| import gradio as gr | |
| import torch | |
| import random | |
| import transformers | |
| from transformers import T5Tokenizer, T5ForConditionalGeneration | |
| if torch.cuda.is_available(): | |
| device = "cuda" | |
| print("Using GPU") | |
| else: | |
| device = "cpu" | |
| print("Using CPU") | |
| tokenizer = T5Tokenizer.from_pretrained("imranali291/flux-prompt-enhancer") | |
| model = T5ForConditionalGeneration.from_pretrained("imranali291/flux-prompt-enhancer", device_map="auto", torch_dtype="auto") | |
| model.to(device) | |
| def generate(your_prompt, max_new_tokens, repetition_penalty, temperature, model_precision_type, top_p, top_k, seed): | |
| if seed == 0: | |
| seed = random.randint(1, 2**32-1) | |
| transformers.set_seed(seed) | |
| if model_precision_type == "fp16": | |
| dtype = torch.float16 | |
| elif model_precision_type == "fp32": | |
| dtype = torch.float32 | |
| model.to(dtype) | |
| repetition_penalty = float(repetition_penalty) | |
| input_text = f"{your_prompt}" | |
| input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to(device) | |
| outputs = model.generate( | |
| input_ids, | |
| max_new_tokens=max_new_tokens, | |
| repetition_penalty=repetition_penalty, | |
| do_sample=True, | |
| temperature=temperature, | |
| top_p=top_p, | |
| top_k=top_k, | |
| ) | |
| better_prompt = tokenizer.decode(outputs[0], skip_special_tokens=True) | |
| return better_prompt | |
| your_prompt = gr.Textbox(label="Your Prompt", info="Your Prompt that you want to enhanced") | |
| max_new_tokens = gr.Slider(value=128, minimum=25, maximum=512, step=1, label="Max New Tokens", info="The maximum numbers of new tokens, controls how long is the output") | |
| repetition_penalty = gr.Slider(value=2.5, minimum=0, maximum=3.0, step=0.05, label="Repetition Penalty", info="Penalize repeated tokens, making the AI repeat less itself") | |
| temperature = gr.Slider(value=0.7, minimum=0, maximum=1, step=0.05, label="Temperature", info="Higher values produce more diverse outputs") | |
| model_precision_type = gr.Dropdown(["fp16", "fp32"], value="fp16", label="Model Precision Type", info="The precision type to load the model, like fp16 which is faster, or fp32 which is more precise but more resource consuming") | |
| top_p = gr.Slider(value=0.9, minimum=0, maximum=2, step=0.05, label="Top P", info="Higher values sample more low-probability tokens") | |
| top_k = gr.Slider(value=50, minimum=1, maximum=100, step=1, label="Top K", info="Higher k means more diverse outputs by considering a range of tokens") | |
| seed = gr.Slider(value=42, minimum=0, maximum=2**32-1, step=1, label="Seed", info="A starting point to initiate the generation process, put 0 for a random one") | |
| examples = [ | |
| ["Futuristic cityscape at twilight descent.", 128, 2.5, 0.5, "fp16", 0.9, 50, 42] | |
| ] | |
| gr.Interface( | |
| fn=generate, | |
| inputs=[your_prompt, max_new_tokens, repetition_penalty, temperature, model_precision_type, top_p, top_k, seed], | |
| outputs=gr.Textbox(label="Prompt Enhancer"), | |
| title="Prompt Enhancer", | |
| description='Make your prompts more detailed!', | |
| examples=examples, | |
| ).launch(share=True) |