Update app.py

app.py

@@ -12,7 +12,7 @@ import matplotlib.pyplot as plt
 import io
 from PIL import Image
 import matplotlib.font_manager as fm
-import util
+
 # global var
 MODEL_NAME = 'yseop/distilbert-base-financial-relation-extraction'
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
@@ -29,23 +29,7 @@ for font in fm.findSystemFonts(font_dir):
     print(font)
     fm.fontManager.addfont(font)
 plt.rcParams["font.family"] = 'NanumGothicCoding'
-def visualize_attention(sent, attention_matrix, n_words=10):
-    def draw(data, x, y, ax):
-        seaborn.heatmap(data, 
-                        xticklabels=x, square=True, yticklabels=y, vmin=0.0, vmax=1.0, 
-                        cbar=False, ax=ax)
-        
-    # make plt figure with 1x6 subplots
-    fig = plt.figure(figsize=(16, 8))
-    # fig.subplots_adjust(hspace=0.7, wspace=0.2)
-    for i, layer in enumerate(range(1, 12, 2)):
-        ax = fig.add_subplot(2, 3, i+1)
-        ax.set_title("Layer {}".format(layer))
-        draw(attention_matrix[layer], sent if layer > 6 else [], sent if layer in [1,7] else [], ax=ax)
- 
-    fig.tight_layout()
-    plt.close()
-    return fig
+
 def change_model_name(name):
     MODEL_BUF["name"] = name
     MODEL_BUF["tokenizer"] = AutoTokenizer.from_pretrained(name)
@@ -59,8 +43,6 @@ def predict(model_name, text):
     model = MODEL_BUF["model"]
     config = MODEL_BUF["config"]
     tokenized_text = tokenizer([text], return_tensors='pt')
-    input_tokens = tokenizer.convert_ids_to_tokens(tokenized_text.input_ids[0])
-    input_tokens = util.bytetokens_to_unicdode(input_tokens) if config.model_type in ['roberta', 'gpt', 'gpt2'] else input_tokens
     model.eval()
     output, attention = model(**tokenized_text, output_attentions=True, return_dict=False)
     output = F.softmax(output, dim=-1)
@@ -68,8 +50,7 @@ def predict(model_name, text):
     
     for idx, label in enumerate(output[0].detach().numpy()):
         result[config.id2label[idx]] = float(label)
-    fig = visualize_attention(input_tokens, attention[0][0].detach().numpy())
-    return result, fig#.logits.detach()#.numpy()#, output.attentions.detach().numpy()
+    return result
 if __name__ == '__main__':
     text = 'An A-B trust is a joint trust created by a married couple for the purpose of minimizing estate taxes.'
     model_name_list = [
@@ -78,7 +59,7 @@ if __name__ == '__main__':
     #Create a gradio app with a button that calls predict()
     app = gr.Interface(
         fn=predict,
-        inputs=[gr.inputs.Dropdown(model_name_list, label="Model Name"), 'text'], outputs=['label', 'plot'], 
+        inputs=[gr.inputs.Dropdown(model_name_list, label="Model Name"), 'text'], outputs=['label'], 
         examples = [[MODEL_BUF["name"], text]],
         title="FReE",
         description="Financial relations classifier"