langchain

Registering a custom tool with `load_tools``

The issue seems to be that load_tools doesn’t really support loading custom tools, but instead complains about TypeError: unhashable type.

A workaround is to simply initialize an empty tools list with load_tools, and then extend it with your custom ones.

tools = load_tools([], llm=llm) + 
    [OhMyCustomTool(target_directory=target_directory, files=files)]

– via this GitHub Issue

Errors when initializing AzureChatOpenAI

Specifically, this beauty: “As of openai>=1.0.0, Azure endpoints should be specified via the azure_endpoint param not openai_api_base (or alias base_url). (type=value_error)”.

Such a frustrating issue, can’t believe it hasn’t been fixed already. What happens is that, if you have ANY kind of OPENAI_API_BASE environment variable set, the overly-eager error checker will throw the error, even if you’re specifying all parameters in AzureChatOpenAI’s constructor, as seen below:

llm = AzureChatOpenAI(
    api_key=config['AZURE_OPENAI_CHAT_KEY'],
    azure_endpoint=config["AZURE_OPENAI_CHAT_ENDPOINT"],
    api_version=config["AZURE_OPENAI_CHAT_API_VERSION"],
    model=config["AZURE_OPENAI_CHAT_MODEL"],
)

One way to debug this, is to simply cycle through all of os.environ, looking for OPENAI_ keys:

import os

for key, value in os.environ.items():
    if "OPENAI_" in key:
        print(f"{key}: {value}")

One way to fix this would be removing the OPENAI_API_BASE from os.environ, one way or another.

Another way would be to instantiate it like this:

llm = AzureChatOpenAI(
    api_key=config['AZURE_OPENAI_CHAT_KEY'],
    azure_endpoint=config["AZURE_OPENAI_CHAT_ENDPOINT"],
    api_version=config["AZURE_OPENAI_CHAT_API_VERSION"],
    model=config["AZURE_OPENAI_CHAT_MODEL"],
    validate_base_url=False
)

So terribly frustrating. Just so you know, here is the guilty code, you can also follow this issue for (hopefully) updates and/or a fix.