Updated model

added_tokens.json

@@ -0,0 +1 @@
+{"<|endoftext|>": 50265}

config.json

@@ -1,7 +1,8 @@
 {
+  "_name_or_path": ".",
   "activation_function": "gelu_new",
   "architectures": [
-    "GPT2Model"
+    "GPT2LMHeadModel"
   ],
   "attn_pdrop": 0.0,
   "bos_token_id": 50256,

evaluate.py

@@ -0,0 +1,8 @@
+from transformers import GPT2Tokenizer, GPT2Model, pipeline, set_seed
+ 
+tokenizer = GPT2Tokenizer.from_pretrained("flax-community/swe-gpt-wiki")
+model = GPT2Model.from_pretrained("flax-community/swe-gpt-wiki")
+
+generator = pipeline('text-generation', model=model)
+set_seed(42)
+generator("Hej, jag är en språkmodell,", max_length=30, num_return_sequences=5)

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f0735a9f1b2acfbb15130aab162d2cc8b323dabdc0399b25a7f348d0bb0de2d
-size 510397417
+oid sha256:226e91a5056e14d4bf8079bfc261958fbf6d8ec09a834091235ba01ca5fbf88e
+size 510401385

save_model.py

@@ -1,6 +1,8 @@
-from transformers.modeling_flax_pytorch_utils import load_flax_checkpoint_in_pytorch_model
-from transformers import GPT2Config, GPT2Model
-config = GPT2Config.from_pretrained("./")
-model = GPT2Model(config)
-load_flax_checkpoint_in_pytorch_model(model, "./flax_model.msgpack")
-model.save_pretrained("./")
+from transformers import AutoTokenizer, GPT2LMHeadModel
+'''
+This is a script to convert the Jax model and the tokenizer to Pytorch model
+'''
+model = GPT2LMHeadModel.from_pretrained(".", from_flax=True)
+model.save_pretrained(".")
+tokenizer = AutoTokenizer.from_pretrained(".")
+tokenizer.save_pretrained(".")

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<|endoftext|>", "bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "add_prefix_space": false, "special_tokens_map_file": null, "name_or_path": ".", "tokenizer_class": "GPT2Tokenizer"}