update app.py

app.py

@@ -83,9 +83,9 @@ def create_bert(cache_dir=None):
     """Creates a GPT2 model, config, and tokenizer from the given name and revision"""
     from transformers import BertConfig
 
-    config = BertConfig.from_pretrained("./cs77_proj/bert_base/checkpoint-3848/config.json")
+    config = BertConfig.from_pretrained("./cs772_proj/bert_base/checkpoint-3848/config.json")
     tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
-    gpt = AutoModelForSequenceClassification.from_pretrained("./cs77_proj/bert_base/checkpoint-3848", config=config, cache_dir=cache_dir)
+    gpt = AutoModelForSequenceClassification.from_pretrained("./cs772_proj/bert_base/checkpoint-3848", config=config, cache_dir=cache_dir)
     print("loaded model")
     return config, tokenizer, gpt
 def interpret(text,label):
@@ -164,9 +164,9 @@ def interpret(text,label):
                         prob_offense = probabilities[0][l].item()
                         data.append({"layer": layer_i, "pos": pos_i, "prob": prob_offense})
                 df = pd.DataFrame(data)
-                df.to_csv(f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.csv")
+                df.to_csv(f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.csv")
             for stream in ["block_output", "mlp_activation", "attention_output"]:
-                    df = pd.read_csv(f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.csv")
+                    df = pd.read_csv(f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.csv")
                     df["layer"] = df["layer"].astype(int)
                     df["pos"] = df["pos"].astype(int)
                     prob_type = "p"+"("+label+")"
@@ -189,13 +189,13 @@ def interpret(text,label):
                         + theme(axis_text_y  = element_text(angle = 90, hjust = 1))
                     )
                     ggsave(
-                        plot, filename=f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.png", dpi=200
+                        plot, filename=f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.png", dpi=200
                     )
                     if stream == "mlp_activation":
-                        mlp_img_path = f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.png"
+                        mlp_img_path = f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.png"
                     elif stream=="block_output":
-                        bo_path = f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.png"
-                    else:attention_path = f"./cs77_proj/tutorial_data/pyvene_rome_{stream}.png"
+                        bo_path = f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.png"
+                    else:attention_path = f"./cs772_proj/tutorial_data/pyvene_rome_{stream}.png"
             return mlp_img_path,bo_path,attention_path
 
 def restore_corrupted_with_interval_config(
@@ -252,9 +252,9 @@ def create_bert(cache_dir=None):
     """Creates a GPT2 model, config, and tokenizer from the given name and revision"""
     from transformers import BertConfig
 
-    config = BertConfig.from_pretrained("./cs77_proj/bert_base/checkpoint-3848/config.json")
+    config = BertConfig.from_pretrained("./cs772_proj/bert_base/checkpoint-3848/config.json")
     tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
-    gpt = AutoModelForSequenceClassification.from_pretrained("./cs77_proj/bert_base/checkpoint-3848", config=config, cache_dir=cache_dir)
+    gpt = AutoModelForSequenceClassification.from_pretrained("./cs772_proj/bert_base/checkpoint-3848", config=config, cache_dir=cache_dir)
     print("loaded model")
     return config, tokenizer, gpt