Update app.py

app.py

@@ -14,28 +14,22 @@ def convert_to_wav(audio_file):
     return wav_file
 
 import torch
-from transformers import AutoProcessor, pipeline
-
-
-# Initialize processor and pipeline
 
 # torch_dtype = torch.float32
+#MODEL_NAME = "NbAiLabBeta/nb-whisper-large"
 
-
-MODEL_NAME = "NbAiLabBeta/nb-whisper-large"
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-pipe = pipeline(
-    task="automatic-speech-recognition",
-    model=MODEL_NAME,
-    chunk_length_s=30,
-    device=device,
+#pipe = pipeline(
+    #task="automatic-speech-recognition",
+    #model=MODEL_NAME,
+    #chunk_length_s=30,
+    #device=device,
 )
 
+#pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="no", task="transcribe")
 
-pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="no", task="transcribe")
-
+from transformers import pipeline
 
+pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large")
 
 def transcribe_audio(audio_file):
     if audio_file.endswith(".m4a"):
@@ -56,7 +50,6 @@ def transcribe_audio(audio_file):
     return text, result
 
 
-# [VERSION 3: full-on w/ 3 styles for summarization]
 import nltk
 from nltk.tokenize import word_tokenize, sent_tokenize
 from nltk.corpus import stopwords
@@ -103,7 +96,8 @@ def preprocess_text(text):
 
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
-summarization_model = AutoModelForSeq2SeqLM.from_pretrained("t5-base", return_dict=True, torch_dtype=torch.float16)
+device = "cuda" if torch.cuda.is_available() else "cpu"
+summarization_model = AutoModelForSeq2SeqLM.from_pretrained("t5-base", torch_dtype=torch.float16)
 summarization_tokenizer = AutoTokenizer.from_pretrained("t5-base")
 summarization_model.to(device)