Skip to content

Pegasus prompt timestamp format is not accepted by FunClip parser #171

Description

@LauraGPT

Problem

PEGASUS_SYSTEM_PROMPT asks TwelveLabs Pegasus to return timestamps as decimal seconds, for example [12.5-15.0]. The downstream extract_timestamps() function only accepts SRT timestamps such as [00:00:12,500-00:00:15,000].

As a result, a response that follows the prompt exactly produces no clip ranges.

Reproduction

On current main (2ee37298):

PYTHONPATH=funclip python3 -c 'from utils.trans_utils import extract_timestamps; print(extract_timestamps("1. [12.5-15.0] highlight"))'

Output:

[]
[]

The first line is the function debug print; the second is the returned empty list.

Expected behavior

A Pegasus response in the format requested by the built-in prompt must be converted into millisecond clip ranges. Existing SRT-formatted LLM responses must remain supported.

Acceptance criteria

  • [12.5-15.0] produces [[12500, 15000]] in the Pegasus clipping path.
  • Existing [00:00:12,500-00:00:15,000] parsing remains unchanged.
  • Multiple segments and integer-second values are covered.
  • Malformed or reversed ranges do not silently create invalid clips.
  • A regression test passes the normalized Pegasus output into the same parser used by the AI clipping pipeline.

This mismatch was also identified during review of #165: #165 (comment)

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions