feat(transformers/model): add InternVL

[AndyZhou952]

What does this PR do?

Adds # (feature)

FEATURE: Add InternVL support in mindone.transformers (compatible w/ ver>=4.52.0)

1. add model internvl in mindone/transformers/models/internvl/modeling_internvl.py with an example
2. added UT in tests/transformers_tests/models/internvl/test_modeling_internvl.py
3. add _checkpoint_conversion_mapping support in modeling_utils.py for 4.50.0 compatibility

Example

from PIL import Image
from transformers import GotOcr2ImageProcessor, InternVLProcessor
import mindspore as ms
from mindone.transformers import InternVLForConditionalGeneration

MODEL_HUB = "OpenGVLab/InternVL3-1B-hf"
image = "demo.jpeg"

processor = InternVLProcessor.from_pretrained(MODEL_HUB)
image_processor = GotOcr2ImageProcessor.from_pretrained(MODEL_HUB)
processor.image_processor = image_processor

model = InternVLForConditionalGeneration.from_pretrained(
    MODEL_HUB,
    mindspore_dtype=ms.bfloat16,
    attn_implementation="eager",
)
image = Image.open(image)
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": image,
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)

inputs = processor(text=prompt, images=[image], return_tensors="np")
for k, v in inputs.items():
    tensor = ms.Tensor(v)
    if tensor.dtype == ms.int64:
        tensor = tensor.astype(ms.int32)
    else:
        tensor = tensor.astype(model.dtype)
    inputs[k] = tensor

generated_ids = model.generate(**inputs, max_new_tokens=500)
texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
print(texts

Performance

model	precision	attn_type	resolution	s/step
OpenGVLab/InternVL3-1B-hf	bp16	eager	2048x1365	0.1960

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
  documentation guidelines
• Did you build and run the code without any errors?
• Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
• Did you write any new necessary tests?

[gemini-code-assist]

Summary of Changes

Hello @AndyZhou952, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces the InternVL (InternVLForConditionalGeneration) model into the mindone.transformers library. This integration includes the full model architecture, necessary utility functions for checkpoint conversion, and comprehensive examples and unit tests to ensure functionality and proper usage.

Highlights

• InternVL Model Integration: Adds the InternVL model, including its vision encoder (InternVLVisionModel) and multimodal capabilities (InternVLModel, InternVLForConditionalGeneration), enabling combined image and text processing.
• Enhanced Checkpoint Loading: Implements _checkpoint_conversion_mapping in modeling_utils.py to facilitate seamless loading of pre-trained InternVL checkpoints by handling potential key mismatches.
• Automated Model Configuration: Updates the auto-configuration and auto-modeling mechanisms to automatically recognize and load InternVL models, streamlining model instantiation.
• Comprehensive Testing and Examples: Provides a new example script (examples/transformers/internvl/generate.py) demonstrating how to use the model for generation and dedicated unit tests (tests/transformers_tests/models/internvl/test_modeling_internvl.py) to validate its behavior.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces the InternVL model, a new vision-language model, along with supporting changes in the modeling utilities. The core model implementation is in mindone/transformers/models/internvl/modeling_internvl.py, and an example usage script is provided. The changes to modeling_utils.py add a flexible key-remapping mechanism for loading checkpoints, which is a valuable general improvement.

My review has identified a few issues. There's a potential bug in the pixel_shuffle method's condition check that could lead to a TypeError. The default value for an argument in the construct method of InternVLForConditionalGeneration might lead to inefficient computations. I've also suggested a refactoring in the example script for better code quality and pointed out a gap in the unit tests, which currently do not cover the main InternVLForConditionalGeneration model.

[SamitHuang]

consider disaggregating the name mapping function into two parts: 1) pt2hf, 2) hf2ms