* Initial notes on TensorFlow/GPU troubleshooting.
This commit is contained in:
20
tensorflow/20220601.GPU-troubleshooting.md
Normal file
20
tensorflow/20220601.GPU-troubleshooting.md
Normal file
@@ -0,0 +1,20 @@
|
|||||||
|
Troubleshooting TensorFlow / KERAS GPU Process
|
||||||
|
==============================================
|
||||||
|
|
||||||
|
|
||||||
|
Common issues:
|
||||||
|
|
||||||
|
* Model too big
|
||||||
|
(print the KERAS model---is the number of parameters doable with the GPU?)
|
||||||
|
|
||||||
|
* Batch size too large
|
||||||
|
|
||||||
|
* Memory not released from previous processing
|
||||||
|
|
||||||
|
(use `with tf.Session() as sess:` code block)
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
Reference in New Issue
Block a user