TenStrip

/

Minimax-h3_Singularity-Lora

Like 29

You're under the Minimax Community license. You have no rights to do anything.

Jesus be more karen

Subject: Unauthorized Model Extraction, Copyright Violation, and Malicious Misrepresentation

Target Repository: TenStrip/Minimax-h3_Singularity-Lora
Original Model Repository: WarmBloodAban/Minimax-h3_Singularity

Detailed Explanation:

I am the creator and maintainer of the original model WarmBloodAban/Minimax-h3_Singularity. I am filing this report regarding unauthorized redistribution and malicious misrepresentation by user TenStrip.

1. Unauthorized Model Extraction and Redistribution:
   The user TenStrip extracted a standalone LoRA from my integrated model (WarmBloodAban/Minimax-h3_Singularity) via difference extraction/demerging without my permission or consent, and publicly re-uploaded it as TenStrip/Minimax-h3_Singularity-Lora.

2. Malicious Misrepresentation and Unfair Technical Attacks:
   Extracting a crude LoRA from a fine-tuned integrated architecture naturally alters or degrades the intended model weights and structural synergy. User TenStrip is actively using this illegally extracted, flawed LoRA in their discussions to undermine my original work, publicly claiming:

"Tbh this isn't really anything. This led me to question what is in Singularity at all..."

"My verdict is: nothing really. A slight few day training run that moves the final_layer of the model a bit..."

Using an unauthorized, reverse-engineered artifact to make false, defamatory technical conclusions about the original work misleads the community and damages my reputation as a creator.

Requested Action:
I kindly request the Hugging Face moderation team to take down the unauthorized repository TenStrip/Minimax-h3_Singularity-Lora and review the user's comments for moderation violations.

Thank you for maintaining a respectful and lawful open-source AI community.

Real galaxy brained report. Your repo is tagged Apache-2.0. Apache 2.0 gives everyone the right to make derivative works and redistribute them, no permission needed, and you can't revoke it. So by your own license tag the extraction is allowed. And if you want to say the tag doesn't count because it's an H3 derivative, fine, then it's under the MiniMax H3 Community License, and Section III says you can't put your own extra terms on the use, reproduction or distribution of your modifications. There's no "ask my permission first" clause in either one. VI.1 says you own your derivative as far as MiniMax is concerned. It doesn't say you get a veto over everyone downstream.

Also "defamatory"? Both repos are public, anyone can compare them and make up their own mind. Somebody not being impressed by your model isn't defamation. And the guy's model card literally lists the extraction's flaws, tells people to run it at 0.5 or lower, and says he'll delete it the second you release an official LoRA. Real malicious stuff.

neither repo has a copy of the license or the NOTICE file that Section III requires from every redistributor, and that includes yours WarmBloodAban. Plus you've got H3 weights tagged Apache-2.0, which drops every restriction the H3 license makes you pass along. So you're reporting someone for violating a license you're not following yourself. That's just hypocrisy. Add the two files before asking HF to enforce a "lawful open-source community" on anybody else.

And be real about what this report is. Half of it is asking mods to punish a guy for saying he wasn't impressed by your model. If negative reviews count as "moderation violations", nobody can say anything about any model except "omg amazing".

Subject: Unauthorized Model Extraction, Copyright Violation, and Malicious Misrepresentation

Target Repository: TenStrip/Minimax-h3_Singularity-Lora
Original Model Repository: WarmBloodAban/Minimax-h3_Singularity

Detailed Explanation:

I am the creator and maintainer of the original model WarmBloodAban/Minimax-h3_Singularity. I am filing this report regarding unauthorized redistribution and malicious misrepresentation by user TenStrip.

1. Unauthorized Model Extraction and Redistribution:
   The user TenStrip extracted a standalone LoRA from my integrated model (WarmBloodAban/Minimax-h3_Singularity) via difference extraction/demerging without my permission or consent, and publicly re-uploaded it as TenStrip/Minimax-h3_Singularity-Lora.

2. Malicious Misrepresentation and Unfair Technical Attacks:
   Extracting a crude LoRA from a fine-tuned integrated architecture naturally alters or degrades the intended model weights and structural synergy. User TenStrip is actively using this illegally extracted, flawed LoRA in their discussions to undermine my original work, publicly claiming:

"Tbh this isn't really anything. This led me to question what is in Singularity at all..."

"My verdict is: nothing really. A slight few day training run that moves the final_layer of the model a bit..."

Using an unauthorized, reverse-engineered artifact to make false, defamatory technical conclusions about the original work misleads the community and damages my reputation as a creator.

Requested Action:
I kindly request the Hugging Face moderation team to take down the unauthorized repository TenStrip/Minimax-h3_Singularity-Lora and review the user's comments for moderation violations.

Thank you for maintaining a respectful and lawful open-source AI community.

@WarmBloodAban
The LORA is barely functional and is only related to your project with it's name, not it's quality. The Lora with base REF2VA BF16 pruned produces much worse results than base REF2VA BF16 pruned model itself, at least for me.

The Lora extraction isn't actually functional, idk what hybrid base was used and it seems to pull off a lot of difference that is caught up in it. Maybe they should just add the bf16 version like it's been promised for days, or better yet, what this should've been and much more useful: an actual Lora. It will take weeks and thousands of dollars to actually effectively finetune H3 so you can stop pretending.

The Lora extraction isn't actually functional, idk what hybrid base was used and it seems to pull off a lot of difference that is caught up in it. Maybe they should just add the bf16 version like it's been promised for days, or better yet, what this should've been and much more useful: an actual Lora. It will take weeks and thousands of dollars to actually effectively finetune H3 so you can stop pretending.

Hi TenStrip, I just wanted to say that I really appreciate your 10 Eros model. The results and overall style are genuinely impressive, and it is clear a lot of thoughtful work went into it.
If possible, would you consider sharing a technical tutorial or write-up about how you optimized the H3 model? I would be very interested in learning more about your workflow, training approach, dataset preparation, parameter choices, and any practical lessons you found important.
Thank you again for your work and for sharing it with the community.