Settings

Theme

Llama3V is suspected to have been stolen from the MiniCPM-Llama3-v2.5 project

github.com

30 points by zinccat 2 years ago · 9 comments

Reader

ammo1662 2 years ago

Related

https://www.reddit.com/r/LocalLLaMA/comments/1d6f1f3/llama3v...

https://aksh-garg.medium.com/llama-3v-building-an-open-sourc...

  • SushiHippie 2 years ago

    > Edit (June 2)

    > A big thank you to people who pointed out similarities to previous research in the comments. We realized that our architecture is very similar to OpenBMB’s “MiniCPM-Llama3-V 2.5: A GPT-4V Level Multimodal LLM on Your Phone,” who beat us to the implementation. We have taken down our original model in respect to the authors.

    > The link to the original author’s repository can be found here: https://github.com/OpenBMB/MiniCPM-V/tree/main?tab=readme-ov...

    > — Aksh Garg, Sid Sharma

allenleee 2 years ago

Related (Chinese)

https://mp.weixin.qq.com/s/_cs2M2Nm1HamkRfEY2F1xg

smarm52 2 years ago

Is there a nice outoftheloop summary out there somewhere?

I don't follow these projects closely. =/

  • TreeDiagramIRC 2 years ago
    • smarm52 2 years ago

      Oh okay, so these are two different academic projects.

      > If the research team from Stanford University is proven to have plagiarized this MiniCPM-V project from Tsinghua University, they should feel ashamed, and also, MiniCPM-V project deserve an apology and acknowledgment.

      I was thinking this might have been a commercial organization stealing from an open source project or something.

      > This is the strongest evidence that llama3-V does not train its model at all, but adds random Gaussian noise to the model parameters of miniCPM-llama3-v2.5

      And not just "built on without attribution", but a literal copy between the products with minimal changes.

zinccatOP 2 years ago

previous post of llama3v on hackernews: https://news.ycombinator.com/item?id=40504827

Keyboard Shortcuts

j
Next item
k
Previous item
o / Enter
Open selected item
?
Show this help
Esc
Close modal / clear selection