From 93d638d8b8b7d5226a8fa56e6d9b96c7710da994 Mon Sep 17 00:00:00 2001 From: Seunghoon Lee Date: Sun, 30 Apr 2023 11:15:09 +0900 Subject: [PATCH] Approx NN works fine. --- modules/dml/hijack/plms.py | 91 ++++++++++++++++++++++++++++++++++++++ modules/shared.py | 2 +- 2 files changed, 92 insertions(+), 1 deletion(-) create mode 100644 modules/dml/hijack/plms.py diff --git a/modules/dml/hijack/plms.py b/modules/dml/hijack/plms.py new file mode 100644 index 000000000..49979db28 --- /dev/null +++ b/modules/dml/hijack/plms.py @@ -0,0 +1,91 @@ +import torch + +from ldm.models.diffusion.ddim import noise_like + +import modules.sd_hijack_inpainting as plms_hijack + + +@torch.no_grad() +def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False, + temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, + unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None, dynamic_threshold=None): + b, *_, device = *x.shape, x.device + + def get_model_output(x, t): + if unconditional_conditioning is None or unconditional_guidance_scale == 1.: + e_t = self.model.apply_model(x, t, c) + else: + x_in = torch.cat([x] * 2) + t_in = torch.cat([t] * 2) + + if isinstance(c, dict): + assert isinstance(unconditional_conditioning, dict) + c_in = dict() + for k in c: + if isinstance(c[k], list): + c_in[k] = [ + torch.cat([unconditional_conditioning[k][i], c[k][i]]) + for i in range(len(c[k])) + ] + else: + c_in[k] = torch.cat([unconditional_conditioning[k], c[k]]) + else: + c_in = torch.cat([unconditional_conditioning, c]) + + e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2) + e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond) + + if score_corrector is not None: + assert self.model.parameterization == "eps" + e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs) + + return e_t + + alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas + alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev + sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas + sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas + + def get_x_prev_and_pred_x0(e_t, index): + # select parameters corresponding to the currently considered timestep + print(alphas[index]) # DML Solution: PLMS Sampling does not work without this print. + a_t = torch.full((b, 1, 1, 1), alphas[index], device=device) + a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device) + sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device) + sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device) + + # current prediction for x_0 + pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() + if quantize_denoised: + pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) + if dynamic_threshold is not None: + from ldm.models.diffusion.sampling_util import norm_thresholding + pred_x0 = norm_thresholding(pred_x0, dynamic_threshold) + # direction pointing to x_t + dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t + noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature + if noise_dropout > 0.: + noise = torch.nn.functional.dropout(noise, p=noise_dropout) + x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise + return x_prev, pred_x0 + + e_t = get_model_output(x, t) + if len(old_eps) == 0: + # Pseudo Improved Euler (2nd order) + x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t, index) + e_t_next = get_model_output(x_prev, t_next) + e_t_prime = (e_t + e_t_next) / 2 + elif len(old_eps) == 1: + # 2nd order Pseudo Linear Multistep (Adams-Bashforth) + e_t_prime = (3 * e_t - old_eps[-1]) / 2 + elif len(old_eps) == 2: + # 3nd order Pseudo Linear Multistep (Adams-Bashforth) + e_t_prime = (23 * e_t - 16 * old_eps[-1] + 5 * old_eps[-2]) / 12 + elif len(old_eps) >= 3: + # 4nd order Pseudo Linear Multistep (Adams-Bashforth) + e_t_prime = (55 * e_t - 59 * old_eps[-1] + 37 * old_eps[-2] - 9 * old_eps[-3]) / 24 + + x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t_prime, index) + + return x_prev, pred_x0, e_t +plms_hijack.p_sample_plms = p_sample_plms diff --git a/modules/shared.py b/modules/shared.py index 7c5916cfd..ee8ca7f0a 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -416,7 +416,7 @@ options_templates.update(options_section(('ui', "Live previews"), { "live_previews_enable": OptionInfo(True, "Show live previews of the created image"), "show_progress_grid": OptionInfo(True, "Show previews of all images generated in a batch as a grid"), "show_progress_every_n_steps": OptionInfo(1, "Show new live preview image every N sampling steps. Set to -1 to show after completion of batch.", gr.Slider, {"minimum": -1, "maximum": 32, "step": 1}), - "show_progress_type": OptionInfo("Approx cheap" if is_device_dml else "Approx NN", "Image creation progress preview mode", gr.Radio, {"choices": ["Full", "Approx NN", "Approx cheap"]}), # DML Solution: Use Approx cheap instead of Approx NN as a default progress type. + "show_progress_type": OptionInfo("Approx NN", "Image creation progress preview mode", gr.Radio, {"choices": ["Full", "Approx NN", "Approx cheap"]}), "live_preview_content": OptionInfo("Combined", "Live preview subject", gr.Radio, {"choices": ["Combined", "Prompt", "Negative prompt"]}), "live_preview_refresh_period": OptionInfo(250, "Progressbar/preview update period, in milliseconds") }))