使用Hugging Face Prompt Injection Identifier遇Pydantic ForwardRef错误求助

包版本 #

• langchain==0.1.14
• langchain-community==0.0.31
• langchain-core==0.1.38
• langchain-experimental==0.0.56
• pydantic==1.10.14

我正尝试按照官方教程实现提示注入识别功能，以下是加载注入识别器的代码片段：

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer, pipeline

tokenizer = AutoTokenizer.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection", 
                                           subfolder="onnx")
tokenizer.model_input_names = ["input_ids", "attention_mask"]
model = ORTModelForSequenceClassification.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection", 
                                                          export=False, 
                                                          subfolder="onnx",
                                                          file_name="model_optimized.onnx"
                                                         )

classifier = pipeline(
                     task="text-classification",
                     model=model,
                     tokenizer=tokenizer,
                     truncation=True,
                     max_length=512,
                    )

from langchain_experimental.prompt_injection_identifier import (
                                                        HuggingFaceInjectionIdentifier,
                                                        )

injection_identifier = HuggingFaceInjectionIdentifier(
                                                     model=classifier,
                                                    )

运行后触发如下错误：

ERROR: pydantic.errors.ConfigError: field "model" not yet prepared so type is still a 
ForwardRef, you might need to call 
HuggingFaceInjectionIdentifier.update_forward_refs().
CONTEXT: Traceback (most recent call last):

解决方案 #

这个错误是Pydantic前向引用未解析导致的，直接在初始化HuggingFaceInjectionIdentifier前调用update_forward_refs()方法就能解决。

修改后的代码：

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer, pipeline
from langchain_experimental.prompt_injection_identifier import HuggingFaceInjectionIdentifier

# 提前导入所有依赖模块，避免前向引用延迟解析
tokenizer = AutoTokenizer.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection", 
                                           subfolder="onnx")
tokenizer.model_input_names = ["input_ids", "attention_mask"]
model = ORTModelForSequenceClassification.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection", 
                                                          export=False, 
                                                          subfolder="onnx",
                                                          file_name="model_optimized.onnx"
                                                         )

classifier = pipeline(
                     task="text-classification",
                     model=model,
                     tokenizer=tokenizer,
                     truncation=True,
                     max_length=512,
                    )

# 解析前向引用
HuggingFaceInjectionIdentifier.update_forward_refs()

injection_identifier = HuggingFaceInjectionIdentifier(
                                                     model=classifier,
                                                    )

另外也可以调整导入顺序，把HuggingFaceInjectionIdentifier的导入放在最开头，确保类定义时相关类型已被加载。如果问题依旧，检查依赖包版本是否匹配，尤其是pydantic和langchain-experimental的兼容性。

内容的提问来源于stack exchange，提问作者AndCh