GTC ON-DEMAND

 
SEARCH SESSIONS
SEARCH SESSIONS

Search All
 
Refine Results:
 
Year(s)

SOCIAL MEDIA

EMAIL SUBSCRIPTION

 
 

GTC ON-DEMAND

HPC and AI
Presentation
Media
Adding a custom CUDA C++ Operations in Tensorflow for boosting BERT Inference
Abstract:
This talk was designed to illustrate how having CUDA knowledge can help DL developers understand and tune their deep learning applications. We explain how to implement Tensorflow custom operations to utilize GPU more efficiently in running DL workloads, esp. BERT Inference for SQuAD. We also deliver the key insights on why the techniques introduced here can achieve better performance by discerning the profiling result.
 
Topics:
HPC and AI
Type:
Talk
Event:
AI Conference Korea
Year:
2019
Session ID:
SKR9108
Download:
Share: