MemoryError when training tensorflow OCR

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

When trying a tensorflow ocr training example here OCR with Keras, TensorFlow, and Deep Learning it produce an error “MemoryError” with no details.
I tried the same example on my PC (16G RAM) with no NVIDIA gpu. it works fine

I’m using the latest JetPacke with the latest tensorflow from NVIDIA.

2

I’m trying to train another tensorflow model but unfortunatlly I got a memory issue too

tensorflow/core/common_runtime/bfc_allocator.cc:434] Allocator (GPU_0_bfc) ran out of memory trying to allocate 675.35MiB (rounded to 708157440)

Is Jetson Nano relly ready to handle tensorflow model training?

4

Hi,

tensorflow/core/common_runtime/bfc_allocator.cc:434] Allocator (GPU_0_bfc) ran out of memory …

According to your log, the issue is out of memory.

Please noticed that Nano only has 4G memory and this will limit you from training a complex model.
In general, we recommend desktop GPU for training job since Nano is designed as an edge device (inference).

Thanks.

1 Like