Update app.py

app.py

@@ -4,11 +4,63 @@ import os
 import string 
 import re
 import torch
-from transformers import pipeline
+from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer, AutoConfig
 from transformers import M2M100Tokenizer, M2M100ForConditionalGeneration
 import fasttext
 from huggingface_hub import hf_hub_download
 
+summarization_model_names = [
+    "google/bigbird-pegasus-large-arxiv",
+    "facebook/bart-large-cnn",
+    "google/t5-v1_1-large",
+    "sshleifer/distilbart-cnn-12-6",
+    "allenai/led-base-16384",
+    "google/pegasus-xsum",
+    "togethercomputer/LLaMA-2-7B-32K"
+]
+
+# Placeholder for the summarizer pipeline, tokenizer, and maximum tokens
+summarizer = None
+tokenizer = None
+max_tokens = None
+
+
+# Function to load the selected model
+def load_summarization_model(model_name):
+    global summarizer, tokenizer, max_tokens
+    try:
+        summarizer = pipeline("summarization", model=model_name, torch_dtype=torch.bfloat16)
+        tokenizer = AutoTokenizer.from_pretrained(model_name)
+        config = AutoConfig.from_pretrained(model_name)
+
+        if  max_tokens = config.max_position_embeddings
+        elif hasattr(config, 'n_positions'):
+            max_tokens = config.n_positions
+        elif hasattr(config, 'd_model'):
+            max_tokens = config.d_model  # for T5 models, d_model is a rough proxy
+        else:
+            max_tokens = "Unknown"
+    
+        return f"Model {model_name} loaded successfully! Max tokens: {max_tokens}"
+    except Exception as e:
+        return f"Failed to load model {model_name}. Error: {str(e)}"
+
+
+def summarize_text(input, min_length, max_length):
+    if summarizer is None:
+        return "No model loaded!"
+
+    input_tokens = tokenizer.encode(input, return_tensors="pt")
+    num_tokens = input_tokens.shape[1]
+    if num_tokens > max_tokens:
+        return f"Error: The input text has {num_tokens} tokens, which exceeds the maximum allowed {max_tokens} tokens. Please enter shorter text."
+
+    min_summary_length = int(num_tokens * (min_length / 100))
+    max_summary_length = int(num_tokens * (max_length / 100))
+
+    output = summarizer(input, min_length=min_summary_length, max_length=max_summary_length)
+    return output[0]['summary_text']
+
 
 model_path = hf_hub_download(repo_id="cis-lmu/glotlid", filename="model.bin")
 identification_model = fasttext.load_model(model_path)
@@ -172,20 +224,33 @@ def print_s(source_lang, target_lang, text0):
 demo = gr.Blocks()
 
 with demo:
-  gr.Markdown("Speech analyzer")
-  audio = gr.Audio(type="filepath", label = "Upload a file")
   text0 = gr.Textbox()
   text = gr.Textbox()
+  #gr.Markdown("Speech analyzer")
+  #audio = gr.Audio(type="filepath", label = "Upload a file")
+  model_dropdown = gr.Dropdown(choices=model_names, label="Choose a model", value="sshleifer/distilbart-cnn-12-6")
+  load_message = gr.Textbox(label="Load Status", interactive=False) 
+  b1 = gr.Button("Load Model")
+  min_length_slider = gr.Slider(minimum=0, maximum=100, step=1, label="Minimum Summary Length (%)", value=10)
+  max_length_slider = gr.Slider(minimum=0, maximum=100, step=1, label="Maximum Summary Length (%)", value=20) 
+
+  summarize_button = gr.Button("Summarize Text")
+
+  b1.click(fn=load_model, inputs=model_dropdown, outputs=load_message)
+  summarize_button.click(fn=summarize_text, inputs=[text0, min_length_slider, max_length_slider],
+                           outputs=text)
+    
+  
   source_lang = gr.Dropdown(label="Source lang", choices=list(lang_id.keys()), value=list(lang_id.keys())[0])
   target_lang = gr.Dropdown(label="target lang", choices=list(lang_id.keys()), value=list(lang_id.keys())[0])
   
   #gr.Examples(examples = list(lang_id.keys()),
    #             inputs=[
     #                source_lang])
-  b1 = gr.Button("convert to text")
+  #b1 = gr.Button("convert to text")
   b3 = gr.Button("translate")
   b3.click(translation_text, inputs = [source_lang, target_lang, text0], outputs = text)
-  b1.click(audio_a_text, inputs=audio, outputs=text)
+  #b1.click(audio_a_text, inputs=audio, outputs=text)
 
   b2 = gr.Button("Classification of language")
   b2.click(lang_ident,inputs  = text0, outputs=text)