Training Performance issues and some feedback

[vtc-fs113371-project]

Hello, we are students from the Hong Kong VTC Academy.

Here are some problems and feedback .

---

Problems

We used 2 devices to run the same project at the same time .

1. i712700 14core (4ghz~) 16G 3200 DDR4 Ram RTX 4050 6GVRAM
2. 2x dual xeon-2696v3 36core (2.8ghz~) 32G 2133 DDR4 ECC Ram RTX 2080 Super 8GVRAM

First device (i7) Always trains very fast, about 5 times faster than the second device(xeon).

Why does this problem ?When training, are CPU threads used? CPU frequency? RAM frequency? I need to know what causes the speed difference?
This way I can assemble the appropriate hardware and server cluster for the training program.

---

Feedback

1. Teachable Machine all cpu threads should be used to achieve better performance and faster speed. Otherwise, it will take a very long time to run the training. : (

2. Web pages should cut out unimportant parts to improve performance, such as「class」The image preview should only show the first 10 photos. Now users can scroll back to see all the 「class」 photos, which will take up a lot of RAM and is not user-friendly.

3. (Maybe this is not important) We are using teachable machine We have created six facial emotion test modes. We already have perfect materials and databases (about 10000 photos). After one month of training with different parameters,We found that the model has some flaws in facial expressions. I think teachable machine It should be possible to develop a project specifically for training facial floating point models. : )