Ditch caching_eval; reduce params more.

egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py

@@ -41,7 +41,6 @@ from scaling import (
     Identity,  # more friendly to backward hooks than nn.Identity(), for diagnostic reasons.
     penalize_abs_values_gt,
     softmax,
-    caching_eval,
     ScheduledFloat,
     FloatLike,
     limit_param_value,
@@ -318,12 +317,7 @@ class Zipformer(EncoderInterface):
             - lengths, a tensor of shape (batch_size,) containing the number
               of frames in `embeddings` before padding.
         """
-        if not torch.jit.is_scripting():
-            # This saves memory during training, at the expense of re-doing the encoder_embed
-            # computation in the backward pass.
-            x = caching_eval(x, self.encoder_embed)
-        else:
-            x = self.encoder_embed(x)
+        x = self.encoder_embed(x)
 
         x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
 
@@ -1693,7 +1687,7 @@ class ConvNeXt(nn.Module):
     """
     def __init__(self,
                  channels: int,
-                 hidden_ratio: int = 4,
+                 hidden_ratio: int = 3,
                  layerdrop_prob: FloatLike = None):
         super().__init__()
         kernel_size = 7