Update Wil's code + typo

apex/amp/_process_optimizer.py

@@ -229,7 +229,7 @@ def post_backward_with_master_weights_fused(self, scaler):
     stash = self._amp_stash
     stash.scale = scaler.loss_scale()
     stash.grads = [[param.grad.data for param in group] for group in self.fp16_groups]
-    stash.output_params = [[param for param in in group] for group in self.fp16_groups]
+    stash.output_params = [[param for param in group] for group in self.fp16_groups]
 
     norm_groups = []
     skip = False

csrc/fused_adam_cuda_kernel.cu

@@ -65,7 +65,7 @@ struct AdamFunctor
     __device__ __forceinline__ void operator()(
         int chunk_size,
         volatile int* noop_gmem,
-        TensorList<DEPTH>& tl,
+        TensorListMetadata<DEPTH>& tl,
         const float b1,
         const float b2,
         const float eps,