Update app.py

app.py

@@ -48,20 +48,22 @@ def get_sim(x):
 # Load pre-trained model 
 
 #model = GPT2LMHeadModel.from_pretrained('distilgpt2', output_hidden_states = True, output_attentions = True)
-model = GPT2LMHeadModel.from_pretrained('gpt2', output_hidden_states = True, output_attentions = True)
+#model = GPT2LMHeadModel.from_pretrained('gpt2', output_hidden_states = True, output_attentions = True)
 #model  =  gr.Interface.load('huggingface/distilgpt2', output_hidden_states = True, output_attentions = True)
 
 #model.eval()
 #tokenizer =  gr.Interface.load('huggingface/distilgpt2')
 
 #tokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')
-tokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')
 #tokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')
+#tokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')
+
+
 
+tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
+model = GPT2LMHeadModel.from_pretrained('gpt2')
 
 
-tokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')
-model = GPT2LMHeadModel.from_pretrained('distilgpt2')
 
 def sentence_prob_mean(text):
     # Tokenize the input text and add special tokens
@@ -85,37 +87,38 @@ def sentence_prob_mean(text):
     # Compute the mean probability across the tokens
     mean_prob = torch.mean(gathered_probs).item()
 
+    return mean_prob
 
 
-def cloze_prob(text):
+# def cloze_prob(text):
 
-	whole_text_encoding = tokenizer.encode(text)
-	text_list = text.split()
-	stem = ' '.join(text_list[:-1])
-	stem_encoding = tokenizer.encode(stem)
-	cw_encoding = whole_text_encoding[len(stem_encoding):]
-	tokens_tensor = torch.tensor([whole_text_encoding])
+# 	whole_text_encoding = tokenizer.encode(text)
+# 	text_list = text.split()
+# 	stem = ' '.join(text_list[:-1])
+# 	stem_encoding = tokenizer.encode(stem)
+# 	cw_encoding = whole_text_encoding[len(stem_encoding):]
+# 	tokens_tensor = torch.tensor([whole_text_encoding])
 	
-	with torch.no_grad():
-		outputs = model(tokens_tensor)
-		predictions = outputs[0]   
-
-	logprobs = []
-	start = -1-len(cw_encoding)
-	for j in range(start,-1,1):
-			raw_output = []
-			for i in predictions[-1][j]:
-					raw_output.append(i.item())
+# 	with torch.no_grad():
+# 		outputs = model(tokens_tensor)
+# 		predictions = outputs[0]   
+
+# 	logprobs = []
+# 	start = -1-len(cw_encoding)
+# 	for j in range(start,-1,1):
+# 			raw_output = []
+# 			for i in predictions[-1][j]:
+# 					raw_output.append(i.item())
 	
-			logprobs.append(np.log(softmax(raw_output)))
+# 			logprobs.append(np.log(softmax(raw_output)))
 			
 	
-	conditional_probs = []
-	for cw,prob in zip(cw_encoding,logprobs):
-			conditional_probs.append(prob[cw])
+# 	conditional_probs = []
+# 	for cw,prob in zip(cw_encoding,logprobs):
+# 			conditional_probs.append(prob[cw])
 
 
-	return np.exp(np.sum(conditional_probs))
+# 	return np.exp(np.sum(conditional_probs))