Add requirements.txt and app.py files

app.py

@@ -0,0 +1,47 @@
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForCausalLM
+import torch
+import pandas as pd
+import numpy as np
+import re
+import gradio as gr
+
+model_repo = "napatswift/mt5-fixpdftext"
+
+tokenizer = AutoTokenizer.from_pretrained(model_repo)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_repo)
+
+embedding = list(model.modules())[1]
+del model
+
+def get_embedding(text):
+    return embedding(tokenizer(text, return_tensors='pt').input_ids[0]).mean(axis=0)
+
+df = pd.read_csv('67_all_ministry.csv')
+
+def get_name(row):
+    for col, val in row.items():
+        if col.startswith('name_') and val and isinstance(val, str):
+            return val
+    return
+
+budget_items = df.apply(get_name, axis=1).unique().tolist()
+
+budget_item_embeddings = torch.stack(list(map(get_embedding, budget_items)))
+
+def get_closest_budget_item(text):
+    text_embedding = get_embedding(text)
+    scores = (budget_item_embeddings * text_embedding).sum(axis=1)
+    top_idx = scores.argsort(descending=True)[:5]
+    return pd.DataFrame({
+        'budget_item': np.array(budget_items)[top_idx],
+        'score': scores[top_idx].tolist()
+    })
+
+demo = gr.Interface(
+    fn=get_closest_budget_item,
+    inputs=gr.inputs.Textbox(lines=5, label="Text"),
+    outputs='dataframe',
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+sentencepiece
+transformers
+pandas
+numpy
+gradio