Fix MiniCPM-o-2_6

[Wovchena]

Description

WWB uses _OVMiniCPMVForCausalLM.preprocess_inputs() for optimum-intel and --hf. That function fails to apply chat_template for MiniCPM-o-2_6 and MiniCPM-V-2_6. A fix is required for optimum-intel. It will take time to validate both of the models with that fix. This PR addresses other identified issues. The original implementation applies chat_template: https://huggingface.co/openbmb/MiniCPM-o-2_6/blob/main/modeling_minicpmo.py#L959 but it's not used in WWB.
WWB results for --weights-format fp16 and unpublished fix for optimum-intel chat_template:

1. --hf vs optimum-intel : 0.986911
2. --hf vs --genai: 0.935212
3. optimum-intel vs --genai 0.942425

Docs: https://wovchena.github.io/openvino.genai-public/docs/supported-models/#visual-language-models-vlms

That should be enough for this release but follow up PRs are needed. The ideal solution includes:

1. Fix optimum-intel chat_template application
2. Request WWB not to use optimum-intel with --hf
3. Align --hf and optimum-intel (likely to be never completed because the results are close: 0.986911)
4. Extend GenAI pre-commit tests
5. Copy image resize implementation from PIL to GenAI
6. Check if it's possible to align sin() and cos() with numpy with hardcoding commit sin() and cos() values

CVS-170987

Checklist:

• Tests have been updated or added to cover the new code
• This patch fully addresses the ticket.
• I have made corresponding changes to the documentation

[Copilot]

Pull Request Overview

This pull request fixes a bug in the WWB implementation where _OVMiniCPMVForCausalLM.preprocess_inputs() failed to apply chat templates for MiniCPM-o-2_6 and MiniCPM-V-2_6 models. The fix changes the image tag format from (<image>./</image>)\n to <image>./</image>\n across documentation and implementation files.

Key Changes

• Updated image tag format for MiniCPM models from (<image>./</image>)\n to <image>./</image>\n
• Added support for new MiniCPM-o-2_6 model variant
• Fixed related algorithm implementations in the MiniCPM vision encoder

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/python_tests/test_vlm_pipeline.py	Updated test to use correct image tag format without parentheses
src/python/py_vlm_pipeline.cpp	Updated documentation strings to show correct image tag format for both MiniCPM variants
src/python/openvino_genai/py_openvino_genai.pyi	Updated type stub documentation to reflect correct image tag format
src/cpp/src/visual_language/minicpm/classes.hpp	Updated resample method signature to use single target size with padding parameter
src/cpp/src/visual_language/minicpm/classes.cpp	Refactored implementation with improved algorithms and removed parentheses from image tag
src/cpp/src/visual_language/inputs_embedder.hpp	Updated comment to reflect correct image tag format
src/cpp/src/debug_utils.hpp	Added support for boolean numpy array type
src/cpp/include/openvino/genai/visual_language/pipeline.hpp	Updated API documentation with correct image tag format
site/docs/use-cases/image-processing/_sections/_usage_options/index.mdx	Updated documentation to separate MiniCPM-o-2_6 and MiniCPM-V-2_6 with correct tags
site/docs/supported-models/index.mdx	Added notes about MiniCPM-o compatibility issues
site/docs/supported-models/_components/vlm-models-table/models.ts	Added MiniCPM-o-2_6 model to supported models table
samples/cpp/visual_language_chat/README.md	Removed specific model reference to make documentation more generic

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}