init

README.md

@@ -1,3 +1,9 @@
 ---
 license: apache-2.0
 ---
+
+## 谷歌bert的tf权重`chinese_L-12_H-768_A-12`转为torch版
+- 测试下来其实和[bert-base-chinese](https://huggingface.co/google-bert/bert-base-chinese)权重是一样的，建议使用`bert-base-chinese`
+- [github](https://github.com/google-research/bert)
+- [tf权重](https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip)
+- [转换命令](https://huggingface.co/docs/transformers/v4.28.1/en/converting_tensorflow_models)

bert4torch_config.json

@@ -0,0 +1,16 @@
+{
+    "attention_probs_dropout_prob": 0.1,
+    "hidden_act": "gelu",
+    "hidden_dropout_prob": 0.1,
+    "hidden_size": 768,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "layer_norm_eps": 1e-12,
+    "max_position_embeddings": 512,
+    "model": "bert",
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12,
+    "pad_token_id": 0,
+    "type_vocab_size": 2,
+    "vocab_size": 21128
+  }

config.json

@@ -0,0 +1,16 @@
+{
+    "attention_probs_dropout_prob": 0.1,
+    "hidden_act": "gelu",
+    "hidden_dropout_prob": 0.1,
+    "hidden_size": 768,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "layer_norm_eps": 1e-12,
+    "max_position_embeddings": 512,
+    "model": "bert",
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12,
+    "pad_token_id": 0,
+    "type_vocab_size": 2,
+    "vocab_size": 21128
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ff197792bc61b19e249391c49e025144e45c88071422f28422d98b080946254
+size 411613091