Enable the predictor masking by default (#5)
Browse files- Enable the predictor masking by default (ad8c07868194774e27efc20f85b668bf179d4c39)
Co-authored-by: Ahmed AbouElhamayed <[email protected]>
- modeling_llama_butler.py +1 -1
modeling_llama_butler.py
CHANGED
|
@@ -918,7 +918,7 @@ class LlamaAttentionExperimental(nn.Module):
|
|
| 918 |
self.num_key_value_groups = self.num_heads // self.num_key_value_heads
|
| 919 |
self.max_position_embeddings = config.max_position_embeddings
|
| 920 |
self.rope_theta = config.rope_theta
|
| 921 |
-
self.inference_mode =
|
| 922 |
self.producer = producer
|
| 923 |
self.layer_idx = layer_idx
|
| 924 |
self.token_sparse_method = None
|
|
|
|
| 918 |
self.num_key_value_groups = self.num_heads // self.num_key_value_heads
|
| 919 |
self.max_position_embeddings = config.max_position_embeddings
|
| 920 |
self.rope_theta = config.rope_theta
|
| 921 |
+
self.inference_mode = True
|
| 922 |
self.producer = producer
|
| 923 |
self.layer_idx = layer_idx
|
| 924 |
self.token_sparse_method = None
|