Sequence last fix (apache#16156)

* seq last fix

* index tensor to have int64

* fix dtypes

* revert unnecessary changes

* if seq len not passed, pass int64 dtype

* dtype comment

* use int32 or int64 as index dtype based on build flag

* Trigger notification

* Trigger notification

* lint fix

src/operator/sequence_last-inl.h

@@ -101,8 +101,8 @@ class SequenceLastOp : public Operator {
     using namespace mshadow::expr;
 
     int axis = param_.axis;
-    int out_size = out.size(0) * out.size(1);
-    int max_seq_len = data.size(axis);
+    index_t out_size = out.size(0) * out.size(1);
+    index_t max_seq_len = data.size(axis);
     index_t offset1 = axis ? out.size(1) : out_size;
     index_t offset2 = axis ? (max_seq_len * out.size(1)) : out.size(1);
 
@@ -121,11 +121,11 @@ class SequenceLastOp : public Operator {
     using namespace mshadow::expr;
 
     auto axis = param_.axis;
-    int batch = out_grad.size(0);
-    int rest = out_grad.size(1);
-    int out_size = batch * rest;
+    index_t batch = out_grad.size(0);
+    index_t rest = out_grad.size(1);
+    index_t out_size = batch * rest;
 
-    int max_seq_len = in_grad.size(axis);
+    index_t max_seq_len = in_grad.size(axis);
     index_t offset1 = axis ? rest : out_size;
     index_t offset2 = axis ? (max_seq_len * rest) : rest;
 

src/operator/sequence_last.cc

@@ -46,8 +46,14 @@ Operator *SequenceLastProp::CreateOperatorEx(Context ctx,
     DO_BIND_DISPATCH(CreateOp, param_, (*in_type)[0], (*in_type)[1]);
   }
 
-  // sequence_length not passed in, so fall back to using input array dtype for second argument
-  DO_BIND_DISPATCH(CreateOp, param_, (*in_type)[0], (*in_type)[0]);
+  // sequence_length not passed in, so fall back to using int32/int64 dtype for second argument
+  // second argument is the dtype of the sequence_length NDArray
+  // use int32 or int64 as index dtype based on build flag
+  #if MXNET_USE_INT64_TENSOR_SIZE == 1
+      DO_BIND_DISPATCH(CreateOp, param_, (*in_type)[0], mshadow::kInt64);
+  #else
+      DO_BIND_DISPATCH(CreateOp, param_, (*in_type)[0], mshadow::kInt32);
+  #endif
 }
 
 DMLC_REGISTER_PARAMETER(SequenceLastParam);

tests/nightly/test_large_vector.py

@@ -356,7 +356,9 @@ def test_sequence_last():
     # test with sequence length
     # parameter sequence_length - NDArray with shape (batch_size)
     # (2,3) indicates 2nd sequence from batch 1 and 3rd sequence from batch 2
-    b = nd.SequenceLast(a, sequence_length=mx.nd.array([2, 3]),
+    # need to mention dtype = int64 for sequence_length ndarray to support large indices
+    # else it defaults to float32 and errors
+    b = nd.SequenceLast(a, sequence_length=mx.nd.array([2, 3], dtype="int64"),
                         use_sequence_length=True)
     # check if it takes 2nd sequence from the first batch
     assert b[0] == a[1][0]