Upload handler.py

handler.py

@@ -10,84 +10,84 @@ model.push_to_hub(repo_name="text-summary-gpt2-short", repo_id="Lin0He/text-summ
 tokenizer.push_to_hub(repo_name="text-summary-gpt2-short", repo_id="Lin0He/text-summary-gpt2-short")
 '''
 import torch
+from typing import Dict, List, Any
 from transformers import pipeline, AutoModel, AutoTokenizer
 
-def topk(probs, n=9):
-    # The scores are initially softmaxed to convert to probabilities
-    probs = torch.softmax(probs, dim= -1)
-
-    # PyTorch has its own topk method, which we use here
-    tokensProb, topIx = torch.topk(probs, k=n)
+class EndpointHandler:
+    def __init__(self, path=""):
+        # load model and tokenizer from path
+        self.tokenizer = AutoTokenizer.from_pretrained("Lin0He/text-summary-gpt2-short")
+        self.model = AutoModel.from_pretrained("Lin0He/text-summary-gpt2-short")
+    
+    def topk(probs, n=9):
+        # The scores are initially softmaxed to convert to probabilities
+        probs = torch.softmax(probs, dim= -1)
 
-    # The new selection pool (9 choices) is normalized
-    tokensProb = tokensProb / torch.sum(tokensProb)
+        # PyTorch has its own topk method, which we use here
+        tokensProb, topIx = torch.topk(probs, k=n)
 
-    # Send to CPU for numpy handling
-    tokensProb = tokensProb.cpu().detach().numpy()
+        # The new selection pool (9 choices) is normalized
+        tokensProb = tokensProb / torch.sum(tokensProb)
 
-    # Make a random choice from the pool based on the new prob distribution
-    choice = np.random.choice(n, 1, p = tokensProb)#[np.argmax(tokensProb)]#
-    tokenId = topIx[choice][0]
+        # Send to CPU for numpy handling
+        tokensProb = tokensProb.cpu().detach().numpy()
 
-    return int(tokenId)
+        # Make a random choice from the pool based on the new prob distribution
+        choice = np.random.choice(n, 1, p = tokensProb)#[np.argmax(tokensProb)]#
+        tokenId = topIx[choice][0]
 
-def model_infer(model, tokenizer, review, max_length=30):
-    # Preprocess the init token (task designator)
-    review_encoded = tokenizer.encode(review)
-    result = review_encoded
-    initial_input = torch.tensor(review_encoded).unsqueeze(0).to(device)
+        return int(tokenId)
 
-    with torch.set_grad_enabled(False):
-        # Feed the init token to the model
-        output = model(initial_input)
+    def model_infer(model, tokenizer, review, max_length=30):
+        # Preprocess the init token (task designator)
+        review_encoded = self.tokenizer.encode(review)
+        result = review_encoded
+        initial_input = torch.tensor(review_encoded).unsqueeze(0).to(device)
 
-        # Flatten the logits at the final time step
-        logits = output.logits[0,-1]
+        with torch.set_grad_enabled(False):
+            # Feed the init token to the model
+            output = self.model(initial_input)
 
-        # Make a top-k choice and append to the result
-        #choices = [topk(logits) for i in range(5)]
-        choices = topk(logits)
-        result.append(choices)
-        
-        # For max_length times:
-        for _ in range(max_length):
-            # Feed the current sequence to the model and make a choice
-            input = torch.tensor(result).unsqueeze(0).to(device)
-            output = model(input)
+            # Flatten the logits at the final time step
             logits = output.logits[0,-1]
-            res_id = topk(logits)
-
-            # If the chosen token is EOS, return the result
-            if res_id == tokenizer.eos_token_id:
-                return tokenizer.decode(result)
-            else: # Append to the sequence
-                result.append(res_id)
 
-    # IF no EOS is generated, return after the max_len
-    return tokenizer.decode(result)
+            # Make a top-k choice and append to the result
+            #choices = [topk(logits) for i in range(5)]
+            choices = self.topk(logits)
+            result.append(choices)
+            
+            # For max_length times:
+            for _ in range(max_length):
+                # Feed the current sequence to the model and make a choice
+                input = torch.tensor(result).unsqueeze(0).to(device)
+                output = self.model(input)
+                logits = output.logits[0,-1]
+                res_id = self.topk(logits)
+
+                # If the chosen token is EOS, return the result
+                if res_id == self.tokenizer.eos_token_id:
+                    return self.tokenizer.decode(result)
+                else: # Append to the sequence
+                    result.append(res_id)
+
+        # IF no EOS is generated, return after the max_len
+        return self.tokenizer.decode(result)
+
+    def predict(text):
+        result_text = []
+        for i in range(6):
+            summary = self.model_infer(self.model, self.tokenizer, input+"TL;DR").strip()
+            result_text.append(summary[len(input)+5:])
+        return sorted(result_text, key=len)[3]
+        #print("summary:", sorted(result_text, key=len)[3])
 
-def predict(text):
-    result_text = []
-    for i in range(6):
-        summary = model_infer(model, tokenizer, input+"TL;DR").strip()
-        result_text.append(summary[len(input)+5:])
-    return sorted(result_text, key=len)[3]
-    #print("summary:", sorted(result_text, key=len)[3])
-
-
-class EndpointHandler:
-    def __init__(self, path=""):
-        # load model and tokenizer from path
-        self.tokenizer = AutoTokenizer.from_pretrained("Lin0He/text-summary-gpt2-short")
-        self.model = AutoModel.from_pretrained("Lin0He/text-summary-gpt2-short")
-    
 
     def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
         # process input
         inputs = data.pop("inputs", data)
         # process input text
-        result = predict(inputs)
-        return {"text":result}
+        prediction = self.predict(inputs)
+        return {"text":prediction}