Skip to content

Randomly getting output full of repetitive newlines on LLaVA invocations #191

Closed

Description

@isidentical

Not sure if anyone else had hit into this but when using liuhaotian/llava-v1.5-13b with llava-hf/llava-1.5-13b-hf tokenizer, randomly I get outputs full of only newlines. The frequency of this happening increases when the temperature is close to 0 but also happens at 1 as well.

Activity

  1. jacob-hansen commented on Apr 24, 2024

    @jacob-hansen

    I've also hit this bug. My issue was that I had the generation outside of the assistant prompt. I was trying to be able to prompt the generation process (e.g. I give it what the model should start with).

    WRONG:
    def image_qa(s, image_path, question, system_prompt="", initial_assistant = ""):
    s += sgl.user(system_prompt + sgl.image(image_path) + question)
    s += sgl.assistant(initial_assistant)
    s += sgl.gen("answer")

    CORRECT:
    def image_qa(s, image_path, question, system_prompt="", initial_assistant = ""):
    s += sgl.user(system_prompt + sgl.image(image_path) + question)
    s += sgl.assistant(initial_assistant + sgl.gen("answer"))

    This fixed the issues.

  2. github-actions commented on Jul 25, 2024

    @github-actions
    Contributor

    This issue has been automatically closed due to inactivity. Please feel free to reopen it if needed.

  3. added a commit that references this issue on Feb 3, 2026
  4. added 2 commits that reference this issue on Jun 7, 2026
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