imvladikon commited on
Commit
8016cd6
1 Parent(s): 44bfa66

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +15 -0
README.md CHANGED
@@ -41,6 +41,21 @@ for token in top_5_tokens:
41
  # 讗讬诇转 讛讬讗 诪讟专讜驻讜诇讬谉 讛诪讛讜讜讛 讗转 诪专讻讝 讛讻诇讻诇讛
42
  ```
43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
 
45
  When using AlephBertGimmel, please reference:
46
 
 
41
  # 讗讬诇转 讛讬讗 诪讟专讜驻讜诇讬谉 讛诪讛讜讜讛 讗转 诪专讻讝 讛讻诇讻诇讛
42
  ```
43
 
44
+ ```
45
+ def ppl_naive(text, model, tokenizer):
46
+ input = tokenizer.encode(text, return_tensors="pt")
47
+ loss = model(input, labels=input)[0]
48
+ return torch.exp(loss).item()
49
+
50
+ text = """{} 讛讬讗 注讬专 讛讘讬专讛 砖诇 诪讚讬谞转 讬砖专讗诇, 讜讛注讬专 讛讙讚讜诇讛 讘讬讜转专 讘讬砖专讗诇 讘讙讜讚诇 讛讗讜讻诇讜住讬讬讛"""
51
+
52
+ for word in ["讞讬驻讛", "讬专讜砖诇讬诐", "转诇 讗讘讬讘"]:
53
+ print(ppl_naive(text.format(word), model, tokenizer))
54
+
55
+ # 10.181422233581543
56
+ # 9.743313789367676
57
+ # 10.171016693115234
58
+ ```
59
 
60
  When using AlephBertGimmel, please reference:
61