trying to add OpenAI compatible z.AI (GLM-4.7)

[farangkao]

i've have tested z.AI with GLM-4.7 (using Coding Plan, which is a very good deal) via my own Python script like this:

from openai import OpenAI

def main():
    client = OpenAI(
        api_key="secret-api-key",
        base_url="https://api.z.ai/api/coding/paas/v4"
    )

    try:
        response = client.chat.completions.create(
            model="glm-4.7",
            messages=[
                {
                    "role": "user",
                    "content": "Hello! Please introduce yourself and tell me what you can do."
                }
            ],
            max_tokens=500,
            temperature=0.7
        )
 ...

This will work.
I then did the following with "llm":
Adding an api key via llm keys set openai

adding the file ~/.config/io.datasette.llm/extra-openai-models.yaml with content:

- model_id: glm-4.7
 model_name: glm-4.7
 api_base: "https://api.z.ai/api/coding/pass/v4"

then trying to test it with:
llm -m glm-4.7 "Short Description of what an llm is"

but got the error:

Error: Error code: 500 - {'error': {'code': '500', 'message': '404 NOT_FOUND'}}

any suggestions?

[drakoai777-creator]

check this:

The error is caused by a typo in the api_base URL within your YAML configuration.

Your Python script used: .../paas/v4 (Double 'a' for Platform-as-a-Service)

Your YAML file used: .../pass/v4 (Double 's')

The server returns a 500/404 because the /pass/ endpoint does not exist.

[cornellsh]

Fixed, use this https://github.com/cornellsh/codex-proxy

[farangkao]

@drakoai777-creator : thanks for the hint, we are a step further, it still doesn't work,
but the message is now:

Error: Error code: 401 - {'error': {'code': '401', 'message': 'token expired or incorrect'}}

I double checked the keys, and also used a key that works with my python script to be sure.
Is there a way to check more debug info (which endpoint is really hit, etc.) ?

Note: i double-checked in the keys.json file, the api-key is correctly stored in "openai", and i use the -m to trigger it.

-- Update: just noticed your code is all python, so i might be able to checkout the source and play around myself to find the culprit :)

@cornellsh : i think this is not related