Форум Привет

elpresidente* писал(а): ↑Ср мар 06, 2024 2:28 pm Практически уверен что "maxOutputTokens": 30720

kyk писал(а): ↑Ср мар 06, 2024 2:50 pm

elpresidente* писал(а): ↑Ср мар 06, 2024 2:28 pm Практически уверен что "maxOutputTokens": 30720

Получается, что при достаточном кол-ве токенов, Gemini API послушно выполняет задание и не "ленится".

В том время, как OpenAI GPT-4 в аналогичной ситуации поступает по своему усмотрению, причём причём иногда может randomly выдать больше, а иногда меньше.

Today, we are releasing an updated GPT-4 Turbo preview model, gpt-4-0125-preview. This model completes tasks like code generation more thoroughly than the previous preview model and is intended to reduce cases of “laziness” where the model doesn’t complete a task.

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "The 2020 World Series was played in Texas at Globe Life Field in Arlington.",
        "role": "assistant"
      },
      "logprobs": null
    }
  ]
  }
}

The finish_reason field in the OpenAI API's response indicates why the model stopped generating text for a particular completion.

"stop": This means that the generation was stopped because the model encountered a stop sequence specified in the prompt or by the parameters in the API call. Stop sequences can be explicit (like a specific word or punctuation mark that signals the end of a text) or implicit based on the prompt structure.

"length": This value indicates that the generation reached the maximum length specified in the request. The API allows users to set a maximum number of tokens to be generated, and if the completion reaches this limit, the generation will stop with "length" as the finish_reason.

"token_limit": Similar to "length", this indicates that the generation has reached the total token limit set for the API call. It's a constraint on the maximum size of the output, ensuring that responses don't exceed the bounds set by the user or the limits of the API

Each of these finish reasons helps in understanding the boundaries of the generated text and can be useful in debugging or refining API requests. For example, if you consistently receive "length" as the finish_reason, you might consider increasing the maximum token limit for your requests if you need longer responses. .

 print(response.choices[0].finish_reason)

stop

kyk писал(а): ↑Ср мар 06, 2024 1:14 pm Gemini Pro с вашей задачей справился без проблем viewtopic.php?p=603502#p603502

response = model.generate_content()

kyk писал(а): ↑Чт мар 07, 2024 2:11 am Вот так получается?
'СhatGPT3.5' < 'Gemini Pro'< 'ChatGPT-4'

kyk писал(а): ↑Чт мар 07, 2024 2:11 am Вы в Gemini Pro API эту фунцию пользуете?

elpresidente* писал(а): ↑Чт мар 07, 2024 8:10 am Плагин на этом сайте использует REST API https://ai.google.dev/tutorials/rest_quickstart