transfer the ema checkpoint to the clean version.

app.py

@@ -66,8 +66,11 @@ def process_multi_wrapper_only_show_rendered(rendered_txt_0, rendered_txt_1, ren
                                      shared_eta, shared_a_prompt, shared_n_prompt, 
                                      only_show_rendered_image=True)  
 
+# cfg = OmegaConf.load("config.yaml")
+# model = load_model_from_config(cfg, "model_states.pt", verbose=True)
+
 cfg = OmegaConf.load("config.yaml")
-model = load_model_from_config(cfg, "model_states.pt", verbose=True)
+model = load_model_from_config(cfg, "model.ckpt", verbose=True)
 
 ddim_sampler = DDIMSampler(model)
 render_tool = Render_Text(model)

config.yaml

@@ -18,7 +18,7 @@ model:
     scale_factor: 0.18215
     only_mid_control: False
     sd_locked: True
-    use_ema: False #True #False #True #False
+    use_ema: False #True #TODO: specify
 
     control_stage_config:
       target: cldm.cldm.ControlNet
@@ -85,4 +85,4 @@ model:
       params:
         freeze: True
         layer: "penultimate"
-        device: "cpu"
+        device: "cpu" #TODO: specify

config_cuda.yaml

@@ -0,0 +1,88 @@
+model:
+  base_learning_rate: 1.0e-6 #1.0e-5 #1.0e-4
+  target: cldm.cldm.ControlLDM
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    control_key: "hint"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false
+    conditioning_key: crossattn
+    monitor: #val/loss_simple_ema
+    scale_factor: 0.18215
+    only_mid_control: False
+    sd_locked: True
+    use_ema: True #TODO: specify
+
+    control_stage_config:
+      target: cldm.cldm.ControlNet
+      params:
+        use_checkpoint: True
+        image_size: 32 # unused
+        in_channels: 4
+        hint_channels: 3
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_head_channels: 64 # need to fix for flash-attn
+        use_spatial_transformer: True
+        use_linear_in_transformer: True
+        transformer_depth: 1
+        context_dim: 1024
+        legacy: False
+
+    unet_config:
+      target: cldm.cldm.ControlledUnetModel
+      params:
+        use_checkpoint: True
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_head_channels: 64 # need to fix for flash-attn
+        use_spatial_transformer: True
+        use_linear_in_transformer: True
+        transformer_depth: 1
+        context_dim: 1024
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          #attn_type: "vanilla-xformers"
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenOpenCLIPEmbedder
+      params:
+        freeze: True
+        layer: "penultimate"
+        # device: "cpu" #TODO: specify

model.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5f82f4af7d69b0ffdff6bf3d1b8dc6b13bbf81e28ea0fbacbf68824d2c1f652
+size 8129070351

transfer.py

@@ -0,0 +1,14 @@
+from omegaconf import OmegaConf
+from scripts.rendertext_tool import Render_Text, load_model_from_config
+import torch
+cfg = OmegaConf.load("config_cuda.yaml")
+model = load_model_from_config(cfg, "model_states.pt", verbose=True)
+
+from pytorch_lightning.callbacks import ModelCheckpoint
+with model.ema_scope("store ema weights"):
+    file_content = {
+        'state_dict': model.state_dict()
+    }
+    torch.save(file_content, "model.ckpt")
+    print("has stored the transfered ckpt.")
+print("trial ends!")