Upload 0001-karras-v2-experimental.patch

patch/0001-karras-v2-experimental.patch

@@ -0,0 +1,57 @@
+From 36078b25801787f0a0f145143637f46d33d8c389 Mon Sep 17 00:00:00 2001
+From: Ashen <git123@gmail.com>
+Date: Fri, 7 Apr 2023 22:04:35 -0700
+Subject: [PATCH] karras v2 experimental
+
+---
+ k_diffusion/sampling.py | 36 ++++++++++++++++++++++++++++++++++++
+ 1 file changed, 36 insertions(+)
+
+diff --git a/k_diffusion/sampling.py b/k_diffusion/sampling.py
+index f050f88..4d5df2a 100644
+--- a/k_diffusion/sampling.py
++++ b/k_diffusion/sampling.py
+@@ -605,3 +605,39 @@ def sample_dpmpp_2m(model, x, sigmas, extra_args=None, callback=None, disable=No
+             x = (sigma_fn(t_next) / sigma_fn(t)) * x - (-h).expm1() * denoised_d
+         old_denoised = denoised
+     return x
++
++
++@torch.no_grad()
++def sample_dpmpp_2m_test(model, x, sigmas, extra_args=None, callback=None, disable=None):
++    """DPM-Solver++(2M)."""
++    extra_args = {} if extra_args is None else extra_args
++    s_in = x.new_ones([x.shape[0]])
++    sigma_fn = lambda t: t.neg().exp()
++    t_fn = lambda sigma: sigma.log().neg()
++    old_denoised = None
++
++    for i in trange(len(sigmas) - 1, disable=disable):
++        denoised = model(x, sigmas[i] * s_in, **extra_args)
++        if callback is not None:
++            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
++        t, t_next = t_fn(sigmas[i]), t_fn(sigmas[i + 1])
++        h = t_next - t
++
++        t_min = min(sigma_fn(t_next), sigma_fn(t))
++        t_max = max(sigma_fn(t_next), sigma_fn(t))
++
++        if old_denoised is None or sigmas[i + 1] == 0:
++            x = (t_min / t_max) * x - (-h).expm1() * denoised
++        else:
++            h_last = t - t_fn(sigmas[i - 1])
++
++            h_min = min(h_last, h)
++            h_max = max(h_last, h)
++            r = h_max / h_min
++
++            h_d = (h_max + h_min) / 2
++            denoised_d = (1 + 1 / (2 * r)) * denoised - (1 / (2 * r)) * old_denoised
++            x = (t_min / t_max) * x - (-h_d).expm1() * denoised_d
++
++        old_denoised = denoised
++    return x
+
+-- 
+2.40.0
+