Update modeling_super_linear.py
Browse files- modeling_super_linear.py +1 -2
modeling_super_linear.py
CHANGED
|
@@ -613,7 +613,6 @@ class SuperLinearForCausalLM(PreTrainedModel, GenerationMixin):
|
|
| 613 |
# 4. Renormalise amplitudes:
|
| 614 |
# irfft divides by `target_len`, whereas the forward rfft used length `L`.
|
| 615 |
# Multiply by (target_len / L) so DC and low-freq amplitudes match input.
|
| 616 |
-
y *= target_len / L
|
| 617 |
|
| 618 |
return y
|
| 619 |
|
|
@@ -641,7 +640,7 @@ class SuperLinearForCausalLM(PreTrainedModel, GenerationMixin):
|
|
| 641 |
x_enc = self.fourier_interp_dim1(x_enc)
|
| 642 |
x_enc = self.revin_layer(x_enc, 'norm')
|
| 643 |
|
| 644 |
-
self.backbone.inf_pred_len =
|
| 645 |
|
| 646 |
# backbone returns (B, pred_len, C)
|
| 647 |
|
|
|
|
| 613 |
# 4. Renormalise amplitudes:
|
| 614 |
# irfft divides by `target_len`, whereas the forward rfft used length `L`.
|
| 615 |
# Multiply by (target_len / L) so DC and low-freq amplitudes match input.
|
|
|
|
| 616 |
|
| 617 |
return y
|
| 618 |
|
|
|
|
| 640 |
x_enc = self.fourier_interp_dim1(x_enc)
|
| 641 |
x_enc = self.revin_layer(x_enc, 'norm')
|
| 642 |
|
| 643 |
+
self.backbone.inf_pred_len = 720
|
| 644 |
|
| 645 |
# backbone returns (B, pred_len, C)
|
| 646 |
|