Fu Yao

PhD student at the University of Edinburgh

[Fu Yao @fuyao386]

A new repo, "flashtensors," claims to be "built from the ground up" and offers 10x faster LLM loading.

Yet, it’s a refactor and rebrand of our work, ServerlessLLM Store (https://t.co/C8C9oyqOzn), based on these facts:

Fact 1: Over 70% of its files were copied from SLLM Store – with all Apache 2.0 license headers removed. @ASF

Fact 2: The claimed 10x performance gains come entirely from our original loader

Fact 3: Half of their "original" contribution is just a CLI wrapper

We’ve archived all evidence at https://t.co/p6qAkbencU. We must be vigilant, as GenAI makes code refactoring—and apparent rebranding—easier than ever.

Since open-sourcing ServerlessLLM, we’ve received incredible support and attention from brilliant minds like @ludovico_bessi @tom_doerr and @hei_nyan , which have in turn supported many amazing projects such as @openEuler to build incredible things.

That’s the true spirit of open source.

[Fu Yao @fuyao386]

Representative diff

[Fu Yao @fuyao386]

A new repo, "flashtensors," claims to be "built from the ground up" and offers 10x faster LLM loading.

Yet, it’s a refactor and rebrand of our work, ServerlessLLM Store (https://t.co/C8C9oyqOzn), based on these facts:

Fact 1: Over 70% of its files were copied from SLLM Store – with all Apache 2.0 license headers removed. @ASF

Fact 2: The claimed 10x performance gains come entirely from our original loader

Fact 3: Half of their "original" contribution is just a CLI wrapper

We’ve archived all evidence at https://t.co/p6qAkbencU. We must be vigilant, as GenAI makes code refactoring—and apparent rebranding—easier than ever.

Since open-sourcing ServerlessLLM, we’ve received incredible support and attention from brilliant minds like @ludovico_bessi @tom_doerr and @hei_nyan , which have in turn supported many amazing projects such as @openEuler to build incredible things.

That’s the true spirit of open source.

[Fu Yao @fuyao386]

@unwind_ai_ We've published automated diff analysis with full reproducibility:
https://t.co/PHUz8yqV1L

Independent verification welcome.

[Fu Yao @fuyao386]

This project's core loader was copied from our academic project, ServerlessLLM [https://t.co/C8C9oyqOzn], with all original credits removed from source files.

This is a clear violation of the Apache 2.0 license. We've opened a public issue to have this corrected:
**https://t.co/oypNzJiyRW**

We appreciate the attention to faster model loading -- ServerlessLLM is always open for contributions! ✨

[Fu Yao @fuyao386]

🚀Doing research in LLM inference systems? Exciting news: ServerlessLLM is now open-sourced on GitHub!

🌟Try experimenting with scheduling, autoscaling, KV-cache management, etc. Join us and help shape the future of LLM! 🔧💡

https://t.co/bQpKw7bN3Y Edinburgh Uni., NTU.

[Fu Yao @fuyao386]

Had an incredible time with Dmitrii in Beijing!

[Dmitrii Ustiugov @DmitriiUstiugov]

ServerlessLLM reunion in Beijing. With the leading author Yao Fu from Uni of Edinburgh and Microsoft Research Asia

[PyTorch @PyTorch]

Welcome to the PyTorch Ecosystem, TorchOpt! 🎉

TorchOpt is revolutionizing differentiable optimization with its unified programming abstraction and powerful differentiation modes.

Check it out: https://t.co/LyHQAeq80U