unilm commited on
Commit
77bc930
1 Parent(s): 7ca686a

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +56 -0
README.md ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Promptist: reinforcement learning for automatic prompt optimization
2
+
3
+ ## News
4
+ - [Demo Release] Dec, 2022: [Demo at HuggingFace Space](https://aka.ms/promptist-demo)
5
+ - [Model Release] Dec, 2022: [link](#load-pretrained-model-for-stable-diffusion-v14)
6
+ - [Paper Release] Dec, 2022: [Optimizing Prompts for Text-to-Image Generation](https://aka.ms/promptist-paper)
7
+
8
+ > - Language models serve as a prompt interface that optimizes user input into model-preferred prompts.
9
+
10
+ > - Learn a language model for automatic prompt optimization via reinforcement learning.
11
+
12
+ ![image](https://user-images.githubusercontent.com/1070872/207856962-02f08d92-f2bf-441a-b1c3-efff1a4b6187.png)
13
+
14
+
15
+ ## Load Pretrained Model for [Stable Diffusion v1.4](https://huggingface.co/CompVis/stable-diffusion-v1-4)
16
+
17
+ You can try the online demo at [https://huggingface.co/spaces/microsoft/Promptist](https://huggingface.co/spaces/microsoft/Promptist).
18
+
19
+ `[Note]` the online demo at HuggingFace Space is using CPU, so slow generation speed would be expected. Please load the model locally with GPUs for faster generation.
20
+
21
+ ```python
22
+ import gradio as grad
23
+ import torch
24
+ from transformers import AutoModelForCausalLM, AutoTokenizer
25
+
26
+ def load_prompter():
27
+ prompter_model = AutoModelForCausalLM.from_pretrained("microsoft/Promptist")
28
+ tokenizer = AutoTokenizer.from_pretrained("gpt2")
29
+ tokenizer.pad_token = tokenizer.eos_token
30
+ tokenizer.padding_side = "left"
31
+ return prompter_model, tokenizer
32
+
33
+ prompter_model, prompter_tokenizer = load_prompter()
34
+
35
+ def generate(plain_text):
36
+ input_ids = prompter_tokenizer(plain_text.strip()+" Rephrase:", return_tensors="pt").input_ids
37
+ eos_id = prompter_tokenizer.eos_token_id
38
+ outputs = prompter_model.generate(input_ids, do_sample=False, max_new_tokens=75, num_beams=8, num_return_sequences=8, eos_token_id=eos_id, pad_token_id=eos_id, length_penalty=-1.0)
39
+ output_texts = prompter_tokenizer.batch_decode(outputs, skip_special_tokens=True)
40
+ res = output_texts[0].replace(plain_text+" Rephrase:", "").strip()
41
+ return res
42
+
43
+ txt = grad.Textbox(lines=1, label="Initial Text", placeholder="Input Prompt")
44
+ out = grad.Textbox(lines=1, label="Optimized Prompt")
45
+ examples = ["A rabbit is wearing a space suit", "Several railroad tracks with one train passing by", "The roof is wet from the rain", "Cats dancing in a space club"]
46
+
47
+ grad.Interface(fn=generate,
48
+ inputs=txt,
49
+ outputs=out,
50
+ title="Promptist Demo",
51
+ description="Promptist is a prompt interface for Stable Diffusion v1-4 (https://huggingface.co/CompVis/stable-diffusion-v1-4) that optimizes user input into model-preferred prompts.",
52
+ examples=examples,
53
+ allow_flagging='never',
54
+ cache_examples=False,
55
+ theme="default").launch(enable_queue=True, debug=True)
56
+ ```