fix(florence): fix text failure

[paul-nechifor]

Problem

Closes DIM-XXX

Solution

Breaking Changes

How to Test

Contributor License Agreement

• I have read and approved the CLA.

[greptile-apps]

Greptile Summary

This PR enhances the Florence2Model.caption() method to accept an optional detail parameter, allowing callers to override the caption detail level (brief/normal/detailed) at call time rather than only at initialization. It also adds a CaptionDetail.from_str() classmethod for convenient string-to-enum resolution.

• Added CaptionDetail.from_str() classmethod with aliases ("brief", "normal", "detailed", "more_detailed") for resolving detail-level strings to enum members
• Extended Florence2Model.caption() signature with an optional detail: str | CaptionDetail | None parameter that falls back to the instance default when None
• The caption_batch() method was not updated with the same detail parameter, creating an API inconsistency between single and batch captioning

Confidence Score: 4/5

• This PR is safe to merge with low risk — it adds an optional parameter with a sensible default, preserving backward compatibility.
• The change is small, backward-compatible, and well-structured. The only concern is the API inconsistency between caption() (supports detail override) and caption_batch() (does not), and a minor gap where from_str can raise an unhelpful KeyError on invalid input.
• No files require special attention; dimos/models/vl/florence.py is the only changed file and the changes are straightforward.

Important Files Changed

Filename	Overview
dimos/models/vl/florence.py	Adds CaptionDetail.from_str() classmethod and extends caption() with optional detail parameter; caption_batch() not updated with same override capability.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["caption(image, detail=None)"] --> B{detail is None?}
    B -- Yes --> C["Use self._task_prompt\n(instance default)"]
    B -- No --> D{isinstance CaptionDetail?}
    D -- Yes --> E["Use detail.value"]
    D -- No --> F["CaptionDetail.from_str(detail)"]
    F --> G{In _ALIASES?}
    G -- Yes --> H["Return alias match"]
    G -- No --> I["cls[name.upper()]\n(KeyError if invalid)"]
    H --> E
    I --> E
    C --> J["Process image with task_prompt"]
    E --> J
    J --> K["Generate caption"]

Loading

Comments Outside Diff (1)

1. dimos/models/vl/florence.py, line 138-148 (link)

   caption_batch missing detail override

   caption() now accepts an optional detail parameter, but caption_batch() still hardcodes self._task_prompt. This creates an API inconsistency — users who discover they can override detail level on single captions will expect the same on batches. Consider adding the same detail: str | CaptionDetail | None = None parameter here as well, with the same resolution logic.

_{Last reviewed commit: 126bb92}