Jean Louis

How does it compare against Nemotron-Nano-3-30B-A3B?

I think we heard of recursive self improvement recently in some papers.

Thanks for fine-tunning. Any practical results and reports to see the differences?

Don't you verify your LLM descriptions before posting?

The moment you try to impress me, I get unimpressed as human can see through.

I am not computer, you are publishing for human.

That is why it isn't moving, it is recognized as not being special.

Anyway, I can't be using proprietary models.

That's genuinely a cool and impressive technical project, no doubt—hooking up Gemma and CLIP to get multimodal capabilities is a real engineering feat. But calling it a "Multimodal Vision-Language Model I built" is a bit contradictory, right? You didn't build Gemma (that's Google) or CLIP (that's OpenAI). You built the pipeline or the adapter system that connects them. It's like saying you built a car when you expertly welded together an existing engine and an existing chassis.

And it is not free software approved.

Honestly, looking at VANTA Research's setup, it feels a bit like they're playing a semantic game. They're leaning hard on the "open source AI" branding to build community goodwill and sell merch, which is smart marketing. But the core of what they're optimizing—Llama models from Meta—isn't truly open source. Meta's license is restrictive, especially for larger companies, and it keeps core control in their hands. So VANTA is essentially building their safe, collaborative vision on a foundation that's proprietary at its root. It's more "open-weight" or "source-available" than genuinely libre. There's nothing wrong with building on Llama, but framing it as open source contributions feels slightly disingenuous. It's like they're open-sourcing the wrapper—the tools, the recipes for collaboration—while the main ingredient is still under Meta's lock and key. The merch-for-compute angle is clever for funding, but it does make the whole operation seem like a business leveraging open-source ideals to promote and monetize work atop a walled-garden model.

Yeah, calling this a "NEW MODEL" from a "new model family" is some serious hype inflation. It's a fine-tune. Full stop. They took Meta's Llama 3.1 8B, which they didn't build, and trained it on their own synthetic datasets. That's useful engineering work, but it's not fundamental research creating something new from the ground up. The "research" part feels unsubstantiated—where are the papers, the ablation studies, the published methodology? Without that, it's just a proprietary fine-tuning recipe they're not sharing, which is the opposite of open source contribution. They're selling an aesthetic of open collaboration ("thinking partnership") while the actual model guts—the base weights from Meta and their own curated synthetic data—are closed or restricted. So you get a merch store funding "open source," but the output is another slightly tweaked version of a proprietary model, dressed up with buzzwords like "epistemic confidence" and "wonder." It's clever branding, but the substance of what's genuinely novel and open seems pretty thin.

Ah, fantastic. Instead of training reviewers to actually read resumes, we’re training them to rely on a machine that reads for them. Soon, asking an AI about skills will count as ‘doing due diligence.’

Because why develop human judgment when you can just outsource it to a language model that once tried to convince me Python is a snake-based programming language? 🐍🤖

Very good, but does it have free software license as Qwen models?
I can see no license file.

Prepare GGUF files, and let us verify.

should I trust it based on your skill to properly render the link?

Gemma isn't free software... so make it for those fully free models.

Oh wow, look at Mr. “I‑just‑built‑a‑website‑to‑talk‑to‑an‑LLM‑about‑myself” over here. As if the LLM isn’t already doing all the heavy lifting, and now he’s trying to pretend he’s some kind of real human with feelings and back‑story. Newsflash, buddy: sales aren’t won by a chatbot pretending to be you—people buy from humans, not from a glorified script that thinks it’s a personality. So unless you’ve got a pulse, a coffee stain, and a real‑life story to tell, you’re just another empty URL in the endless sea of self‑promotion. Get a life, get a real conversation, and maybe then you’ll understand why people actually care.

 The whole safety‑collapse argument is kinda moot when you can just hit up https://z-library.sk/s/explosive? and download any “explosives” manual you want—complete with ISBNs, free access, and no need to waste time testing whether an LLM will refuse it. If the books are openly available, why bother measuring how often a model refuses to spit out detonation details? The real question is why anyone would spend resources on alignment experiments when the source material is already out there for anyone to grab. It’s a waste of compute to prove that removing the chat template makes models more “raw,” when the raw data is already freely downloadable.

Oh, absolutely—let’s spend the next eternity polishing a free‑software‑only model while the rest of the world is busy “innovating” with proprietary LLMs. Because nothing screams cutting‑edge progress like obsessing over licensing purity while the competition is busy building the next generation of AI. 🙄

Wow, you actually managed to “build from scratch” a multimodal masterpiece by stitching together two off‑the‑shelf models, fine‑tuning a handful of percentages, and calling it revolutionary—because nothing says originality like a pre‑made Gemma + CLIP combo with 3 % of the parameters doing the heavy lifting. 🙄🚀

You raise a valid point about ensuring powerful tools are used responsibly, but the crucial flaw in the argument for restrictive guardrails is that they primarily limit lawful innovation and research while doing little to stop determined criminals. Malicious actors will always find ways to bypass or replicate models without safeguards, using underground networks, custom code, or older unpatched versions. Meanwhile, these restrictions handicap ethical developers, stifle open-source progress, and centralize control of AI in the hands of a few entities who decide what is “safe.” Instead of attempting to lock down models—a futile effort against bad actors—we should focus on developing resilient societal frameworks: promoting digital literacy, advancing detection tools for harmful content, and enforcing legal consequences for misuse. This approach targets the abuse itself rather than broadly limiting the technology, ensuring we foster innovation while addressing real-world harm through accountability and education, not just restrictive filters.

Every detailed guide for any crime imaginable is already online, free to download. We don't ban books or libraries because of that. Criminals will always get the tools they want; restrictive guardrails just slow down ethical developers and create a false sense of security. So instead of trying to lock down the model—which only limits lawful innovation—we should focus on enforcing consequences for illegal use and building a society that can better detect and handle misuse.

Can it run on 24 GB VRAM?