Memory leak in MacOS version

[avijit-chakroborty]

Describe the bug
The memory consumption keeps growing when using the library.

To Reproduce
Steps to reproduce the behavior:

1. Build the phi3 command line application from the examples/c/src directory
2. Run phi3 in one terminal window
3. On a separate terminal window, run the ps -p $(pgrep phi3) -o rss= command in a loop
4. Now ask questions and watch the phi3 output.
5. Simultaneously, in the second window observe that the memory usage grows over time with new questions asked.

Here's a small shell script that helps you to see the memory usage in 5 second intervals:

$ while true; do
while> mem_usage=$(ps -p $(pgrep phi3) -o rss=)
while> mem_usage_mb=$(echo "scale=2; $mem_usage/1024" | bc)
while> echo "Memory usage: $mem_usage_mb MB"
while> sleep 5
while> done

Observed memory usage:

Memory usage: 680.28 MB
Memory usage: 680.28 MB
Memory usage: 927.48 MB
Memory usage: 927.48 MB
Memory usage: 927.48 MB
Memory usage: 1019.68 MB
Memory usage: 1019.68 MB
Memory usage: 1019.68 MB
Memory usage: 1019.68 MB
Memory usage: 1019.68 MB
Memory usage: 1054.14 MB
Memory usage: 1056.70 MB
Memory usage: 1056.70 MB
Memory usage: 1061.62 MB
Memory usage: 1062.57 MB
Memory usage: 1127.75 MB
Memory usage: 1127.93 MB
Memory usage: 1128.82 MB
Memory usage: 1129.09 MB

The Android version of the same release doesn't show this behavior, just an FYI.

Expected behavior
Memory usage will stay within the certain bounds.

Desktop (please complete the following information):

• OS: macOS Sequoia
• Browser N/A
• Version 15.1

Smartphone (please complete the following information):

• N/A

Additional context
N/A

[skyline75489]

Thanks for the report! I'll take a look.

[skyline75489]

@avijit-chakroborty Are you using Phi-3-mini-4k-instruct-onnx? I'm seeing a memory usage capped at ~2.5GB on a M1 MacBook Pro 16GB.

Could you provide more information? For example the model of the device you're using?