import streamlit as st from PIL import Image from diffusers import StableDiffusionPipeline, ControlNetModel, DDIMScheduler, LMSDiscreteScheduler, UNet2DConditionModel, DiffusionPipeline from diffusers import DDPMScheduler, DDPMSchedulerV2, PNDMScheduler from transformers import AutoModelForCausalLM, AutoTokenizer, AutoModelForSeq2SeqLM, LlamaTokenizerFast, LlamaForCausalLM from accelerate import Accelerator import torch from peft import PeftModel, LoraConfig, get_peft_model, prepare_model_for_int8_training, prepare_model_for_int8_bf16_training # Define a dictionary with all available models, schedulers, features, weights, and adapters model_dict = { "Stable Diffusion": { "Models": [ "stabilityai/stable-diffusion-3-medium" "CompVis/stable-diffusion-v1-4", "stabilityai/stable-diffusion-2-1", "runwayml/stable-diffusion-v1-5", "runwayml/stable-diffusion-inpainting", "runwayml/stable-diffusion-video-v1-5", "stabilityai/stable-diffusion-2-base" ], "Schedulers": [ "DDIMScheduler", "LMSDiscreteScheduler" ], "Features": [ "Unconditional image generation", "Text-to-image", "Image-to-image", "Inpainting", "Text or image-to-video", "Depth-to-image" ], "Adapters": [ "ControlNet", "T2I-Adapter" ], "Weights": [ "Stable Diffusion XL", "SDXL Turbo", "Kandinsky", "IP-Adapter", "ControlNet", "Latent Consistency Model", "Textual inversion", "Shap-E", "DiffEdit", "Trajectory Consistency Distillation-LoRA", "Stable Video Diffusion", "Marigold Computer Vision" ] }, "Llama": { "Models": [ "decapoda-research/llama-7b-hf", "decapoda-research/llama-13b-hf", "decapoda-research/llama-30b-hf", "decapoda-research/llama-65b-hf" ], "Tokenizers": [ "LlamaTokenizerFast" ], "Features": [ "AutoPipeline", "Train a diffusion model", "Load LoRAs for inference", "Accelerate inference of text-to-image diffusion models", "LOAD PIPELINES AND ADAPTERS", "Load community pipelines and components", "Load schedulers and models", "Model files and layouts", "Load adapters", "Push files to the Hub", "GENERATIVE TASKS", "Unconditional image generation", "Text-to-image", "Image-to-image", "Inpainting", "Text or image-to-video", "Depth-to-image", "INFERENCE TECHNIQUES", "Overview", "Distributed inference with multiple GPUs", "Merge LoRAs", "Scheduler features", "Pipeline callbacks", "Reproducible pipelines", "Controlling image quality", "Prompt techniques", "ADVANCED INFERENCE", "Outpainting", "SPECIFIC PIPELINE EXAMPLES", "Stable Diffusion XL", "SDXL Turbo", "Kandinsky", "IP-Adapter", "ControlNet", "T2I-Adapter", "Latent Consistency Model", "Textual inversion", "Shap-E", "DiffEdit", "Trajectory Consistency Distillation-LoRA", "Stable Video Diffusion", "Marigold Computer Vision" ], "Weights": [ "LoRA weights" ] } } model_type = st.selectbox("Select a model type:", list(model_dict.keys())) if model_type == "Stable Diffusion": model = st.selectbox("Select a Stable Diffusion model:", model_dict[model_type]["Models"]) scheduler = st.selectbox("Select a scheduler:", model_dict[model_type]["Schedulers"]) feature = st.selectbox("Select a feature:", model_dict[model_type]["Features"]) adapter = st.selectbox("Select an adapter:", model_dict[model_type]["Adapters"]) weight = st.selectbox("Select a weight:", model_dict[model_type]["Weights"]) if st.button("Generate Images"): st.write("Generating images...") pipe = StableDiffusionPipeline.from_pretrained(model) pipe.scheduler = eval(scheduler)() if adapter == "ControlNet": controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11e_sd15_openpose") pipe = pipe.to_controlnet(controlnet) # Define the prompt and number of images to generate prompt = st.text_input("Enter a prompt:") num_images = st.slider("Number of images to generate", min_value=1, max_value=10, value=1) # Generate the images images = pipe(prompt, num_images=num_images, guidance_scale=7.5).images # Display the generated images cols = st.columns(num_images) for i, image in enumerate(images): cols[i].image(image, caption=f"Image {i+1}", use_column_width=True) if model_type == "Llama": model = st.selectbox("Select a Llama model:", model_dict[model_type]["Models"]) tokenizer = st.selectbox("Select a tokenizer:", model_dict[model_type]["Tokenizers"]) feature = st.selectbox("Select a feature:", model_dict[model_type]["Features"]) weight = st.selectbox("Select a weight:", model_dict[model_type]["Weights"]) if st.button("Generate Text"): st.write("Generating text...") tokenizer = AutoTokenizer.from_pretrained(tokenizer) model = AutoModelForCausalLM.from_pretrained(model) input_text = st.text_area("Enter a prompt:") # Tokenize the input text inputs = tokenizer(input_text, return_tensors="pt") # Generate the text output = model.generate(**inputs) # Decode the generated text generated_text = tokenizer.decode(output[0], skip_special_tokens=True) st.write("Generated Text:") st.write(generated_text)