Add Tensor Input Support: Enable .pt file processing with <tensor> tags for latent representations

[Marshall-mk]

PR type

• [ ] Bug Fix
• [x] New Feature
• [ ] Document Updates
• [ ] More Models or Datasets Support

PR information

Write the detail information belongs to this PR.

Experiment results

Paste your experiment result here(if needed).

[gemini-code-assist[bot]]

Summary of Changes

Hello @Marshall-mk, I'm Gemini Code Assist[^1]! 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 significantly enhances the ms-swift framework by adding native support for PyTorch tensor files (.pt) as a new multimodal input. This feature allows for greater flexibility in handling specialized data, such as pre-processed medical scans, scientific data, or custom feature representations, by seamlessly integrating them into the existing multimodal processing pipeline. The change includes updates to the core processing logic, data structures, and comprehensive documentation to guide users on its implementation and usage.

Highlights

• Tensor Input Support: Introduced the ability to process PyTorch tensor files (.pt) as a new multimodal input type, alongside images, videos, and audios.
• Flexible Tagging and Loading: Users can now specify tensor inputs using <tensor> tags within messages and provide file paths via a new tensors parameter in InferRequest and dataset formats.
• Automatic Tensor-to-Image Conversion: Tensors are automatically loaded and converted into PIL Images for model processing, supporting various shapes (2D, 3D, 4D) and channel formats (grayscale, RGB), with proper normalization.
• Comprehensive Documentation: Added a new dedicated documentation page (Tensor-support.md) detailing the overview, supported formats, usage examples, dataset integration, creation of tensor files, technical details, use cases, and limitations of the tensor input feature.
• Framework Integration and Testing: Integrated tensor support into the dataset preprocessor, template system (special tokens, keys, placeholders), and added extensive unit tests to ensure robust functionality.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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 :thumbsup: and :thumbsdown: 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.

[^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.

[Jintao-Huang]

Hello. Thanks for your PR. Could you tell me what additional advantages using <tensor> has compared to using <image> ?

[Marshall-mk]

Hello. Thanks for your PR. Could you tell me what additional advantages using <tensor> has compared to using <image> ?

The key difference here is that you can't load tensors (.pt files) using the or

You might be wondering why we’d want to load tensors at all. In my case, I work with medical images such as CT or MRI scans, which often contain multiple slices and have large dimensions. These images make training VLMs very compute-intensive. To efficiently process these images, we need to compress them using an encoder, save the resulting embeddings as tensors, and then use the VLMs available in ms-swift.

Now, a naive approach would be to convert the embeddings (the tensors) into videos (from .pt files to .mp4), but that introduces the need for additional normalizations and scaling, which actually leads to some disparity in the actual embeddings. This is why using directly is a better approach, as it avoids the unnecessary conversion and preserves the embeddings.