修改seq_len_to_mask的jittor实现及测试例中的一处传参错误

fastNLP/core/utils/seq_len_to_mask.py

@@ -74,7 +74,7 @@ def seq_len_to_mask(seq_len, max_len: Optional[int]=None):
         if isinstance(seq_len, jittor.Var):
             assert seq_len.ndim == 1, f"seq_len can only have one dimension, got {seq_len.ndim == 1}."
             batch_size = seq_len.shape[0]
-            broad_cast_seq_len = jittor.arange(max_len).expand(batch_size, -1)
+            broad_cast_seq_len = jittor.arange(max_len).reshape(1, max_len).expand(batch_size, -1)
             mask = broad_cast_seq_len < seq_len.unsqueeze(1)
             return mask
     except NameError as e:

tests/core/utils/test_seq_len_to_mask.py

@@ -78,7 +78,7 @@ class TestSeqLenToMask:
             mask = seq_len_to_mask(seq_len)
 
         # 3. pad到指定长度
-        seq_len = paddle.randint(1, 10, size=(10,))
+        seq_len = paddle.randint(1, 10, shape=(10,))
         mask = seq_len_to_mask(seq_len, 100)
         assert 100 == mask.shape[1]