Fix uninitialised rotary inv_freq on transformers v5

rotary.py

@@ -534,8 +534,10 @@ class RotaryEmbedding(torch.nn.Module):
             # We want fp32 here, not self.inv_freq.dtype, since the model could be loaded in bf16
             # And the output of arange can be quite large, so bf16 would lose a lot of precision.
             # However, for compatibility reason, we add an option to use the dtype of self.inv_freq.
-            if rotary_base_changed:
-                self.inv_freq = self._compute_inv_freq(device=device)
+            # NOTE: transformers>=5 builds the model on the meta device and materialises
+            # non-persistent buffers without re-running __init__, so the inv_freq computed
+            # there can be uninitialised memory. Recompute whenever the cache is (re)built.
+            self.inv_freq = self._compute_inv_freq(device=device)
             if self.pos_idx_in_fp32:
                 t = torch.arange(seqlen, device=device, dtype=torch.float32)
                 # We want fp32 here as well since inv_freq will be multiplied with t, and the output