Upload 2 files

app.py

@@ -0,0 +1,64 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+import nltk
+import torch
+import math
+
+model_name = "AGIvan/t5-base-title-generation"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+nltk.download("punkt")
+
+def generate_titles(text, num_titles=3, temperature=0.7):
+    # tokenize text
+    inputs = ["summarize: " + text]
+    inputs = tokenizer(inputs, return_tensors="pt")
+
+    num_tokens = len(inputs["input_ids"][0])
+    max_input_length = 512
+    num_spans = math.ceil(num_tokens / max_input_length)
+    overlap = math.ceil((num_spans * max_input_length - num_tokens) / max(num_spans - 1, 1))
+
+    spans_boundaries = []
+    start = 0
+    for i in range(num_spans):
+        spans_boundaries.append([start + max_input_length * i, start + max_input_length * (i + 1)])
+        start -= overlap
+
+    spans_boundaries_selected = []
+    j = 0
+    for _ in range(num_titles):
+        spans_boundaries_selected.append(spans_boundaries[j])
+        j += 1
+        if j == len(spans_boundaries):
+            j = 0
+
+    tensor_ids = [inputs["input_ids"][0][boundary[0]:boundary[1]] for boundary in spans_boundaries_selected]
+    tensor_masks = [inputs["attention_mask"][0][boundary[0]:boundary[1]] for boundary in spans_boundaries_selected]
+
+    inputs = {
+        "input_ids": torch.stack(tensor_ids),
+        "attention_mask": torch.stack(tensor_masks),
+    }
+
+    outputs = model.generate(**inputs, do_sample=True, temperature=temperature)
+    decoded_outputs = tokenizer.batch_decode(outputs, skip_special_tokens=True)
+    predicted_titles = [nltk.sent_tokenize(decoded_output.strip())[0] for decoded_output in decoded_outputs]
+
+    return predicted_titles
+
+# Gradio interface
+demo = gr.Interface(
+    fn=generate_titles,
+    inputs=[
+        gr.Textbox(label="Article text", lines=10, placeholder="Paste your article text here"),
+        gr.Slider(1, 10, value=3, step=1, label="Number of titles"),
+        gr.Slider(0.1, 1.5, value=0.7, step=0.05, label="Temperature"),
+    ],
+    outputs=gr.List(label="Generated titles"),
+    title="📰 T5 Title Generator",
+    description="Generate candidate titles for articles using a fine-tuned T5 model."
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+transformers
+torch
+nltk
+gradio
+sentencepiece