Azure Open AI GPT-4 model

Question

Azure Open AI GPT-4 model

Viachaslau Lapitski 0

Azure Open AI GPT-4 model (after deployment) says that it is actually GPT-3. How can i make sure that i work with 4th version of the model?

[{"role":"system","content":"You are an AI assistant that helps people find information."},{"role":"user","content":"What is you model name"},{"role":"assistant","content":"As an AI language model, I am powered by OpenAI's GPT-3."}]

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2023-04-28T08:01:22.0266667+00:00

@Viachaslau Lapitski The model is running GPT4 but the data it is trained on is until September 2021, please see reference here. This is an expected behavior until the training data is refreshed and a new model version is released under Azure OpenAI.

You might see a different response on similar model from OpenAI, this is because with OpenAI there are additional context and information passed as part of the generation which makes the response different. Azure OpenAI provides direct access to the underlying model so responses will be different at times.
Taivo Pungas 0 Reputation points

2023-08-09T08:48:06.8933333+00:00

Hi! I have a similar question - although my starting point is different. It seems that Azure gpt-4 (which I just got access to) is exactly as fast as Azure gpt-3.5-turbo, which seems way too good to be true - GPT-4 has so many more parameters that it just seems hard to believe.

However, the model deployment (and API responses) show that the correct gpt-4 model is being used.

Is there any way I can definitively tell whether it is GPT-4 or 3.5? Is there any chance there's a bug somewhere and my GPT-4 request is actually routed to GPT-3.5?

1 answer

Your answer

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2023-04-28T08:01:22.0266667+00:00

@Viachaslau Lapitski The model is running GPT4 but the data it is trained on is until September 2021, please see reference here. This is an expected behavior until the training data is refreshed and a new model version is released under Azure OpenAI.

You might see a different response on similar model from OpenAI, this is because with OpenAI there are additional context and information passed as part of the generation which makes the response different. Azure OpenAI provides direct access to the underlying model so responses will be different at times.
Taivo Pungas 0 Reputation points

2023-08-09T08:48:06.8933333+00:00

Hi! I have a similar question - although my starting point is different. It seems that Azure gpt-4 (which I just got access to) is exactly as fast as Azure gpt-3.5-turbo, which seems way too good to be true - GPT-4 has so many more parameters that it just seems hard to believe.

However, the model deployment (and API responses) show that the correct gpt-4 model is being used.

Is there any way I can definitively tell whether it is GPT-4 or 3.5? Is there any chance there's a bug somewhere and my GPT-4 request is actually routed to GPT-3.5?

Answer 1

强何 5

My deployed gpt-4 Why say it's gpt-3 when I suspect Microsoft is selling fakes

Share via

Azure Open AI GPT-4 model

1 answer

Your answer