Update README.md
Browse files
README.md
CHANGED
@@ -9,6 +9,7 @@ metrics:
|
|
9 |
- rouge
|
10 |
library_name: transformers
|
11 |
pipeline_tag: text-generation
|
|
|
12 |
tags:
|
13 |
- clickbait
|
14 |
- noticia
|
@@ -68,12 +69,28 @@ If you are looking for a smaller model, check out [ClickbaitFighter-2B](https://
|
|
68 |
- 🔌 Online Demo: [https://iker-clickbaitfighter.hf.space/](https://iker-clickbaitfighter.hf.space/)
|
69 |
|
70 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
# Usage example:
|
|
|
|
|
72 |
```python
|
73 |
import torch # pip install torch
|
74 |
-
from
|
75 |
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig # pip install transformers
|
76 |
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
|
78 |
def prompt(
|
79 |
headline: str,
|
@@ -107,10 +124,81 @@ def prompt(
|
|
107 |
f"{body}\n"
|
108 |
)
|
109 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
dataset = load_dataset("Iker/NoticIA")
|
111 |
example = dataset["test"][0]
|
|
|
|
|
112 |
|
113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
114 |
|
115 |
tokenizer = AutoTokenizer.from_pretrained("Iker/ClickbaitFighter-10B")
|
116 |
model = AutoModelForCausalLM.from_pretrained(
|
@@ -140,13 +228,6 @@ summary = tokenizer.batch_decode(model_output,skip_special_tokens=True)[0]
|
|
140 |
print(summary.strip().split("\n")[-1]) # Get only the summary, without the prompt.
|
141 |
```
|
142 |
|
143 |
-
# Evaluation Results
|
144 |
-
<table>
|
145 |
-
<tr>
|
146 |
-
<td style="width:100%"><img src="https://github.com/ikergarcia1996/NoticIA/raw/main/results/Results.png" align="right" width="100%"> </td>
|
147 |
-
</tr>
|
148 |
-
</table>
|
149 |
-
|
150 |
|
151 |
# Citation
|
152 |
|
|
|
9 |
- rouge
|
10 |
library_name: transformers
|
11 |
pipeline_tag: text-generation
|
12 |
+
base_model: NousResearch/Nous-Hermes-2-SOLAR-10.7B
|
13 |
tags:
|
14 |
- clickbait
|
15 |
- noticia
|
|
|
69 |
- 🔌 Online Demo: [https://iker-clickbaitfighter.hf.space/](https://iker-clickbaitfighter.hf.space/)
|
70 |
|
71 |
|
72 |
+
# Evaluation Results
|
73 |
+
<table>
|
74 |
+
<tr>
|
75 |
+
<td style="width:100%"><img src="https://github.com/ikergarcia1996/NoticIA/raw/main/results/Results.png" align="right" width="100%"> </td>
|
76 |
+
</tr>
|
77 |
+
</table>
|
78 |
+
|
79 |
+
|
80 |
# Usage example:
|
81 |
+
|
82 |
+
## Summarize a web article
|
83 |
```python
|
84 |
import torch # pip install torch
|
85 |
+
from newspaper import Article #pip3 install newspaper3k
|
86 |
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig # pip install transformers
|
87 |
|
88 |
+
article_url ="https://www.huffingtonpost.es/virales/le-compra-abrigo-abuela-97nos-reaccion-fantasia.html"
|
89 |
+
article = Article(article_url)
|
90 |
+
article.download()
|
91 |
+
article.parse()
|
92 |
+
headline=article.title
|
93 |
+
body = article.text
|
94 |
|
95 |
def prompt(
|
96 |
headline: str,
|
|
|
124 |
f"{body}\n"
|
125 |
)
|
126 |
|
127 |
+
prompt = prompt(headline=headline, body=body)
|
128 |
+
|
129 |
+
tokenizer = AutoTokenizer.from_pretrained("Iker/ClickbaitFighter-10B")
|
130 |
+
model = AutoModelForCausalLM.from_pretrained(
|
131 |
+
"Iker/ClickbaitFighter-2B", torch_dtype=torch.bfloat16, device_map="auto"
|
132 |
+
)
|
133 |
+
|
134 |
+
formatted_prompt = tokenizer.apply_chat_template(
|
135 |
+
[{"role": "user", "content": prompt}],
|
136 |
+
tokenize=False,
|
137 |
+
add_generation_prompt=True,
|
138 |
+
)
|
139 |
+
|
140 |
+
model_inputs = tokenizer(
|
141 |
+
[formatted_prompt], return_tensors="pt", add_special_tokens=False
|
142 |
+
)
|
143 |
+
|
144 |
+
model_output = model.generate(**model_inputs.to(model.device), generation_config=GenerationConfig(
|
145 |
+
max_new_tokens=32,
|
146 |
+
min_new_tokens=1,
|
147 |
+
do_sample=False,
|
148 |
+
num_beams=1,
|
149 |
+
use_cache=True
|
150 |
+
))
|
151 |
+
|
152 |
+
summary = tokenizer.batch_decode(model_output,skip_special_tokens=True)[0]
|
153 |
+
|
154 |
+
print(summary.strip().split("\n")[-1]) # Get only the summary, without the prompt.
|
155 |
+
```
|
156 |
+
|
157 |
+
## Run inference in the NoticIA dataset
|
158 |
+
```python
|
159 |
+
import torch # pip install torch
|
160 |
+
from newspaper import Article #pip3 install newspaper3k
|
161 |
+
from datasets import load_dataset # pip install datasets
|
162 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig # pip install transformers
|
163 |
+
|
164 |
dataset = load_dataset("Iker/NoticIA")
|
165 |
example = dataset["test"][0]
|
166 |
+
headline = example["web_headline"]
|
167 |
+
body = example["web_text"]
|
168 |
|
169 |
+
def prompt(
|
170 |
+
headline: str,
|
171 |
+
body: str,
|
172 |
+
) -> str:
|
173 |
+
"""
|
174 |
+
Generate the prompt for the model.
|
175 |
+
|
176 |
+
Args:
|
177 |
+
headline (`str`):
|
178 |
+
The headline of the article.
|
179 |
+
body (`str`):
|
180 |
+
The body of the article.
|
181 |
+
Returns:
|
182 |
+
`str`: The formatted prompt.
|
183 |
+
"""
|
184 |
+
|
185 |
+
return (
|
186 |
+
f"Ahora eres una Inteligencia Artificial experta en desmontar titulares sensacionalistas o clickbait. "
|
187 |
+
f"Tu tarea consiste en analizar noticias con titulares sensacionalistas y "
|
188 |
+
f"generar un resumen de una sola frase que revele la verdad detrás del titular.\n"
|
189 |
+
f"Este es el titular de la noticia: {headline}\n"
|
190 |
+
f"El titular plantea una pregunta o proporciona información incompleta. "
|
191 |
+
f"Debes buscar en el cuerpo de la noticia una frase que responda lo que se sugiere en el título. "
|
192 |
+
f"Siempre que puedas cita el texto original, especialmente si se trata de una frase que alguien ha dicho. "
|
193 |
+
f"Si citas una frase que alguien ha dicho, usa comillas para indicar que es una cita. "
|
194 |
+
f"Usa siempre las mínimas palabras posibles. No es necesario que la respuesta sea una oración completa. "
|
195 |
+
f"Puede ser sólo el foco de la pregunta. "
|
196 |
+
f"Recuerda responder siempre en Español.\n"
|
197 |
+
f"Este es el cuerpo de la noticia:\n"
|
198 |
+
f"{body}\n"
|
199 |
+
)
|
200 |
+
|
201 |
+
prompt = prompt(headline=headline, body=body)
|
202 |
|
203 |
tokenizer = AutoTokenizer.from_pretrained("Iker/ClickbaitFighter-10B")
|
204 |
model = AutoModelForCausalLM.from_pretrained(
|
|
|
228 |
print(summary.strip().split("\n")[-1]) # Get only the summary, without the prompt.
|
229 |
```
|
230 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
231 |
|
232 |
# Citation
|
233 |
|