feature/document_translation

[DaleMG]

Currently translates PDF,TXT, and DOCX files.

[kristapratico]

The filename says azure_document_translation, but it looks like we're importing from azure.ai.translation.text. The Azure AI document translation service translates full documents (e.g. pdf, docx, etc) from a source --> target container while the latter does realtime translation on text. I thought I saw a PR that already added the realtime translation feature, if this is a wrapper around the same service, maybe they should go into the same tool class?

[kristapratico]

Additionally, we should probably import azure.ai.translation.text and azure.core types under the init and raise any import errors so the user knows that they need to install the library to use this tool

[DaleMG]

I can move it in there for now but I have also been trying to get the document translation to work. Ill try to get it done before the meeting.

[DaleMG]

I am not entirely sure how to format the big try block with the unstructured imports.

[kristapratico]

Thanks for addressing the initial feedback, making good progress! A few more comments, I think the next step would be to add some tests and examples

[DaleMG]

I have no idea why the "make_spell_check" is failing.

[DaleMG]

I will work on the notebook/examples and tests now

[kristapratico]

I checked out your branch and gave the code a try myself. I think with the following changes we should be able to invoke the tool, or use with an agent:

file_path = "hello.txt"  # has the text "hola mundo"

ft = AzureFileTranslateTool(
    text_translation_key=os.environ["AZURE_TRANSLATE_API_KEY"],
    text_translation_endpoint=os.environ["AZURE_TRANSLATE_ENDPOINT"],
    region="westus2",
    file_path=file_path,
)
answer = ft.invoke({"query": f"What does the text in the file say in english?"})

print(answer)  # hello world

or

ft = AzureFileTranslateTool(
    text_translation_key=os.environ["AZURE_TRANSLATE_API_KEY"],
    text_translation_endpoint=os.environ["AZURE_TRANSLATE_ENDPOINT"],
    region="westus2",
)

tools = [ft]

model = AzureChatOpenAI(
    openai_api_version=os.environ["OPENAI_API_VERSION"],
    azure_deployment=os.environ["COMPLETIONS_MODEL"],
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    api_key=os.environ["AZURE_OPENAI_API_KEY"],
)

agent = create_structured_chat_agent(model, tools, prompt)

agent_executor = AgentExecutor(
    agent=agent, tools=tools, verbose=True, handle_parsing_errors=True
)

file_path = "hello.txt"

answer = agent_executor.invoke(
    {"input": f"What does the text in the file say in english? : {file_path}"}
)
print(answer) 

{
"action": "Final Answer",
"action_input": "The text in the file says 'Hello world' in English."
}

[kristapratico]

I have no idea why the "make_spell_check" is failing.

My guess is that this has nothing to do with your code changes and is a bug on the repo. I suspect if you pull in the latest from master, it may resolve.

[kristapratico]

Looks good, nice job! I think we just need to add an example of using the tool and showing its output

[kristapratico]

validate_environment is called before this function so we'll already have raised an ImportError if the translation library is not installed. We can remove the try/except here