ModuFlow/mask-ddpm
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update ks

Browse Source
This commit is contained in:
MingzheYang
2026-01-25 17:55:28 +08:00
parent 5ef1e465f9
commit cc10125fbf
4 changed files with 18 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/config.json
View File

@@ -35,9 +35,9 @@
"disc_mask_scale": 0.9, "disc_mask_scale": 0.9,
"cont_loss_weighting": "inv_std", "cont_loss_weighting": "inv_std",
"cont_loss_eps": 1e-6, "cont_loss_eps": 1e-6,
"cont_target": "x0", "cont_target": "v",
"cont_clamp_x0": 5.0, "cont_clamp_x0": 5.0,
"quantile_loss_weight": 0.3, "quantile_loss_weight": 0.1,
"quantile_points": [0.05, 0.25, 0.5, 0.75, 0.95], "quantile_points": [0.05, 0.25, 0.5, 0.75, 0.95],
"shuffle_buffer": 256, "shuffle_buffer": 256,
"sample_batch_size": 8, "sample_batch_size": 8,

4
example/export_samples.py
View File

@@ -201,6 +201,10 @@ def main():
if cont_clamp_x0 > 0: if cont_clamp_x0 > 0:
x0_pred = torch.clamp(x0_pred, -cont_clamp_x0, cont_clamp_x0) x0_pred = torch.clamp(x0_pred, -cont_clamp_x0, cont_clamp_x0)
eps_pred = (x_cont - torch.sqrt(a_bar_t) * x0_pred) / torch.sqrt(1.0 - a_bar_t) eps_pred = (x_cont - torch.sqrt(a_bar_t) * x0_pred) / torch.sqrt(1.0 - a_bar_t)
elif cont_target == "v":
v_pred = eps_pred
x0_pred = torch.sqrt(a_bar_t) * x_cont - torch.sqrt(1.0 - a_bar_t) * v_pred
eps_pred = torch.sqrt(1.0 - a_bar_t) * x_cont + torch.sqrt(a_bar_t) * v_pred
coef1 = 1.0 / torch.sqrt(a_t) coef1 = 1.0 / torch.sqrt(a_t)
coef2 = (1 - a_t) / torch.sqrt(1 - a_bar_t) coef2 = (1 - a_t) / torch.sqrt(1 - a_bar_t)
mean_x = coef1 * (x_cont - coef2 * eps_pred) mean_x = coef1 * (x_cont - coef2 * eps_pred)

11
example/sample.py
View File

@@ -114,15 +114,18 @@ def main():
t_batch = torch.full((batch_size,), t, device=DEVICE, dtype=torch.long) t_batch = torch.full((batch_size,), t, device=DEVICE, dtype=torch.long)
eps_pred, logits = model(x_cont, x_disc, t_batch, cond) eps_pred, logits = model(x_cont, x_disc, t_batch, cond)
# Continuous reverse step (DDPM): x_{t-1} mean
a_t = alphas[t]
a_bar_t = alphas_cumprod[t]
if cont_target == "x0": if cont_target == "x0":
x0_pred = eps_pred x0_pred = eps_pred
if cont_clamp_x0 > 0: if cont_clamp_x0 > 0:
x0_pred = torch.clamp(x0_pred, -cont_clamp_x0, cont_clamp_x0) x0_pred = torch.clamp(x0_pred, -cont_clamp_x0, cont_clamp_x0)
eps_pred = (x_cont - torch.sqrt(a_bar_t) * x0_pred) / torch.sqrt(1.0 - a_bar_t) eps_pred = (x_cont - torch.sqrt(a_bar_t) * x0_pred) / torch.sqrt(1.0 - a_bar_t)
elif cont_target == "v":
# Continuous reverse step (DDPM): x_{t-1} mean v_pred = eps_pred
a_t = alphas[t] x0_pred = torch.sqrt(a_bar_t) * x_cont - torch.sqrt(1.0 - a_bar_t) * v_pred
a_bar_t = alphas_cumprod[t] eps_pred = torch.sqrt(1.0 - a_bar_t) * x_cont + torch.sqrt(a_bar_t) * v_pred
coef1 = 1.0 / torch.sqrt(a_t) coef1 = 1.0 / torch.sqrt(a_t)
coef2 = (1 - a_t) / torch.sqrt(1 - a_bar_t) coef2 = (1 - a_t) / torch.sqrt(1 - a_bar_t)
mean = coef1 * (x_cont - coef2 * eps_pred) mean = coef1 * (x_cont - coef2 * eps_pred)

6
example/train.py
View File

@@ -62,7 +62,7 @@ DEFAULTS = {
"shuffle_buffer": 256, "shuffle_buffer": 256,
"cont_loss_weighting": "none", # none | inv_std "cont_loss_weighting": "none", # none | inv_std
"cont_loss_eps": 1e-6, "cont_loss_eps": 1e-6,
"cont_target": "eps", # eps | x0 "cont_target": "eps", # eps | x0 | v
"cont_clamp_x0": 0.0, "cont_clamp_x0": 0.0,
"quantile_loss_weight": 0.0, "quantile_loss_weight": 0.0,
"quantile_points": [0.05, 0.25, 0.5, 0.75, 0.95], "quantile_points": [0.05, 0.25, 0.5, 0.75, 0.95],
@@ -259,6 +259,10 @@ def main():
if float(config.get("cont_clamp_x0", 0.0)) > 0: if float(config.get("cont_clamp_x0", 0.0)) > 0:
x0_target = torch.clamp(x0_target, -float(config["cont_clamp_x0"]), float(config["cont_clamp_x0"])) x0_target = torch.clamp(x0_target, -float(config["cont_clamp_x0"]), float(config["cont_clamp_x0"]))
loss_base = (eps_pred - x0_target) ** 2 loss_base = (eps_pred - x0_target) ** 2
elif cont_target == "v":
a_bar_t = alphas_cumprod[t].view(-1, 1, 1)
v_target = torch.sqrt(a_bar_t) * noise - torch.sqrt(1.0 - a_bar_t) * x_cont
loss_base = (eps_pred - v_target) ** 2
else: else:
loss_base = (eps_pred - noise) ** 2 loss_base = (eps_pred - noise) ** 2