From 3577811abaf172996b732db7f42055a897b5c94f Mon Sep 17 00:00:00 2001
From: Ubuntu
 <zbzscript@zbgpu01.ndj4anicnnlexeer5gqrjnt1hc.ax.internal.cloudapp.net>
Date: Tue, 21 Jan 2025 13:15:00 +0000
Subject: [PATCH] :sparkles: implement sliding window

---
 compute_pppl.py | 73 +++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 58 insertions(+), 15 deletions(-)

diff --git a/compute_pppl.py b/compute_pppl.py
index 95c2ecb..7498f06 100644
--- a/compute_pppl.py
+++ b/compute_pppl.py
@@ -4,7 +4,7 @@ from transformers import AutoTokenizer, AutoModelForMaskedLM
 
 
 
-def calculate_pseudo_perplexity(text, model_name):
+def calculate_pseudo_perplexity(text, model_name, window_size):
     # Load the model and tokenizer
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     model = AutoModelForMaskedLM.from_pretrained(model_name)
@@ -19,11 +19,12 @@ def calculate_pseudo_perplexity(text, model_name):
     # Get tokenized words
     words = tokenizer.convert_ids_to_tokens(input_ids[0])
     num_tokens = input_ids.size(1)
+    trg_pos = window_size // 2
 
-    print(words)
-    print(num_tokens)
-    print(tokens)
-    print(word_ids)
+    #print(words)
+    #print(num_tokens)
+    #print(tokens)
+    #print(word_ids)
 
 
     pppl_per_word = []
@@ -34,21 +35,59 @@ def calculate_pseudo_perplexity(text, model_name):
 
     # Iteratively mask each token
     for i in range(1, num_tokens - 1):  # Skip [CLS] and [SEP] tokens
-        masked_input_ids = input_ids.clone()
-        masked_input_ids[0, i] = tokenizer.mask_token_id  # Mask the i-th token
-        print(i, masked_input_ids)
-        print(tokenizer.convert_ids_to_tokens(masked_input_ids[0]))
+
+        # Define the sliding window range
+        start_idx = max(1, i - trg_pos)
+        end_idx = min(num_tokens - 1, i + trg_pos + 1)
+
+        # Extract the window
+        window_ids = input_ids[0, start_idx:end_idx]
+        window_attention_mask = attention_mask[0, start_idx:end_idx]
+        #print("WID:", window_ids)
+        #print("WAM:", window_attention_mask)
+
+        # Pad the window if it's smaller than window_size
+        if end_idx - start_idx < window_size:
+            padding_left = max(0, trg_pos - i + 1)
+            #print("pos:", trg_pos, "i:", i, "num_tokens:", num_tokens)
+            #print(padding_left)
+            padding_right = max(0, trg_pos - ((num_tokens - 2) - i))
+            #print(padding_right)
+            window_ids = torch.cat(
+                [
+                    input_ids[0, :1].repeat(padding_left),  # [CLS] token for padding
+                    window_ids,
+                    input_ids[0, -1:].repeat(padding_right),  # [SEP] token for padding
+                ]
+            )
+            window_attention_mask = torch.cat(
+                [
+                    attention_mask[0, :1].repeat(padding_left),
+                    window_attention_mask,
+                    attention_mask[0, -1:].repeat(padding_right),
+                ]
+            )
+
+
+        #print("WID:", window_ids)
+        #print("WAM:", window_attention_mask)
+
+        # Mask the middle token in the window
+        masked_window_ids = window_ids.clone()
+        masked_window_ids[trg_pos] = tokenizer.mask_token_id
+        print(i, masked_window_ids)
+        print(tokenizer.convert_ids_to_tokens(masked_window_ids))
 
         # Predict masked token
         with torch.no_grad():
-            outputs = model(input_ids=masked_input_ids, attention_mask=attention_mask)
+            outputs = model(input_ids=masked_window_ids.unsqueeze(0), attention_mask=window_attention_mask.unsqueeze(0))
             logits = outputs.logits
 
             print(logits)
         
         # Compute log probability of the target token
-        target_token_id = input_ids[0, i]
-        log_probs = torch.log_softmax(logits[0, i], dim=-1)
+        target_token_id = window_ids[trg_pos]
+        log_probs = torch.log_softmax(logits[0, trg_pos], dim=-1)
         print("log prob mask:", log_probs[target_token_id].item())
         trg_log_prob = log_probs[target_token_id].item()
 
@@ -80,12 +119,16 @@ def calculate_pseudo_perplexity(text, model_name):
 
 def main():
     model_name = "bert-base-multilingual-cased"
+    window_size =11
     #example_sent = "Man träffc hochmüthige Leute an, die davor angesehen seyn wollten, sie wären dazu gesetzt, daß sie den Erdboden richten sollten, und dieser Wahn."
     #example_sent = "Hallo Welt!"
-    #example_sent= "Morgen gehen wir ins Kino."
-    example_sent = "Man träffc hochmüthige Leute an."
+    #example_sent= "Mofgen gehen win ims Kino und alle bekomen eine Tute Popcorn, weil si heute brv die Hausaufgben gemact haben."
+    example_sent= "Morgen gehen wir ins Kino und alle bekommen eine Tüte Popcorn, weil sie heute brav die Hausaufgaben gemacht haben."
+    #example_sent = "Ein grosser Baum steht vor dem Haus."
+
+    #example_sent = "Man träffc hochmüthige Leute an , die davor angesehen seyn wollten ."
     #example_sent = "Good morning!"
-    words, pseudo_perplexity = calculate_pseudo_perplexity(example_sent, model_name)
+    words, pseudo_perplexity = calculate_pseudo_perplexity(example_sent, model_name, window_size)
     for word, pppl in zip(words, pseudo_perplexity):
         print(word, ":", pppl)
 
-- 
GitLab