update to schell fp8

app.py

@@ -1,46 +1,46 @@
 import gradio as gr
 import numpy as np
 import random
-from diffusers import DiffusionPipeline
+import spaces
 import torch
+from diffusers import DiffusionPipeline
+
+# Enable TF32 for A100 (this is a form of FP8 computation)
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.float16  # Use float16 for loading
 
-if torch.cuda.is_available():
-    torch.cuda.max_memory_allocated(device=device)
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
-    pipe.enable_xformers_memory_efficient_attention()
-    pipe = pipe.to(device)
-else: 
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", use_safetensors=True)
-    pipe = pipe.to(device)
+pipe = DiffusionPipeline.from_pretrained(
+    "Kijai/flux-fp8",
+    torch_dtype=dtype,
+    revision="main",
+    filename="flux1-schnell-fp8.safetensors"
+).to(device)
 
 MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
+MAX_IMAGE_SIZE = 2048
 
+@spaces.GPU()
+def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, progress=gr.Progress(track_tqdm=True)):
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
-        
     generator = torch.Generator().manual_seed(seed)
-    
     image = pipe(
-        prompt = prompt, 
-        negative_prompt = negative_prompt,
-        guidance_scale = guidance_scale, 
-        num_inference_steps = num_inference_steps, 
-        width = width, 
-        height = height,
-        generator = generator
-    ).images[0] 
-    
-    return image
+        prompt=prompt,
+        width=width,
+        height=height,
+        num_inference_steps=num_inference_steps,
+        generator=generator,
+        guidance_scale=0.0
+    ).images[0]
+    return image, seed
 
 examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
+    "a tiny astronaut hatching from an egg on the moon",
+    "a cat holding a sign that says hello world",
+    "an anime illustration of a wiener schnitzel",
 ]
 
 css="""
@@ -50,21 +50,13 @@ css="""
 }
 """
 
-if torch.cuda.is_available():
-    power_device = "GPU"
-else:
-    power_device = "CPU"
-
 with gr.Blocks(css=css) as demo:
-    
     with gr.Column(elem_id="col-container"):
-        gr.Markdown(f"""
-        # Text-to-Image Gradio Template
-        Currently running on {power_device}.
+        gr.Markdown(f"""# FLUX.1 [schnell] FP8
+        12B param rectified flow transformer distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/) for 4 step generation
+        [[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/Kijai/flux-fp8)]
         """)
-        
         with gr.Row():
-            
             prompt = gr.Text(
                 label="Prompt",
                 show_label=False,
@@ -72,20 +64,9 @@ with gr.Blocks(css=css) as demo:
                 placeholder="Enter your prompt",
                 container=False,
             )
-            
             run_button = gr.Button("Run", scale=0)
-        
         result = gr.Image(label="Result", show_label=False)
-
         with gr.Accordion("Advanced Settings", open=False):
-            
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-            
             seed = gr.Slider(
                 label="Seed",
                 minimum=0,
@@ -93,54 +74,42 @@ with gr.Blocks(css=css) as demo:
                 step=1,
                 value=0,
             )
-            
             randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-            
             with gr.Row():
-                
                 width = gr.Slider(
                     label="Width",
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=512,
+                    value=1024,
                 )
-                
                 height = gr.Slider(
                     label="Height",
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=512,
+                    value=1024,
                 )
-            
             with gr.Row():
-                
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,
-                )
-                
                 num_inference_steps = gr.Slider(
                     label="Number of inference steps",
                     minimum=1,
-                    maximum=12,
+                    maximum=50,
                     step=1,
-                    value=2,
+                    value=4,
                 )
-        
         gr.Examples(
-            examples = examples,
-            inputs = [prompt]
+            examples=examples,
+            fn=infer,
+            inputs=[prompt],
+            outputs=[result, seed],
+            cache_examples="lazy"
+        )
+        gr.on(
+            triggers=[run_button.click, prompt.submit],
+            fn=infer,
+            inputs=[prompt, seed, randomize_seed, width, height, num_inference_steps],
+            outputs=[result, seed]
         )
 
-    run_button.click(
-        fn = infer,
-        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result]
-    )
-
-demo.queue().launch()
+demo.launch()

requirements.txt

@@ -1,6 +1,7 @@
 accelerate
-diffusers
+git+https://github.com/huggingface/diffusers.git@flux-pipeline
 invisible_watermark
 torch
-transformers
-xformers
+transformers==4.42.4
+xformers
+sentencepiece