Cross-attention tied to dbmdz/bert-base-german-cased

config.json

@@ -1,6 +1,7 @@
 {
-  "_name_or_path": "dbmdz/german-gpt2",
+  "_name_or_path": "josh-oo/german-gpt2-easy",
   "activation_function": "gelu_new",
+  "add_cross_attention": true,
   "architectures": [
     "GPT2LMHeadModel"
   ],
@@ -10,6 +11,7 @@
   "eos_token_id": 1,
   "gradient_checkpointing": false,
   "initializer_range": 0.02,
+  "is_decoder": true,
   "layer_norm_epsilon": 1e-05,
   "model_type": "gpt2",
   "n_ctx": 1024,

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd0841fa8bdd2afb18a1b73972a84da2caaf09e56e007a8253d407f78d12b2cb
-size 510421097
+oid sha256:ebc33e88b9e55cc9f0686af4790da8c8cfa5f041aef3a45e868dbbb4cf81fe46
+size 636511097