fix: 修复特殊token 问题

This commit is contained in:
2026-04-02 16:16:02 +08:00
parent e01ec081b3
commit f44ad6912e
28 changed files with 334 additions and 163 deletions
+6 -5
View File
@@ -1,4 +1,5 @@
"""ChatML format strategy."""
from typing import List
from pipeline.tokenizer import BpeTokenizer
@@ -8,15 +9,15 @@ from pipeline.strategies.factory import StrategyFactory
@StrategyFactory.register("chatml")
class ChatMLStrategy(PromptStrategy):
"""ChatML format: ``<|im_start|>user ... <|im_end|> <|im_start|>assistant ... <|im_end|> <eos>``"""
"""ChatML format strategy."""
def __init__(
self,
tokenizer: BpeTokenizer,
user_start: str = "<|im_start|>user\n",
user_end: str = "<|im_end|>\n",
assistant_start: str = "<|im_start|>assistant\n",
assistant_end: str = "<|im_end|>\n<eos>",
user_start: str = "<imstart>user\n",
user_end: str = "<imend>\n",
assistant_start: str = "<imstart>assistant\n",
assistant_end: str = "<imend>\n",
):
super().__init__(tokenizer)