAPH=1 flag in intel extension for tensorflow for hugging face bert and transformer model, But when I check onednn log, graph apis are not seen. Do I need to change anything in code?

Try to export _ITEX_ONEDNN_GRAPH_ALL_TYPE=1 flag. Only the quantized model would run into graph API by default in intel extension for tensorflow. For other float datatypes, need export the internal flag to enable graph API.

[akote123]

@gyhintel , Thank you . I will check with this flag.