Merge pull request meta-llama#127 from MarkSchmidty/patch-1

changed max_seq_len 1024 to 2048
drakejwong · Mar 5, 2023 · e22bdb2 · e22bdb2
2 parents 2845af9 + a81fb4e
commit e22bdb2
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/llama/model.py b/llama/model.py
@@ -27,7 +27,7 @@ class ModelArgs:
     norm_eps: float = 1e-5
 
     max_batch_size: int = 32
-    max_seq_len: int = 1024
+    max_seq_len: int = 2048
 
 
 class RMSNorm(torch.nn.Module):