Revert to original behavior, specifying no upstream revision by default

config.json

@@ -8,7 +8,7 @@
   },
   "embedding_size": 512,
   "k_bucket_size": 1024,
-  "upstream_transformer_revision": "48bffbbd27bf028ecdd0cd55abb51236ec12ef1b",
+  "upstream_transformer_revision": null,
   "model_type": "LUAR",
   "q_bucket_size": 512,
   "torch_dtype": "float32",

config.py

@@ -1,4 +1,6 @@
 
+from typing import Optional
+
 from transformers import PretrainedConfig
 
 class LUARConfig(PretrainedConfig):
@@ -9,7 +11,7 @@ class LUARConfig(PretrainedConfig):
         use_memory_efficient_attention=False,
         q_bucket_size=512,
         k_bucket_size=1024,
-        upstream_transformer_revision=None,
+        upstream_transformer_revision: Optional[str] = None,
         **kwargs,
     ):
         self.embedding_size = embedding_size

model.py

@@ -1,6 +1,7 @@
 
 import math
 from functools import partial
+from typing import Optional
 
 import torch
 import torch.nn as nn
@@ -147,7 +148,7 @@ class LUAR(PreTrainedModel):
         )
         self.linear = nn.Linear(self.hidden_size, config.embedding_size)
 
-    def create_transformer(self, revision=None):
+    def create_transformer(self, revision: Optional[str] = None):
         """Creates the Transformer backbone.
         """
         kwargs = {"revision": revision} if revision else {}