Merge branch 'develop' into think

luukunn · web-flow · commit beeef52442ef · 2025-09-29T00:33:35.000+08:00
diff --git a/fastdeploy/cache_manager/cache_messager.py b/fastdeploy/cache_manager/cache_messager.py
@@ -152,8 +152,8 @@ def __init__(
         cache_v = []
         self.messager = {}
         for layer_idx in range(self.num_layers):
-            key_cache = self.gpu_cache_kvs[f"key_caches_{layer_idx}_rank{self.rank}.device{gpu_id}"]
-            val_cache = self.gpu_cache_kvs[f"value_caches_{layer_idx}_rank{self.rank}.device{gpu_id}"]
+            key_cache = self.gpu_cache_kvs[f"key_caches_{layer_idx}_rank{self.rank}_device{gpu_id}"]
+            val_cache = self.gpu_cache_kvs[f"value_caches_{layer_idx}_rank{self.rank}_device{gpu_id}"]
             cache_k.append(key_cache)
             cache_v.append(val_cache)
             cache_k_ptr_list.append(key_cache.data_ptr())
diff --git a/fastdeploy/entrypoints/cli/main.py b/fastdeploy/entrypoints/cli/main.py
@@ -25,10 +25,12 @@ def main():
     import fastdeploy.entrypoints.cli.openai
     import fastdeploy.entrypoints.cli.run_batch
     import fastdeploy.entrypoints.cli.serve
+    import fastdeploy.entrypoints.cli.tokenizer
     from fastdeploy.utils import FlexibleArgumentParser
 
     CMD_MODULES = [
         fastdeploy.entrypoints.cli.run_batch,
+        fastdeploy.entrypoints.cli.tokenizer,
         fastdeploy.entrypoints.cli.openai,
         fastdeploy.entrypoints.cli.benchmark.main,
         fastdeploy.entrypoints.cli.serve,
diff --git a/fastdeploy/entrypoints/cli/tokenizer.py b/fastdeploy/entrypoints/cli/tokenizer.py
@@ -0,0 +1,249 @@
+"""
+# Copyright (c) 2025  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import typing
+from pathlib import Path
+
+from fastdeploy.entrypoints.cli.types import CLISubcommand
+from fastdeploy.input.preprocess import InputPreprocessor
+
+if typing.TYPE_CHECKING:
+    from fastdeploy.utils import FlexibleArgumentParser
+
+
+class TokenizerSubcommand(CLISubcommand):
+    """The `tokenizer` subcommand for the FastDeploy CLI."""
+
+    name = "tokenizer"
+
+    @staticmethod
+    def cmd(args: argparse.Namespace) -> None:
+        main(args)
+
+    def subparser_init(self, subparsers: argparse._SubParsersAction) -> FlexibleArgumentParser:
+        tokenizer_parser = subparsers.add_parser(
+            name=self.name,
+            help="Start the FastDeploy Tokenizer Server.",
+            description="Start the FastDeploy Tokenizer Server.",
+            usage="fastdeploy tokenizer [--encode/-e TEXT] [--decode/-d TEXT]",
+        )
+
+        # 添加通用参数
+        tokenizer_parser.add_argument(
+            "--model_name_or_path",
+            "--model",
+            "-m",
+            type=str,
+            default="baidu/ERNIE-4.5-0.3B-PT",
+            help="Path to model or model identifier",
+        )
+        tokenizer_parser.add_argument("--enable-mm", "-mm", action="store_true", help="Enable multi-modal support")
+        tokenizer_parser.add_argument("--vocab-size", "-vs", action="store_true", help="Show vocabulary size")
+        tokenizer_parser.add_argument("--info", "-i", action="store_true", help="Show tokenizer information")
+        tokenizer_parser.add_argument(
+            "--vocab-export", "-ve", type=str, metavar="FILE", help="Export vocabulary to file"
+        )
+        tokenizer_parser.add_argument("--encode", "-e", default=None, help="Encode text to tokens")
+        tokenizer_parser.add_argument("--decode", "-d", default=None, help="Decode tokens to text")
+
+        return tokenizer_parser
+
+
+def cmd_init() -> list[CLISubcommand]:
+    return [TokenizerSubcommand()]
+
+
+def get_vocab_size(tokenizer) -> int:
+    """获取词表大小"""
+    try:
+        if hasattr(tokenizer, "vocab_size"):
+            return tokenizer.vocab_size
+        elif hasattr(tokenizer, "get_vocab_size"):
+            return tokenizer.get_vocab_size()
+        else:
+            return 100295  # Ernie4_5Tokenizer的固定词表大小
+    except Exception:
+        return 0
+
+
+def get_tokenizer_info(tokenizer) -> dict:
+    """获取tokenizer的元信息"""
+    info = {}
+
+    try:
+        # 基本属性
+        info["vocab_size"] = get_vocab_size(tokenizer)
+
+        # 模型类型和路径
+        if hasattr(tokenizer, "name_or_path"):
+            info["model_name"] = tokenizer.name_or_path
+
+        # tokenizer类型
+        info["tokenizer_type"] = type(tokenizer).__name__
+
+        # 特殊符号
+        special_tokens = {}
+        for attr in ["bos_token", "eos_token", "unk_token", "sep_token", "pad_token", "cls_token", "mask_token"]:
+            if hasattr(tokenizer, attr):
+                token = getattr(tokenizer, attr)
+                if token:
+                    special_tokens[attr] = token
+        info["special_tokens"] = special_tokens
+
+        # 特殊token IDs
+        special_token_ids = {}
+        for attr in [
+            "bos_token_id",
+            "eos_token_id",
+            "unk_token_id",
+            "sep_token_id",
+            "pad_token_id",
+            "cls_token_id",
+            "mask_token_id",
+        ]:
+            if hasattr(tokenizer, attr):
+                token_id = getattr(tokenizer, attr)
+                if token_id is not None:
+                    special_token_ids[attr] = token_id
+        info["special_token_ids"] = special_token_ids
+
+        # 模型最大长度
+        if hasattr(tokenizer, "model_max_length"):
+            info["model_max_length"] = tokenizer.model_max_length
+
+    except Exception as e:
+        info["error"] = f"Failed to get tokenizer info: {e}"
+
+    return info
+
+
+def get_vocab_dict(tokenizer) -> dict:
+    """获取词表字典"""
+    try:
+        if hasattr(tokenizer, "vocab"):
+            return tokenizer.vocab
+        elif hasattr(tokenizer, "get_vocab"):
+            return tokenizer.get_vocab()
+        elif hasattr(tokenizer, "tokenizer") and hasattr(tokenizer.tokenizer, "vocab"):
+            return tokenizer.tokenizer.vocab
+        elif hasattr(tokenizer, "encoder"):
+            return tokenizer.encoder
+        else:
+            return {}
+    except Exception:
+        return {}
+
+
+def export_vocabulary(tokenizer, file_path: str) -> None:
+    """导出词表到文件"""
+    try:
+        vocab = get_vocab_dict(tokenizer)
+        if not vocab:
+            print("Warning: Could not retrieve vocabulary from tokenizer")
+            return
+
+        path = Path(file_path)
+        path.parent.mkdir(parents=True, exist_ok=True)
+
+        # 根据文件扩展名选择格式
+        if path.suffix.lower() == ".json":
+            with open(path, "w", encoding="utf-8") as f:
+                json.dump(vocab, f, ensure_ascii=False, indent=2)
+        else:
+            # 默认格式：每行一个token
+            with open(path, "w", encoding="utf-8") as f:
+                for token, token_id in sorted(vocab.items(), key=lambda x: x[1]):
+                    # 处理不可打印字符
+                    try:
+                        f.write(f"{token_id}\t{repr(token)}\n")
+                    except:
+                        f.write(f"{token_id}\t<unprintable>\n")
+
+        print(f"Vocabulary exported to: {file_path}")
+        print(f"Total tokens: {len(vocab)}")
+
+    except Exception as e:
+        print(f"Error exporting vocabulary: {e}")
+
+
+def main(args: argparse.Namespace) -> None:
+
+    def print_separator(title=""):
+        if title:
+            print(f"\n{'='*50}")
+            print(f" {title}")
+            print(f"{'='*50}")
+        else:
+            print(f"\n{'='*50}")
+
+    # 检查参数
+    if not any([args.encode, args.decode, args.vocab_size, args.info, args.vocab_export]):
+        print("请至少指定一个参数：--encode, --decode, --vocab-size, --info, --export-vocab")
+        return
+
+    # 初始化tokenizer
+    preprocessor = InputPreprocessor(model_name_or_path=args.model_name_or_path, enable_mm=args.enable_mm)
+    tokenizer = preprocessor.create_processor().tokenizer
+
+    # 执行操作
+    operations_count = 0
+
+    if args.encode:
+        print_separator("ENCODING")
+        print(f"Input text: {args.encode}")
+        encoded_text = tokenizer.encode(args.encode)
+        print(f"Encoded tokens: {encoded_text}")
+        operations_count += 1
+
+    if args.decode:
+        print_separator("DECODING")
+        print(f"Input tokens: {args.decode}")
+        try:
+            if isinstance(args.decode, str):
+                if args.decode.startswith("[") and args.decode.endswith("]"):
+                    tokens = eval(args.decode)
+                else:
+                    tokens = [int(x.strip()) for x in args.decode.split(",")]
+            else:
+                tokens = args.decode
+
+            decoded_text = tokenizer.decode(tokens)
+            print(f"Decoded text: {decoded_text}")
+        except Exception as e:
+            print(f"Error decoding tokens: {e}")
+        operations_count += 1
+
+    if args.vocab_size:
+        print_separator("VOCABULARY SIZE")
+        print(f"Vocabulary size: {get_vocab_size(tokenizer)}")
+        operations_count += 1
+
+    if args.info:
+        print_separator("TOKENIZER INFO")
+        print(json.dumps(get_tokenizer_info(tokenizer), indent=2))
+        operations_count += 1
+
+    if args.vocab_export:
+        print_separator("EXPORT VOCABULARY")
+        export_vocabulary(tokenizer, args.vocab_export)
+        operations_count += 1
+
+    print_separator()
+    print(f"Completed {operations_count} operation(s)")
diff --git a/tests/entrypoints/cli/test_tokenizer_cli.py b/tests/entrypoints/cli/test_tokenizer_cli.py
