feat: change ivy.lstm to always return sequences and states (ivy-llc#…

…28155)

ivy/functional/backends/torch/layers.py

@@ -883,8 +883,6 @@ def lstm(
     weights_transposed: bool = False,
     has_ih_bias: bool = True,
     has_hh_bias: bool = True,
-    return_sequences: bool = True,
-    return_states: bool = True,
 ):
     if weights_transposed:
         # transpose the weights if they are in the wrong format
@@ -924,15 +922,4 @@ def lstm(
         batch_first,
     )
 
-    if return_states:
-        if return_sequences:
-            return ret
-        else:
-            return tuple(
-                [ret[0][:, -1], ret[1], ret[2]]
-            )  # TODO: this depends on batch_first
-    else:
-        if return_sequences:
-            return ret[0]
-        else:
-            return ret[0][:, -1]
+    return ret[0][:, -1], ret[0], (ret[1], ret[2])

ivy/functional/frontends/torch/nn/functional/layer_functions.py

@@ -26,7 +26,7 @@ def _lstm_full(
     bidirectional,
     batch_first,
 ):
-    return ivy.lstm(
+    ret = ivy.lstm(
         input,
         hx,
         params,
@@ -38,6 +38,7 @@ def _lstm_full(
         has_ih_bias=has_biases,
         has_hh_bias=has_biases,
     )
+    return ret[1], ret[2][0], ret[2][1]
 
 
 def _lstm_packed(
@@ -51,7 +52,7 @@ def _lstm_packed(
     train,
     bidirectional,
 ):
-    return ivy.lstm(
+    ret = ivy.lstm(
         data,
         hx,
         params,
@@ -63,6 +64,7 @@ def _lstm_packed(
         has_ih_bias=has_biases,
         has_hh_bias=has_biases,
     )
+    return ret[1], ret[2][0], ret[2][1]
 
 
 # --- Main --- #

ivy/functional/ivy/layers.py

@@ -2397,8 +2397,6 @@ def lstm(
     weights_transposed: bool = False,
     has_ih_bias: bool = True,
     has_hh_bias: bool = True,
-    return_sequences: bool = True,
-    return_states: bool = True,
 ):
     """Applies a multi-layer long-short term memory to an input sequence.
 
@@ -2442,11 +2440,6 @@ def lstm(
         whether the `all_weights` argument includes a input-hidden bias
     has_hh_bias
         whether the `all_weights` argument includes a hidden-hidden bias
-    return_sequences
-        whether to return the last output in the output sequence,
-        or the full sequence
-    return_states
-        whether to return the final hidden and carry states in addition to the output
 
     Returns
     -------
@@ -2567,16 +2560,7 @@ def lstm(
     if batch_sizes is not None:
         output = _pack_padded_sequence(output, batch_sizes)[0]
 
-    if return_states:
-        if return_sequences:
-            return output, h_outs, c_outs
-        else:
-            return output[:, -1], h_outs, c_outs  # TODO: this depends on batch_first
-    else:
-        if return_sequences:
-            return output
-        else:
-            return output[:, -1]
+    return output[:, -1], output, (h_outs, c_outs)
 
 
 # Helpers #

ivy_tests/test_ivy/test_functional/test_nn/test_layers.py

@@ -90,8 +90,6 @@ def _lstm_helper(draw):
     has_ih_bias = draw(st.booleans())
     has_hh_bias = draw(st.booleans())
     weights_transposed = draw(st.booleans())
-    return_sequences = draw(st.booleans())
-    return_states = draw(st.booleans())
     bidirectional = draw(st.booleans())
     dropout = draw(st.floats(min_value=0, max_value=0.99))
     train = draw(st.booleans()) and not dropout
@@ -217,8 +215,6 @@ def _lstm_helper(draw):
             "weights_transposed": weights_transposed,
             "has_ih_bias": has_ih_bias,
             "has_hh_bias": has_hh_bias,
-            "return_sequences": return_sequences,
-            "return_states": return_states,
         }
     else:
         dtypes = dtype
@@ -234,8 +230,6 @@ def _lstm_helper(draw):
             "weights_transposed": weights_transposed,
             "has_ih_bias": has_ih_bias,
             "has_hh_bias": has_hh_bias,
-            "return_sequences": return_sequences,
-            "return_states": return_states,
         }
     return dtypes, kwargs