Spaces:
Sleeping
Sleeping
File size: 3,723 Bytes
4f48282 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 |
import modal
import random
from datetime import datetime
import os
from config.config import models, prompts
# Define the Modal image (same as in modal_app.py)
image = modal.Image.debian_slim(python_version="3.11").pip_install(
"diffusers",
"transformers",
"torch>=2.0.1",
"accelerate",
"gradio",
"safetensors",
"pillow",
)
# Create a Modal app
app = modal.App("ctb-ai-img-gen-modal", image=image)
# Define a volume for caching models
volume = modal.Volume.from_name("flux-model-vol")
@app.cls(
gpu="H100", # Use H100 GPU for maximum performance
container_idle_timeout=20 * 60, # 20 minutes
timeout=60 * 60, # 1 hour
volumes={"/cache": volume},
)
class Model:
def __init__(self):
self.device = "cuda"
self.torch_dtype = torch.bfloat16
self.model_dir = "/cache/models"
@modal.enter()
def setup(self):
import torch
from diffusers import StableDiffusionPipeline
# Load the model
self.pipe = StableDiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-2-1",
torch_dtype=self.torch_dtype,
safety_checker=None,
feature_extractor=None,
).to(self.device)
# Optimize the model
self.pipe = self.optimize(self.pipe)
def optimize(self, pipe):
import torch
# Fuse QKV projections
pipe.unet.fuse_qkv_projections()
pipe.vae.fuse_qkv_projections()
# Switch memory layout
pipe.unet.to(memory_format=torch.channels_last)
pipe.vae.to(memory_format=torch.channels_last)
# Compile the model
pipe.unet = torch.compile(pipe.unet, mode="max-autotune", fullgraph=True)
pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=True)
return pipe
@modal.method()
def generate(self, prompt_alias, team_color, model_alias, custom_prompt):
import torch
from diffusers import StableDiffusionPipeline
# Find the selected prompt and model
try:
prompt = next(p for p in prompts if p["alias"] == prompt_alias)["text"]
model_name = next(m for m in models if m["alias"] == model_alias)["name"]
except StopIteration:
return None, "ERROR: Invalid prompt or model selected."
# Format the prompt
enemy_color = "blue" if team_color.lower() == "red" else "red"
prompt = prompt.format(team_color=team_color.lower(), enemy_color=enemy_color)
if custom_prompt.strip():
prompt += " " + custom_prompt.strip()
# Set seed
seed = random.randint(0, 2**32 - 1)
torch.manual_seed(seed)
# Generate the image
try:
image = self.pipe(
prompt,
guidance_scale=2.0,
num_inference_steps=20,
width=640,
height=360,
generator=torch.Generator(self.device).manual_seed(seed)
).images[0]
# Save the image
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_filename = f"{timestamp}_{model_alias.replace(' ', '_').lower()}_{prompt_alias.replace(' ', '_').lower()}_{team_color.lower()}.png"
image.save(output_filename)
return output_filename, "Image generated successfully!"
except Exception as e:
return None, f"ERROR: Failed to generate image. Details: {e}"
# Function to be called from the Gradio interface
def generate(prompt_alias, team_color, model_alias, custom_prompt):
model = Model()
return model.generate.remote(prompt_alias, team_color, model_alias, custom_prompt) |