[Bug] The "decode_first_stage" function in DDPM does not respect the "force_not_quantize" parameter

[wywywywy]

The decode_first_stage function in the ldm/models/diffusion/ddpm.py file looks like this.

    def decode_first_stage(self, z, predict_cids=False, force_not_quantize=False):
        if predict_cids:
            if z.dim() == 4:
                z = torch.argmax(z.exp(), dim=1).long()
            z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None)
            z = rearrange(z, 'b h w c -> b c h w').contiguous()

        z = 1. / self.scale_factor * z
        return self.first_stage_model.decode(z)

The predict_cids and force_not_quantize parameters are accepted but never used.

The last line in the old repo looks like this, which makes more sense: return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)

So the question is, will it break anything applying this change?

[richservo]

is this why it uses SO MUCH VRAM to decode? Above 2k isn't possible....so the new upscaller can't even be used on the 768 model.