Model Training ‐ Comparison ‐ [Number of CPU Threads per Core]

_{Models | Logs | Graphs | Configs}

---

The name of the parameter speaks for itself. I have always wondered what it is and why it is needed.

Compared values:

• 1,

• 2 - BD,

• 4,

• 8,

• 16.

---

_DLR(step)

_Loss(epoch)

All the graphs merge into one of two according to a completely random principle.

---

To avoid the hassle of looking at a million grids of identical images, I kept only two.

Even from the graphs it was clear that the parameter most likely did not affect anything, and that turned out to be the case. When it is increased, the CPU load increases, but it doesn't have any impact on the quality of the results or the training speed.

---

CONCLUSION

The parameter does not affect the result. It is easier to leave the default value.

---

Next - Model Training ‐ Comparison - [Checkpoint]