Hacker Remix

Ollama violating llama.cpp license for over a year

174 points by Jabrov 13 hours ago | 67 comments

rlpb 12 hours ago

I don't see how this claimed issue is valid.

https://github.com/ollama/ollama/blob/main/llama/llama.cpp/L... says:

"The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software."

The issue submitter claims:

"The terms of the MIT license require that it distribute the copyright notice in both source and binary form."

But: a) that doesn't seem to be in the license text as far I can see; b) I see no evidence that upstream arranged to ship any notice in their binaries, so I don't see how it's reasonable to expect downstreams to do it; and c) in the distribution world (Debian, etc) that takes great care about license compliance, patching upstreams to include copyright notices in binaries isn't a thing. It's not the norm, and this is considered acceptable in our ecosystem.

Maybe I'm missing something, but the issue linked does not make the case that there's anything unacceptable going on here.

Tomte 10 hours ago

The need to extract license and copyright information for binary distribution is universally accepted among Open Source license compliance practitioners and lawyers.

There is a whole industry of tools around it (Fossid, Fossa, BlackDuck, Snyk), as well as Open Source projects ( FOSSology, scancode, oss-review-toolkit).

Re: Debian, they have copyright files in their packaged that are manually curated by Debian Developers and should include all those license texts and copyright notices.

rlpb 3 hours ago

Ah - we're talking about different things.

I was concerned about the implication (or so I thought) that a binary executable should provide the required documentation (eg. via --version or similar). You are thinking about the text being included as part of a binary redistribution. That did not occur to me, because to me, GitHub issues refer to sources, not binary redistributions.

But of course GitHub does have a Releases page. If those binary redistributions do not contain the license text, then I accept that's something that Debian does do, and is the norm in our ecosystem.

But as other commenters have said, it's not completely clear that this is actually a violation of the license, since https://github.com/ollama/ollama/releases/tag/v0.7.0 for example bundles both source and binary downloads and the bundle does contain the license text via the source file download. Certainly anyone who downloads the binary from the maintainer via GitHub does have the required notice made available to them.

fn-mote 12 hours ago

I won't address the rest, but:

> b) I see no evidence that upstream arranged to ship any notice in their binaries, so I don't see how it's reasonable to expect downstreams to do it

Downstream is not in compliance. The fact that upstream has made that compliance hard/impossible is not relevant to the fact that downstream is infringing.

int_19h 7 hours ago

And it's not hard at all. You just include a text file with a third party software notice that has all the licenses, alongside the binary. All major companies shipping F/OSS in their products somehow manage to do this just fine (I have personally done so for three different products at two different companies).

mrguyorama 5 hours ago

It's so normal and common that your car's infotainment screen has a page for it, and it causes the guy who built a useful open source project to get hate mail, because his email address is listed there.

slavik81 3 hours ago

> in the distribution world (Debian, etc) that takes great care about license compliance, patching upstreams to include copyright notices in binaries isn't a thing

On Debian, you will find the llama.cpp copyright notice in /usr/share/doc/llama.cpp/copyright if you have installed the llama.cpp binary package.

pama 12 hours ago

You cannot argue successfully in court that the copy of the binary compiled code is not a copy of a substantial portion of the software. The fix is very trivial. This should not be an open issue.

levifig 10 hours ago

FWIW, llama.cpp links to and fetches models from ollama (https://github.com/ggml-org/llama.cpp/blob/master/tools/run/...).

This issue seems to be the typical case of someone being bothered for someone else, because it implies there's no "recognition of source material" when there's quite a bit of symbiosis between the projects.

diggan 9 hours ago

Well, llama.cpp supports fetching models from a bunch of different sources according to that file, Hugging Face, ModelScope, Ollama, any HTTP/local source. Seems fair to say they've added support for any source one most likely will find the LLM model you're looking for at.

Not sure I'd say there is "symbiosis" between ModelScope and llama.cpp just because you could download models from there via llama.cpp, just like you wouldn't say there is symbiosis between LM Studio and Hugging Face, or even more fun example: YouTube <> youtube-dl/yt-dlp.

gopher_space 6 hours ago

Symbiosis states that a relationship exists. Subcategories of symbiosis state how useful that relationship is to either party, and they're determined by the observer rather than the organisms involved.

ActionHank 10 hours ago

Yes and no, the problem with not expecting that a prominent project follow the rules is that it makes it easier and more likely that no one will follow the rules.

Broken window theory.

cwmoore 2 hours ago

Police broke my window. No theory needed.

int_19h 7 hours ago

The fact that Ollama has been downplaying their reliance on llama.cpp has been known in the local LLM community for a long time now. Describing the situation as "symbiosis" is very misleading IMO.

moralestapia 8 hours ago

>FWIW

It's not worth much. That is a compeltely different thing.

What you mention equates to downloading a file from the web.

Ollama using code from llama.cpp without complying with the license terms is illegal.

Havoc 12 hours ago

I'm continually puzzled by their approach - it's such self inflicted negative PR.

Building on llama is perfectly valid and they're adding value on ease of use here. Just give the llama team appropriately prominent and clearly worded credit for their contributions and call it a day.

nimbius 11 hours ago

i dont find it puzzling at all. the website is basically a blank canvas. contact information is nonexistent.

ignoring the issue is just another way of saying "catch me if you can." and even then open source lawsuits are rather toothless anyway, so the company clearly expects there to be zero consequence.

Maxious 10 hours ago

Ollama is YC21 https://www.ycombinator.com/companies/ollama and was founded by the engineers of what became Docker Desktop

You know, the tool that very famously had a massive rug pull once it gained marketshare https://www.servethehome.com/docker-abruptly-starts-charging...

KronisLV 8 hours ago

> First here, we understand that Docker needs to generate revenue. Creating a foundational technology and not having revenue to grow the business is hard. At the same time, the notice period is what one may consider short.

If the money was starting to run dry, with everyone using the tech (and Docker Hub in particular) but not really giving them any money for it, then something was bound to change.

It's cool that there are other alternatives to Docker Hub though and projects like Podman. I feel like with a bigger grace period, the Docker pricing changes wouldn't have been a big deal.

Koshima 10 hours ago

I think it’s fair to push for clear attribution in these cases, but it’s also important to remember that the MIT license is intentionally permissive. It was designed to make sharing code easy without too many hoops. If Ollama is genuinely trying to be part of the open-source community, a little transparency and acknowledgment can avoid a lot of bad blood.

Etheryte 10 hours ago

MIT is permissive given you follow the license. You can't just copy the code and omit the license and copyright, that's not fine, even if many people like to pretend that it is.

montebicyclelo 9 hours ago

It's permissive; but that doesn't imply not crediting people for their work