Update src/pipeline.py

src/pipeline.py

@@ -9,7 +9,7 @@ from torch import Generator
 from diffusers import FluxPipeline
 from torchao.quantization import quantize_, int8_weight_only, fpx_weight_only
 import torch.nn as nn 
-from model import Model, Decoder
+from model import Model, Decoder, Encoder
 import torchvision
 
 Pipeline = None
@@ -28,15 +28,26 @@ def load_pipeline() -> Pipeline:
     # vae.to(dtype=DTYPE)
 
     vae = AutoencoderTiny.from_pretrained("madebyollin/taef1")
+    vae.encoder = Encoder(16)
     vae.decoder = Decoder(16)
+
+    encoder_path = "encoder.pth"
     decoder_path = "decoder.pth"
-    decoder_state_dict = torch.load(decoder_path,  weights_only=True) #map_location="cpu",
-    filtered_state_dict = {k: v for k, v in decoder_state_dict.items() if k in vae.decoder.state_dict() and v.size() == vae.decoder.state_dict()[k].size()}
-    print(decoder_state_dict.keys())
-    print(filtered_state_dict.keys())
-    print(f"num of keys in pth: {len(decoder_state_dict)} and decoder: {len(vae.decoder.state_dict())}")
-    vae.decoder.load_state_dict(filtered_state_dict, strict=False)
+
+    if encoder_path is not None:
+        encoder_state_dict = torch.load(encoder_path, map_location="cpu", weights_only=True)
+        filtered_state_dict = {k.strip('encoder.'): v for k, v in encoder_state_dict.items() if k.strip('encoder.') in vae.encoder.state_dict() and v.size() == vae.encoder.state_dict()[k.strip('encoder.')].size()}
+        print(f" num of keys in filtered: {len(filtered_state_dict)} and in decoder: {len(vae.encoder.state_dict())}")
+        vae.encoder.load_state_dict(filtered_state_dict, strict=False)
+            
+    if decoder_path is not None:
+        decoder_state_dict = torch.load(decoder_path, map_location="cpu", weights_only=True)
+        filtered_state_dict = {k.strip('decoder.'): v for k, v in decoder_state_dict.items() if k.strip('decoder.') in vae.decoder.state_dict() and v.size() == vae.decoder.state_dict()[k.strip('decoder.')].size()}
+        print(f" num of keys in filtered: {len(filtered_state_dict)} and in decoder: {len(vae.decoder.state_dict())}")
+        vae.decoder.load_state_dict(filtered_state_dict, strict=False)
+
     vae.decoder.requires_grad_(False)
+    vae.encoder.requires_grad_(False)
     vae.to(dtype=DTYPE)
     
     pipeline = FluxPipeline.from_pretrained(MODEL_ID,vae=vae,
@@ -68,11 +79,13 @@ def infer(request: TextToImageRequest, pipeline: Pipeline) -> Image:
     generator = Generator("cuda").manual_seed(request.seed)
     image = None
     # with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
-    image=pipeline(request.prompt,generator=generator, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256, height=request.height, width=request.width, output_type="pt").images[0]
+    # image=pipeline(request.prompt,generator=generator, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256, height=request.height, width=request.width, output_type="pt").images[0]
+    image=pipeline(request.prompt,generator=generator, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256, height=request.height, width=request.width, output_type="pil").images[0]
     # image = image / 255.
     # image = image.mul_(2).sub_(1)
     # image = ((image + 1) / 2) * 255
     # image = image.clamp(0, 255)
     # image = image.to(torch.float32)
     # return torchvision.transforms.functional.to_pil_image(image)
-    return torchvision.transforms.functional.to_pil_image(image.to(torch.float32).mul_(2).sub_(1))
+    # return torchvision.transforms.functional.to_pil_image(image.to(torch.float32).mul_(2).sub_(1))
+    return image