Add pytorch model

config.json

@@ -19,6 +19,7 @@
   "num_hidden_layers": 24,
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
   "transformers_version": "4.11.0",
   "type_vocab_size": 1,
   "use_cache": true,

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4e949e396236fb74ea6aa041483146851bb721aa3babd0c42b017832e0efad2
-size 711588089
+oid sha256:2a878489306599e17d93ae80b2eed555b4c6b13fbd306d4fd88c06b1c2b8e7a0
+size 1421662309

flax_model_to_pytorch.py

@@ -0,0 +1,27 @@
+from transformers import RobertaForMaskedLM, FlaxRobertaForMaskedLM, AutoTokenizer
+import torch
+import numpy as np
+import jax
+import jax.numpy as jnp
+
+def to_f32(t):
+    return jax.tree_map(lambda x: x.astype(jnp.float32) if x.dtype == jnp.bfloat16 else x, t)
+
+jax.config.update('jax_platform_name', 'cpu')
+MODEL_PATH = "./"
+model = FlaxRobertaForMaskedLM.from_pretrained(MODEL_PATH)
+model.params = to_f32(model.params)
+model.save_pretrained(MODEL_PATH)
+
+pt_model = RobertaForMaskedLM.from_pretrained(
+    MODEL_PATH, from_flax=True).to('cpu')
+
+input_ids = np.asarray(2 * [128 * [0]], dtype=np.int32)
+input_ids_pt = torch.tensor(input_ids)
+
+logits_pt = pt_model(input_ids_pt).logits
+print(logits_pt)
+logits_fx = model(input_ids).logits
+print(logits_fx)
+
+pt_model.save_pretrained(MODEL_PATH)

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c90748f0745624d7d96b645cbb4fe73b836f6091ea1daae31871fc138cf302b8
+size 1421780139