Skip to content

Implement parse_dllm_trajectory — multi-step aggregation #265

@chicham

Description

@chicham

Changes

  • parse_dllm_trajectory(outputs, steps: list[int] | None = None) -> list[dict[int, list[float]]] in artefactual/preprocessing/dllm_parser.py
  • Aggregates mask entropy across denoising steps; steps=None uses all steps
  • steps=[-1] behaviour must match parse_dllm_logprobs

Acceptance criteria

  • Function importable from artefactual.preprocessing
  • Multi-step synthetic output tested
  • steps=[-1] produces identical output to parse_dllm_logprobs(outputs, step=-1)

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpriority:lowBacklog/Nice to havetaskConcrete implementation issue, child of an epic

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions