[fix] Rotary embeddings respecting input types (#326)

* tentative fix

* fixing arange not knowing half on cpu

CHANGELOG.md

@@ -8,6 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## TBD
 ### Fixed
 - Removed dupliacated biases in the FusedMLP layers [#317]
+- Rotary embeddings respecting input types [#326]
 
 ### Added
 - Four blocksparsity layouts from DeepSpeed [#320]

tests/test_rotary_embeddings.py

@@ -52,16 +52,25 @@ def test_helper_methods():
 
 
 @pytest.mark.parametrize("device", DEVICES)
-def test_rotary_embeddings(device):
+@pytest.mark.parametrize("dtype", [torch.float16, torch.float32])
+def test_rotary_embeddings(device, dtype):
     rotary = RotaryEmbedding(EMB).to(device)
 
     # Generate dummy inputs
-    q = torch.ones((BATCH, HEADS, SEQ, EMB), device=device)  # uniform on purpose
+    q = torch.ones(
+        (BATCH, HEADS, SEQ, EMB), device=device, dtype=dtype
+    )  # uniform on purpose
     k = q.clone()
 
     q_rot, k_rot = rotary(q, k)
 
+    assert q_rot.dtype == q.dtype
+    assert k_rot.dtype == k.dtype
+
     # Check that the sequences now encode relative position information
+    q, k = q.float(), k.float()
+    q_rot, k_rot = q_rot.float(), k_rot.float()
+
     att = torch.einsum("bhne,bhme->bhnm", q, k)
     att_rot = torch.einsum("bhne,bhme->bhnm", q_rot, k_rot)
 

xformers/components/positional_embedding/rotary.py

@@ -61,16 +61,20 @@ def _update_cos_sin_tables(self, x, seq_dimension=1):
 
         # Reset the tables if the sequence length has changed,
         # or if we're on a new device (possibly due to tracing for instance)
-        if seq_len != self._seq_len_cached or self._cos_cached.device != x.device:
+        if (
+            seq_len != self._seq_len_cached
+            or self._cos_cached.device != x.device
+            or self._cos_cached.dtype != x.dtype
+        ):
             self._seq_len_cached = seq_len
-            t = torch.arange(x.shape[seq_dimension], device=x.device).type_as(
-                self.inv_freq
+            t = torch.arange(
+                x.shape[seq_dimension], device=x.device, dtype=torch.float32
             )
-            freqs = torch.einsum("i,j->ij", t, self.inv_freq)
+            freqs = torch.einsum("i,j->ij", t, self.inv_freq.to(x.dtype))
             emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
 
-            self._cos_cached = emb.cos()[None, None, :, :]
-            self._sin_cached = emb.sin()[None, None, :, :]
+            self._cos_cached = emb.cos()[None, None, :, :].to(x.dtype)
+            self._sin_cached = emb.sin()[None, None, :, :].to(x.dtype)
 
         return self._cos_cached, self._sin_cached