feat: Using the HuggingFace tokenizer (#4329)

apps/common/config/tokenizer_manage_config.py

@@ -6,6 +6,18 @@
     @date：2024/4/28 10:17
     @desc:
 """
+import os
+from pathlib import Path
+
+BASE_DIR = Path(__file__).resolve().parent.parent.parent
+
+
+class MKTokenizer:
+    def __init__(self, tokenizer):
+        self.tokenizer = tokenizer
+
+    def encode(self, text):
+        return self.tokenizer.encode(text).ids
 
 
 class TokenizerManage:
@@ -13,12 +25,8 @@ class TokenizerManage:
 
     @staticmethod
     def get_tokenizer():
-        from transformers import BertTokenizer
-        if TokenizerManage.tokenizer is None:
-            TokenizerManage.tokenizer = BertTokenizer.from_pretrained(
-                'bert-base-cased',
-                cache_dir="/opt/maxkb-app/model/tokenizer",
-                local_files_only=True,
-                resume_download=False,
-                force_download=False)
-        return TokenizerManage.tokenizer
+        from tokenizers import Tokenizer
+        # 创建Tokenizer
+        s = os.path.join(BASE_DIR.parent, 'tokenizer', 'bert-base-cased', 'tokenizer.json')
+        TokenizerManage.tokenizer = Tokenizer.from_file(s)
+        return MKTokenizer(TokenizerManage.tokenizer)

apps/models_provider/impl/tencent_cloud_model_provider/model/llm.py

@@ -6,9 +6,7 @@
     @date：2024/4/18 15:28
     @desc:
 """
-from typing import List, Dict
-
-from langchain_core.messages import BaseMessage, get_buffer_string
+from typing import Dict
 
 from common.config.tokenizer_manage_config import TokenizerManage
 from models_provider.base_model_provider import MaxKBBaseModel

tokenizer/bert-base-cased/config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "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_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.6.0.dev0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 28996
+}

tokenizer/bert-base-cased/tokenizer_config.json

@@ -0,0 +1 @@
+{"do_lower_case": false, "model_max_length": 512}