[WIP]: vl prefix caching

[RunningLeon]

Motivation

Please describe the motivation of this PR and the goal you want to achieve through this PR.

Modification

Please briefly describe what modification is made in this PR.

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a dependency on downstream projects of a newer version, this PR should be tested with all supported versions of downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[Copilot]

Copilot reviewed 33 out of 33 changed files in this pull request and generated no comments.

Comments suppressed due to low confidence (3)

lmdeploy/vl/model/deepseek.py:91

• [nitpick] Consider renaming the variable 'params' to 'image_params' for improved clarity in the context of image preprocessing.

for image, params in images:

lmdeploy/pytorch/paging/block_trie.py:81

• Accessing 'self.parent' in repr may raise an AttributeError since no getter for 'parent' is defined; consider using 'self._parent' or adding an appropriate property.

f'num_children={len(self.children)}, is_root={self.parent is None}, '

lmdeploy/api.py:71

• [nitpick] The automatic disabling of prefix caching for VLM has been removed here; please confirm that the VLM backend indeed supports prefix caching to avoid unexpected behavior.

_, pipeline_class = get_task(model_path)