Skip to content

非常好奇使用GPT4之外的LLMs效果怎么样? #12

Description

@lizhixi212

如果使用GPT4之外的LLMs,处理一般复杂的任务,能有一个不错的效果,
那真是令人惊艳和兴奋。

Activity

  1. BlitherBoom812 commented on Dec 27, 2023

    @BlitherBoom812
    Collaborator

    首先感谢您对 ProAgent 的支持和期待!
    目前非GPT4的基座模型主要存在两个瓶颈:

    • 工具调用能力:很多基座模型不能很好地 follow 给定的 JSON Schema。在 RPA 应用场景下,节点生成、参数配置、工作流生成都是通过工具调用实现的,目前llama等开源模型在调用工具时比gpt4有更严重的幻觉问题以及格式不正确(JSON括号、缺少必要参数),在测试过程中,我们发现甚至 gpt3.5 也时常不能处理好结构复杂、深度较深的 Schema。
    • 长文本能力:目前 gpt4-32k 的上下文文本长度最适合于 ProAgent 场景的。单条workflow生成所需的 tokens 常常在 10k ~20k 之间,gpt3.5 在上下文长度上已经捉襟见肘。一方面期待基座模型继续升级上下文文本长度,另一方面我们也在努力解决 workflow 生成流程中的开销问题。
  2. Tridu33 commented on Feb 5, 2024

    @Tridu33

    can i use any other LLM but not openai?is there any other local LLM like llama instead? all just some Mock data instead of calling ChatGPT4?

    Unable to generate ChatCompletion response
    Exception: No API key provided. You can set your API key in code using 'openai.api_key = ', or you can set the environment variable OPENAI_API_KEY=). If your API key is stored in a file, you can point the openai module at it with 'openai.api_key_path = '. You can generate API keys in the OpenAI web interface. See https://platform.openai.com/account/api-keys for details.

  3. MMXC commented on Apr 26, 2024

    @MMXC

    如果使用GPT4之外的LLMs,处理一般复杂的任务,能有一个不错的效果, 那真是令人惊艳和兴奋。

    我用智谱替换了下,生成一个示例的workflow会卡在中间,不过也还行(指耗费接近1000w token的调试后)

  4. wangyunii commented on Jan 10, 2025

    @wangyunii

    如果使用GPT4之外的LLMs,处理一般复杂的任务,能有一个不错的效果, 那真是令人惊艳和兴奋。

    我用智谱替换了下,生成一个示例的workflow会卡在中间,不过也还行(指耗费接近1000w token的调试后)

    替换后可以在n8n上生成工作流了吗?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions