Add pytorch, tf version

events.out.tfevents.1626033670.t1v-n-a95a71e5-w-0.441100.3.v2

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec4230aa8ba86cfc41429756c8141fcc040e8fccd2a1d4ed6cc4289cc9d0d9af
-size 48083343
+oid sha256:7663ff60a0b876debfd15a5101a510a59721b473d85bf2e605a223328e96e047
+size 48307053

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cae1f3c7d3627c1b1ce2ebc6991542e88781e525b4ef12041c25900aea411d12
+size 1443523865

src/convert_flax_to_pytorch.py

@@ -0,0 +1,31 @@
+import torch
+import numpy as np
+
+import jax
+import jax.numpy as jnp
+
+from transformers import AutoTokenizer
+from transformers import FlaxGPT2LMHeadModel
+from transformers import GPT2LMHeadModel
+
+tokenizer = AutoTokenizer.from_pretrained("../")
+tokenizer.pad_token = tokenizer.eos_token
+
+model_fx = FlaxGPT2LMHeadModel.from_pretrained("../")
+
+# def to_f32(t):
+#     return jax.tree_map(lambda x: x.astype(jnp.float32) if x.dtype == jnp.bfloat16 else x, t)
+
+# model_fx.params = to_f32(model_fx.params)
+# model_fx.save_pretrained("./fx")
+
+model_pt = GPT2LMHeadModel.from_pretrained("../", from_flax=True)
+model_pt.save_pretrained("./pt")
+
+input_ids = np.asarray(2 * [128 * [0]], dtype=np.int32)
+input_ids_pt = torch.tensor(input_ids)
+
+logits_pt = model_pt(input_ids_pt).logits
+print(logits_pt)
+logits_fx = model_fx(input_ids).logits
+print(logits_fx)

src/convert_flax_to_tf.py

@@ -0,0 +1,24 @@
+import torch
+import numpy as np
+
+import jax
+import jax.numpy as jnp
+
+from transformers import AutoTokenizer
+from transformers import GPT2LMHeadModel
+from transformers import TFGPT2LMHeadModel
+
+tokenizer = AutoTokenizer.from_pretrained("../")
+tokenizer.pad_token = tokenizer.eos_token
+
+model_pt = GPT2LMHeadModel.from_pretrained("./pt")
+model_tf = TFGPT2LMHeadModel.from_pretrained("./pt", from_pt=True)
+model_tf.save_pretrained("./tf")
+
+input_ids = np.asarray(2 * [128 * [0]], dtype=np.int32)
+input_ids_pt = torch.tensor(input_ids)
+
+logits_pt = model_pt(input_ids_pt).logits
+print(logits_pt)
+logits_tf = model_tf(input_ids).logits
+print(logits_tf)

tf_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc7ac4bf62ab348f0729b3f0aebd539b072bb75a5bcffb7f5ec7778185f305f2
+size 1418594792