Translation Interceptor

Application Example: Prompt Interceptor

In TextLLMPromptRunner, there's an interface called "Interceptor (PromptInterceptor)" for pre-processing and post-processing prompts.

By using the PromptInterceptor, you can modify the input before executing the prompt and modify the output after the prompt has been executed.

For instance, by using the ValueTranslationInterceptor, while it appears that the user is communicating in Japanese, the actual communication with the LLM is done in English.

graph TB
    JPInput[Input in Japanese]
    TranslateInput[Translation interceptor translates prompt input from Japanese to English]
    LLMCommunication["Communicate with LLM and obtain output in English"]
    TranslateOutput[Translation interceptor translates output from English to Japanese]
    JPOutput[Obtain results in Japanese]

    JPInput --> TranslateInput
    TranslateInput --> LLMCommunication
    LLMCommunication --> TranslateOutput
    TranslateOutput --> JPOutput

Input/Output Translation Interceptor Implementation Example

This complex process can be achieved with just a few changes in PromptoGen. The actual LLM for prompt execution and the LLM for translation can use different models.

import promptogen as pg
from promptogen.prompt_interceptor.translation_interceptor import ValueTranslationInterceptor

formatter = pg.KeyValuePromptFormatter()
llm = YourTextLLM(model="your-model")
translator_llm = YourTextLLM(model="your-model-translator")

interceptors = [
    ValueTranslationInterceptor(llm=translator_llm, from_lang="Japanese", to_lang="English"),
]

prompt_runner = pg.TextLLMPromptRunner(llm=llm, formatter=formatter, interceptors=interceptors)
# ...(omitted)

Implementation Example (GitHub)

translation_interceptor.py (GitHub)