Open hongyi-zhao opened 1 year ago
It doesn't.
You can tell it to by setting API_REVERSE_PROXY="https://bypass.churchless.tech/conversation"
or something like that
churchless.tech
are constructed based on https://github.com/acheong08/ChatGPT-Proxy-V4?https://bypass.churchless.tech/conversation https://free.churchless.tech/v1/chat/completions
churchless.tech
?See https://github.com/binary-husky/gpt_academic/issues/900#issuecomment-1615961556 for the related discussion.
Do you mean that all the endpoints exposed under the domain churchless.tech are constructed based on https://github.com/acheong08/ChatGPT-Proxy-V4?
No. The bypass endpoint can be used if your IP is banned by Cloudflare/OpenAI
What is the difference between the following endpoints?
Bypass: Gets around cloudflare. API format is identical to chat.openai.com
Free: Gets around cloudflare and makes the API format identical to api.openai.com
Bypass: Gets around cloudflare. API format is identical to chat.openai.com
If I understand correctly, the endpoint of this method is only applicable to applications that are compatible with web-based access, aka, without using API key.
For example, the https://github.com/binary-husky/gpt_academic works based on the API key, so the Bypass
endpoint cannot be used directly with it, while the Free
endpoint does, as you have commented here.
For example, the https://github.com/binary-husky/gpt_academic works based on the API key, so the Bypass endpoint cannot be used directly with it, while the Free endpoint does, as you have commented https://github.com/binary-husky/gpt_academic/issues/900#issuecomment-1615961556.
It's not a matter of whether it uses API key or not. The difference is in the API response format:
{"id":"chatcmpl-QXlha2FBbmROaXhpZUFyZUF3ZXNvbWUK","object":"chat.completion","created":0,"model":"gpt-3.5-turbo-0301","usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0},"choices":[{"index":0,"message":{"role":"assistant","content":"1"},"finish_reason":null}]}
vs
{"message": {"id": "787e1451-6c7d-4a07-bf33-16a99ec51755", "author": {"role": "assistant",
"name": null, "metadata": {}}, "create_time": 1688381993.772501, "update_time": null,
"content": {"content_type": "text", "parts": ["Hello! How can I assist you today?\n"]},
"status": "finished_successfully", "end_turn": true, "weight": 1.0, "metadata":
{"message_type": "next", "model_slug": "gpt-4", "finish_details": {"type": "stop", "stop":
"<|diff_marker|>"}}, "recipient": "all"}, "conversation_id":
"c6639e58-2ebc-470e-a57e-d68ec5dcce0d", "error": null}
Why did the returned content, aka, ["Hello! How can I assist you today?"]
, not appear in the former response?
Oh. It was a different prompt
The response was "1"
The response was "1"
I see.
But how can I use https://bypass.churchless.tech/conversation as the endpoint of https://github.com/binary-husky/gpt_academic, as what you have commented here?
cant
IMO, the following two packages are wonderful and key tools when working with OpenAI API:
https://github.com/acheong08/ChatGPT-to-API https://github.com/acheong08/ChatGPT-Proxy-V4
So, I want to know if the ChatGPT-TO-API calls ChatGPT-PROXY-V4 in the background to further simplify the access workflow?