colin
/
Witllm
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update unsuper.

This commit is contained in:
Colin 2024-10-05 16:17:41 +08:00
parent 45d5701835
commit 22464e7724
1 changed files with 9 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
unsuper/minist.py
View File

@ -115,43 +115,27 @@ for epoch in range(epochs):
images = images.to(device)
outputs = model.forward_unsuper(images)
# outputs = outputs.permute(0, 2, 3, 1) # 64 8 24 24 -> 64 24 24 8
# sample = outputs.reshape(-1, outputs.shape[3]) # -> 36864 8
# abs = torch.abs(sample)
# max, max_index = torch.max(abs, dim=1)
# min, min_index = torch.min(abs, dim=1)
# label = sample * 0.9
# all = range(0, label.shape[0])
# label[all, max_index] = label[all, max_index]*1.1
# loss = F.l1_loss(sample, label)
# model.conv1.weight.grad = None
# loss.backward()
outputs = outputs.permute(0, 2, 3, 1) # 64 8 24 24 -> 64 24 24 8
sample = outputs.reshape(outputs.shape[0], -1, outputs.shape[3]) # -> 64 24x24 8
sample = outputs.reshape(-1, outputs.shape[3]) # -> 36864 8
abs = torch.abs(sample)
sum = torch.sum(abs, dim=1, keepdim=False)
max, max_index = torch.max(sum, dim=1)
max, max_index = torch.max(abs, dim=1)
label = sample * 0.9
all = range(0, label.shape[0])
all_wh = range(0, 24 * 24)
label[all, :, max_index] = label[all, :, max_index] * 1.1
label[all, max_index] = label[all, max_index] * 1.1
loss = F.l1_loss(sample, label)
model.conv1.weight.grad = None
loss.backward()
# show.DumpTensorToImage(images.view(-1, images.shape[2], images.shape[3]), "input_image.png", Contrast=[0, 1.0])
# w = model.conv1.weight.data
# show.DumpTensorToImage(w.view(-1, w.shape[2], w.shape[3]), "conv1_weight.png", Contrast=[-1.0, 1.0])
# w = model.conv1.weight.grad
# show.DumpTensorToImage(w.view(-1, w.shape[2], w.shape[3]).cpu(), "conv1_weight_grad.png")
model.conv1.weight.data = model.conv1.weight.data - model.conv1.weight.grad * 1000
# w = model.conv1.weight.data
# show.DumpTensorToImage(w.view(-1, w.shape[2], w.shape[3]), "conv1_weight_update.png", Contrast=[-1.0, 1.0])
model.conv1.weight.data = model.conv1.weight.data - model.conv1.weight.grad * 100
if (i + 1) % 100 == 0:
print(f"Epoch [{epoch+1}/{epochs}], Step [{i+1}/{n_total_steps}], Loss: {loss.item():.8f}")
w = model.conv1.weight.grad
show.DumpTensorToImage(w.view(-1, w.shape[2], w.shape[3]).cpu(), "conv1_weight_grad.png")
w = model.conv1.weight.data
show.DumpTensorToImage(w.view(-1, w.shape[2], w.shape[3]), "conv1_weight_update.png", Contrast=[-1.0, 1.0])
# Train the model
model.conv1.weight.requires_grad = False
model.conv2.weight.requires_grad = True