GPT-3でAPI呼び出し時におけるコンテクストの総量上限を取り払う

概要

GPT-3でCompletionのEndPointを叩く際，トークンが2049を超えるとエラーがでます．
これを取り除く方法です．

GPT-3のansewer APIでエラーが出る

APIのレスポンスに下記のテキストが返されるようになります．

　　　 import openai
      response = openai.Completion.create(
            engine=engine,
            prompt=prompt,
            temperature=float(temperature),
            max_tokens=int(max_tokens),
            top_p=1.0,
            frequency_penalty=0.0,
            presence_penalty=0.0,
            stop=["###"]
            )
        export_text = response["choices"][0]['text']
        print(export_text)

You requested a completion of 2076 tokens 
(you supplied a prompt of length 1576 and requested to sample 500), 
but this model's maximum context length is 2049. 
If you would like us to add a feature to auto-truncate server-side, 
let us know at [e-mail].

対処

そのままだが，サポートにいってマニュアルで撤廃してもらいます．

送ったメール

Hello. My name is and GPT-3 works great.When I use the factual answering API I got an error saying that 
'''You requested a completion of 2076 tokens (you supplied a prompt of length 1576 and requested to sample 500), but this model's maximum context length is 2049. If you would like us to add a feature to auto-truncate server-side, let us know at [e-mail].'''
Can you please stop auto-truncate service on server side, please?
Information about my account is as follows.

• mail: [e-mail]
• Organization title:
  Best regards, 

その後

月曜日の8時に連絡したら，４時間後の12時半に連絡がきて解除されました．

Hi , we are not currently auto-truncating server-side so you should be all set. This message is asking for feedback if you would like us to add auto-truncation as a feature. I hope that helps!