fix transformers
Browse files
app.py
CHANGED
@@ -32,6 +32,8 @@ from operator import itemgetter
|
|
32 |
|
33 |
import gradio as gr
|
34 |
|
|
|
|
|
35 |
global df
|
36 |
bearer_token = 'AAAAAAAAAAAAAAAAAAAAACEigwEAAAAACoP8KHJYLOKCL4OyB9LEPV00VB0%3DmyeDROUvw4uipHwvbPPfnTuY0M9ORrLuXrMvcByqZhwo3SUc4F'
|
37 |
client = tweepy.Client(bearer_token=bearer_token)
|
@@ -392,11 +394,10 @@ def full_lda(df):
|
|
392 |
return top_tweets
|
393 |
|
394 |
def topic_summarization(topic_groups):
|
395 |
-
|
396 |
|
397 |
-
|
398 |
-
|
399 |
-
model = model.to(device)
|
400 |
translator = Translator()
|
401 |
|
402 |
headlines = []
|
@@ -410,8 +411,8 @@ def topic_summarization(topic_groups):
|
|
410 |
max_len = 256
|
411 |
|
412 |
encoding = tokenizer.encode_plus(text, return_tensors = "pt")
|
413 |
-
input_ids = encoding["input_ids"]
|
414 |
-
attention_masks = encoding["attention_mask"]
|
415 |
|
416 |
beam_outputs = model.generate(
|
417 |
input_ids = input_ids,
|
|
|
32 |
|
33 |
import gradio as gr
|
34 |
|
35 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
36 |
+
|
37 |
global df
|
38 |
bearer_token = 'AAAAAAAAAAAAAAAAAAAAACEigwEAAAAACoP8KHJYLOKCL4OyB9LEPV00VB0%3DmyeDROUvw4uipHwvbPPfnTuY0M9ORrLuXrMvcByqZhwo3SUc4F'
|
39 |
client = tweepy.Client(bearer_token=bearer_token)
|
|
|
394 |
return top_tweets
|
395 |
|
396 |
def topic_summarization(topic_groups):
|
397 |
+
|
398 |
|
399 |
+
tokenizer = AutoTokenizer.from_pretrained("Michau/t5-base-en-generate-headline")
|
400 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("Michau/t5-base-en-generate-headline")
|
|
|
401 |
translator = Translator()
|
402 |
|
403 |
headlines = []
|
|
|
411 |
max_len = 256
|
412 |
|
413 |
encoding = tokenizer.encode_plus(text, return_tensors = "pt")
|
414 |
+
input_ids = encoding["input_ids"]
|
415 |
+
attention_masks = encoding["attention_mask"]
|
416 |
|
417 |
beam_outputs = model.generate(
|
418 |
input_ids = input_ids,
|