Simple slope for hyper network training

[Heathen]

If anyone wants to implement this or do it better and create a pull request, this is just a quick and dirty linear slope display for training. If anyone wants to really tackle this, could use gradients or curve fitting.

Change

stable-diffusion-webui/modules/hypernetworks/hypernetwork.py

Lines 437 to 442 in e725474

	if len(previous_mean_losses) > 1:
	std = stdev(previous_mean_losses)
	else:
	std = 0
	dataset_loss_info = f"dataset loss:{mean(previous_mean_losses):.3f}" + u"\u00B1" + f"({std / (len(previous_mean_losses) ** 0.5):.3f})"
	pbar.set_description(dataset_loss_info)

To

        if len(previous_mean_losses) > 1:
            std = stdev(previous_mean_losses)
        else:
            std = 0
        if len(previous_mean_losses) > 4:
            loss_slope = mean(previous_mean_losses[len(previous_mean_losses)//2:]) - mean(previous_mean_losses[:len(previous_mean_losses)//2])
        else:
            loss_slope = 0.0
        dataset_loss_info = f"dataset loss:{mean(previous_mean_losses):.3f}" + u"\u00B1" + f"({std / (len(previous_mean_losses) ** 0.5):.3f})" + f" Slope: {loss_slope:.6f}"
        pbar.set_description(dataset_loss_info)

And change

stable-diffusion-webui/modules/hypernetworks/hypernetwork.py

Lines 499 to 505 in e725474

	<p>
	Loss: {previous_mean_loss:.7f}<br/>
	Step: {hypernetwork.step}<br/>
	Last prompt: {html.escape(entries[0].cond_text)}<br/>
	Last saved hypernetwork: {html.escape(last_saved_file)}<br/>
	Last saved image: {html.escape(last_saved_image)}<br/>
	</p>

To

<p>
Loss: {previous_mean_loss:.7f}<br/>
Slope: {loss_slope:.7f}<br/>
Step: {hypernetwork.step}<br/>
Last prompt: {html.escape(entries[0].cond_text)}<br/>
Last saved hypernetwork: {html.escape(last_saved_file)}<br/>
Last saved image: {html.escape(last_saved_image)}<br/>
</p>

[Heathen]

Actually, implementing TensorBoard would be a better endeavor:

https://pytorch.org/tutorials/recipes/recipes/tensorboard_with_pytorch.html