Hello,
Is here anyone who got access to API 32k? If so how are you achived that? I want it so much and Im wondering whatt kind of ideas you guys had and got it.
nda… u can just make it if u have the hardware… with minimal modifications to the basic api main scrpts
1 Like
Is it really possible to be done only this way? I have seen people talking that there are several ways to get it, however they didn’t mention any other option.
No, there is no way to make external modification to access the 32k model, it is a different endpoint and without your account having it enabled it will not work.
32k access is extremely limited as it requires a large amount of resources, it will become more accessible as time goes on and additional compute is added to the backend server infrastructure.
2 Likes
Can GPT-4 be deployed on a private server? Do you have any suggestions on how I can obtain such access?
OpenAI is not open source. Loading the model would require significant computational resources, likely around 20 A100 GPUs just for VRAM. There is a model in Poe with support for 100k tokens, but it is less advanced than gpt4. Additionally, API access is available through an application process only. While I can’t explicitly mention the company providing the API, you should be able to figure it out.
It may be possible to have OpenAI create a private instance for you, but you should be willing to pay a hundred thousand dollars upfront. If that’s the case you should contact the sales team at sales@openai.com
Another alternative could be using the API though Microsoft azure:
I hope that helps.
As I’m sure everyone is finding out, all paying API developers have GPT-4 API now 
I tested access on poe 100k but it is not suitable for my needs. Any idea what else I can do to access the GPT-4 API?
Okay I see, however is it possible to get an api key on azure?
Yes, as far as I know, but I don’t personally use azure.
Maybe someone else could elaborate on using the API through Microsoft Azure? 