Skip to main content

Quick monkey-patching fix for OpenAI's hallucinated multi_tool_use.parallel

Project description

Patching multi_tool_use.parallel in OpenAI

Quick monkey-patching fix for OpenAI's hallucinated multi_tool_use.parallel issue.

This is a simple workaround for this issue where the GPT ChatCompletions API sometimes hallucinates a strange alternative way of calling multiple tools at once. In my experience, these calls at least always follow a predictable structure and can be hotfixed to look like normal, legal tool calls. This patch does that, and also rewrites the message accordingly (which, in my experience, makes the model perform calls work in the normal expected way.)

This fix will be redundant soon, but it looks like I am not the only one who occasionally runs into this error, so I thought I would share this workaround.

Installation and usage

pip install openai-multi-tool-use-parallel-patch

In your code:

import openai_multi_tool_use_parallel_patch  # import applies the patch
import openai

client = openai.AsyncOpenAI(...)  # sync client will be patched too

...

response = await client.chat.completions.create(...)  # no changes to the call signature or response vs vanilla OpenAI client

That's it - theoretically this should work even if you import openai first, but at the least you will need to make sure not to create your openai.OpenAI or openai.AsyncOpenAI instances before importing this patch.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

Built Distribution

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page