Support for o1-mini and o1-preview

[jame25]

Since there is now API access to these models, could you please add support for them. I've tried generating a new API key, however I still get the following error: state=finished raised KeyError

[hibobmaster]

Doc: https://platform.openai.com/docs/guides/reasoning

There is no compatible issue i think. Do you have the required tier to use this model?

[jame25]

Ah, that is probably the issue. I am only tier 1. I have however recently become able to interact with the API via the playground:

Thanks for taking the time to respond.

[jame25]

FYI, I got an email an hour ago confirming that I now have API access to both models (o1-mini and o1-preview). Also the related webpage has been updated:

I am still facing the same error: state=finished raised KeyError

[hibobmaster]

Can you share me your config with sensitive information redacted.

Besides, what's in container log?

[jame25]

The container log:

2024-11-20 12:15:07,453 - INFO - matrix chatgpt bot start.....
2024-11-20 12:15:10,841 - INFO - Successfully login via password
Mismatch in keys payload of device MatrixChatGPTBot (xxxxx) of user @bot:matrix.org (@bot:matrix.org).
2024-11-20 12:15:25,930 - INFO - Message received in room XXXX
User | !gpt hello
2024-11-20 12:15:30,729 - gpt - ERROR - RetryError[<Future at 0x7fb025e2f790 state=finished raised KeyError>]
2024-11-20 12:15:30,729 - ERROR - RetryError[<Future at 0x7fb025e2f790 state=finished raised KeyError>]
2024-11-20 12:15:32,003 - INFO - Message received in room XXXX
bot | > <@user:matrix.org> !gpt hello
Something went wrong, please try again or contact admin.

My config:

HOMESERVER="https://matrix.org"
USER_ID="@bot:matrix.org"
PASSWORD=xxxxx
DEVICE_ID="MatrixChatGPTBot"
ROOM_ID="!xxxxx:matrix.org"
OPENAI_API_KEY="xxxxxxxxxxxxxx"
GPT_API_ENDPOINT="https://api.openai.com/v1/chat/completions"
GPT_MODEL="o1-preview"
MAX_TOKENS=4000
TOP_P=1.0
PRESENCE_PENALTY=0.0
FREQUENCY_PENALTY=0.0
REPLY_COUNT=1
SYSTEM_PROMPT="You are ChatGPT, a large language model trained by OpenAI. Respond conversationally"
TEMPERATURE=0.8
TIMEOUT=120.0

[hibobmaster]

Does gpt-3.5-turbo work on well?

Can you use curl to do some test as well? If curl work on well, please paste the output here with sensitive info redacted.
https://platform.openai.com/docs/api-reference/chat/create?lang=curl

[jame25]

Yes, other models including gpt-3.5-turbo work fine. The curl request for o1-preview returned the following:

BadRequestError: Error code: 400 - {'error': {'message': "Unsupported value: 'messages[0].role' does not support 'system' with this model.", 'type': 'invalid_request_error', 'param': 'messages[0].role', 'code': 'unsupported_value'}}

Which in turn, lead me to stackoverflow. And subsequently the creation of the following python script - which does work with the o1-mini and o1-preview models:

import os
import requests
import json

# Retrieve your OpenAI API key from an environment variable
api_key = os.getenv("OPENAI_API_KEY")

# Check if the API key is available
if not api_key:
    raise ValueError("Please set the OPENAI_API_KEY environment variable.")

# Set up the URL and headers for the request
url = "https://api.openai.com/v1/chat/completions"

headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {api_key}"
}

# Define your prompt
prompt = "Can you tell me a joke?"

# Structure the messages accordingly
messages = [
    {
        "role": "user",
        "content": prompt  # Use the prompt directly as a string
    }
]

# Define the data payload for the request
data = {
    "model": "o1-preview",
    "messages": messages
}

# Send the POST request
response = requests.post(url, headers=headers, json=data)

# Check if the request was successful
if response.status_code == 200:
    # Extract the assistant's reply content
    reply_content = response.json()["choices"][0]["message"]["content"]

    # Print the assistant's reply (since it's already a string)
    print("Response from GPT:", reply_content)
else:
    # Print the error
    print(f"Status Code: {response.status_code}")
    print("Response:")
    print(response.json())

[hibobmaster]

Try: ghcr.io/hibobmaster/matrixchatgptbot:sha-c2d6ecfaf62f539f9e5ac0d321aa6c56002ec437

[jame25]

That version works great with both o1 models. Thank you.