Attempt to fix the clip guidance

app.py

@@ -19,7 +19,6 @@ if not (path_exists(f"CLIP")):
     Repo.clone_from("https://github.com/openai/CLIP", "CLIP")
 sys.path.append('v-diffusion-pytorch')
 
-
 from huggingface_hub import hf_hub_download
 
 from CLIP import clip
@@ -62,18 +61,23 @@ model = model.half().cuda().eval().requires_grad_(False)
 #model_small.load_state_dict(torch.load(cc12m_model, map_location='cpu'))
 #model_small = model_small.half().cuda().eval().requires_grad_(False)
 
-print(model.clip_model)
 clip_model = clip.load(model.clip_model, jit=False, device='cuda')[0]
 clip_model.eval().requires_grad_(False)
 normalize = transforms.Normalize(mean=[0.48145466, 0.4578275, 0.40821073],
                                      std=[0.26862954, 0.26130258, 0.27577711])
 make_cutouts = MakeCutouts(clip_model.visual.input_resolution, 16, 1.)
+gc.collect()
+torch.cuda.empty_cache()
+
 def run_all(prompt, steps, n_images, weight, clip_guided):
+    gc.collect()
+    torch.cuda.empty_cache()
     import random
     seed = int(random.randint(0, 2147483647))
     target_embed = clip_model.encode_text(clip.tokenize(prompt).to('cuda')).float()#.cuda()
     
     if(clip_guided):
+        n_images = 1
         steps = steps*5
         clip_guidance_scale = weight*100
         prompts = [prompt]
@@ -109,8 +113,7 @@ def run_all(prompt, steps, n_images, weight, clip_guided):
         clip_embed_in = torch.cat([torch.zeros_like(clip_embed_repeat), clip_embed_repeat])
         v_uncond, v_cond = model(x_in, t_in, clip_embed_in).chunk(2, dim=0)
         v = v_uncond + (v_cond - v_uncond) * weight
-        return v
-    
+        return v   
     def make_cond_model_fn(model, cond_fn):
         def cond_model_fn(x, t, **extra_args):
             with torch.enable_grad():
@@ -132,12 +135,9 @@ def run_all(prompt, steps, n_images, weight, clip_guided):
         grad = -torch.autograd.grad(loss, x)[0]
         return grad
     
-    gc.collect()
-    torch.cuda.empty_cache()
     torch.manual_seed(seed)
     x = torch.randn([n_images, 3, side_y, side_x], device='cuda')
     t = torch.linspace(1, 0, steps + 1, device='cuda')[:-1]
-    #step_list = utils.get_spliced_ddpm_cosine_schedule(t)
     if model.min_t == 0:
         step_list = utils.get_spliced_ddpm_cosine_schedule(t)
     else:
@@ -164,11 +164,10 @@ iface = gr.Interface(
     gr.inputs.Slider(label="Steps - more steps can increase quality but will take longer to generate",default=40,maximum=80,minimum=1,step=1),
     gr.inputs.Slider(label="Number of images in parallel", default=2, maximum=4, minimum=1, step=1),
     gr.inputs.Slider(label="Weight - how closely the image should resemble the prompt", default=5, maximum=15, minimum=0, step=1),
-    gr.inputs.Checkbox(label="CLIP Guided - improves coherence with complex prompts, makes it slower"),
+    gr.inputs.Checkbox(label="CLIP Guided - improves coherence with complex prompts, makes it slower (with CLIP Guidance only one image is generated)"),
     ], 
     outputs=gallery,
     title="Generate images from text with V-Diffusion",
     description="<div>By typing a prompt and pressing submit you can generate images based on this prompt. <a href='https://github.com/crowsonkb/v-diffusion-pytorch' target='_blank'>V-Diffusion</a> is diffusion text-to-image model created by <a href='https://twitter.com/RiversHaveWings' target='_blank'>Katherine Crowson</a> and <a href='https://twitter.com/jd_pressman'>JDP</a>, trained on the <a href='https://github.com/google-research-datasets/conceptual-12m'>CC12M dataset</a>. The UI to the model was assembled by <a style='color: rgb(99, 102, 241);font-weight:bold' href='https://twitter.com/multimodalart' target='_blank'>@multimodalart</a>, keep up with the <a style='color: rgb(99, 102, 241);' href='https://multimodal.art/news' target='_blank'>latest multimodal ai art news here</a> and consider <a style='color: rgb(99, 102, 241);' href='https://www.patreon.com/multimodalart' target='_blank'>supporting us on Patreon</a></div>",
-    #article="<h4 style='font-size: 110%;margin-top:.5em'>Biases acknowledgment</h4><div>Despite how impressive being able to turn text into image is, beware to the fact that this model may output content that reinforces or exarcbates societal biases. According to the <a href='https://arxiv.org/abs/2112.10752' target='_blank'>Latent Diffusion paper</a>:<i> \"Deep learning modules tend to reproduce or exacerbate biases that are already present in the data\"</i>. The model was trained on an unfiltered version the LAION-400M dataset, which scrapped non-curated image-text-pairs from the internet (the exception being the the removal of illegal content) and is meant to be used for research purposes, such as this one. <a href='https://laion.ai/laion-400-open-dataset/' target='_blank'>You can read more on LAION's website</a></div><h4 style='font-size: 110%;margin-top:1em'>Who owns the images produced by this demo?</h4><div>Definetly not me! Probably you do. I say probably because the Copyright discussion about AI generated art is ongoing. So <a href='https://www.theverge.com/2022/2/21/22944335/us-copyright-office-reject-ai-generated-art-recent-entrance-to-paradise' target='_blank'>it may be the case that everything produced here falls automatically into the public domain</a>. But in any case it is either yours or is in the public domain.</div>"
     )
 iface.launch(enable_queue=True)