Fakypedia
Browse files- app.py +59 -0
- requirements.txt +7 -0
app.py
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
+
import spaces
|
5 |
+
|
6 |
+
title = "Fakypedia"
|
7 |
+
|
8 |
+
DESCRIPTION = """\
|
9 |
+
# Genarate a silly article
|
10 |
+
A bilingual (English and Hebrew) [nonsense generation model](https://huggingface.co/Norod78/SmolLM-135M-FakyPedia-EngHeb) which produces silly Wikipedia-like abstract text.
|
11 |
+
Tap on the \"Submit\" button to generate a silly and/or fake \"Wikipedia-Like\" article based on the input title
|
12 |
+
"""
|
13 |
+
|
14 |
+
article = "<p>This model extended the tokenizer of and is a fine-tuned of [SmolLM-135M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM-135M-Instruct)</p>"
|
15 |
+
|
16 |
+
CUDA_AVAILABLE = torch.cuda.is_available()
|
17 |
+
device = torch.device("cuda" if CUDA_AVAILABLE else "cpu")
|
18 |
+
|
19 |
+
model_id = "Norod78/SmolLM-135M-FakyPedia-EngHeb"
|
20 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
21 |
+
tokenizer.pad_token_id = tokenizer.eos_token_id
|
22 |
+
bos_token = tokenizer.bos_token
|
23 |
+
model = AutoModelForCausalLM.from_pretrained(model_id).to(device)
|
24 |
+
model.generation_config.pad_token_id = tokenizer.pad_token_id
|
25 |
+
|
26 |
+
torch.manual_seed(1234)
|
27 |
+
|
28 |
+
@spaces.GPU
|
29 |
+
def generate_fakypedia(article_title: str):
|
30 |
+
with torch.no_grad():
|
31 |
+
result = ""
|
32 |
+
string_to_tokenize= f"{bos_token}\\%{article_title}"
|
33 |
+
input_ids = tokenizer( string_to_tokenize, return_tensors="pt").input_ids.to(device)
|
34 |
+
sample_outputs = model.generate(input_ids, do_sample=True,repetition_penalty=1.2, temperature=0.5, max_length=96, num_return_sequences=3)
|
35 |
+
if article_title == None or len(article_title) == 0:
|
36 |
+
result += f"# Fakypedia results with random titles \n"
|
37 |
+
article_title = ""
|
38 |
+
else:
|
39 |
+
result += f"# Fakypedia results for \"{article_title}\" \n"
|
40 |
+
for i, sample_output in enumerate(sample_outputs):
|
41 |
+
decoded_output = tokenizer.decode(sample_output, skip_special_tokens=True)
|
42 |
+
decoded_output = decoded_output.replace(f"\%{article_title}", f"## {article_title}").replace("\%", " ").replace("\\n", " \n")
|
43 |
+
decoded_output = decoded_output.replace("## \n", "\n")
|
44 |
+
result += "{}\n".format(decoded_output)
|
45 |
+
return result
|
46 |
+
|
47 |
+
demo = gr.Interface(
|
48 |
+
generate_fakypedia,
|
49 |
+
inputs=gr.Textbox(lines=1, label="Enter a title for the article (or leave blank for a random one)"),
|
50 |
+
outputs=gr.Markdown(label="Generated fakypedia article"),
|
51 |
+
title=title,
|
52 |
+
description=DESCRIPTION,
|
53 |
+
article=article,
|
54 |
+
examples=["Hugging face", "A socially awkward potato", "ืืืจืื ืืืืจ", ""],
|
55 |
+
allow_flagging="never",
|
56 |
+
)
|
57 |
+
|
58 |
+
demo.queue()
|
59 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio
|
2 |
+
accelerate
|
3 |
+
torch
|
4 |
+
transformers
|
5 |
+
tokenizers
|
6 |
+
spaces
|
7 |
+
numpy
|