So many here are trashing on Ollama, saying it's "just" nice porcelain around llama.cpp and it's not doing anything complicated. Okay. Let's stipulate that.
So where's the non-sketchy, non-for-profit equivalent? Where's the nice frontend for llama.cpp that makes it trivial for anyone who wants to play around with local LLMs without having to know much about their internals? If Ollama isn't doing anything difficult, why isn't llama.cpp as easy to use?
Making local LLMs accessible to the masses is an essential job right now—it's important to normalize owning your data as much as it can be normalized. For all of its faults, Ollama does that, and it does it far better than any alternative. Maybe wait to trash it for being "just" a wrapper until someone actually creates a viable alternative.
I totally agree with this. I wanted to make it really easy for non-technical users with an app that hid all the complexities. I basically just wanted to embed the engine without making users open their terminal, let alone make them configure. I started with llama.cpp amd almost gave up on the idea before I stumbled upon Ollama, which made the app happen[1]
There are many flaws in Ollama but it makes many things much easier esp. if you don’t want to bother building and configuring. They do take a long time to merge any PRs though. One of my PRs has been waiting for 8 months and there was this another PR about KV cache quantization that took them 6 months to merge.
> They do take a long time to merge any PRs though.
I guess you have a point there, seeing as after many months of waiting we finally have a comment on this PR from someone with real involvement in Ollama - see https://github.com/ollama/ollama/pull/5059#issuecomment-2628... . Of course this is very welcome news.
It's not really welcome news, he is just saying they're putting it on the long finger because they think other stuff is more important. He's the same guy that kept ignoring the KV cache quant merge.
And the actual patch is tiny..
I think it's about time for a bleeding-edge fork of ollama. These guys are too static and that is not what AI development is all about.
He specifically says that they're reworking the Ollama server implementation in order to better support other kinds of models, and that such work has priority and is going to be a roadblock for this patch. This is not even news to those who were following the project, and it seems reasonable in many ways - users will want Vulkan to work across the board if it's made available at all, not for it to be limited to the kinds of models that exist today.
>So where's the non-sketchy, non-for-profit equivalent
llama.cpp, kobold.cpp, oobabooga, llmstudio, etc. There are dozens at this point.
And while many chalk the attachment to ollama up to a "skill issue", that's just venting frustration that all something has to do to win the popularity contest is to repackage and market it as an "app".
I prefer first-party tools, I'm comfortable managing a build environment and calling models using pytorch, and ollama doesn't really cover my use cases, so I'm not it's audience. I still recommend it to people who might want the training wheels while they figure out how not-scary local inference actually is.
I would argue that kobold.cpp is even easier to use than Ollama. You click on the link in the README to download an .exe and doubleclick it and select your model file. No command line involved.
Which part of the user experience did you have problems with when using it?
You’re coming at it from a point of knowledge. Read the first sentence of the Ollama website against the first paragraph of kobold’s GitHub. Newcomers don’t have a clue what “running a GGUF model..” means. It’s written by tech folk without an understanding of the audience.
Ollama is also written for technical/developer users, by accident (it seems), even though they don't want it to be strictly for technical users. I've opened a issue asking them to make it more clear that Ollama is for technical users, but they seem confident people with no terminal experience can and will also use Ollama: https://github.com/ollama/ollama/issues/7116
Why do you care? They're the ones who will deal with the support burden of people who don't understand how to use it—if that support burden is low enough that they're happy with where they're at, what motivation do you have to tell them to deliberately restrict their audience?
Like many in FOSS I care about making the experience better for everyone. Slightly weird question, why do you care that I care?
> what motivation do you have to tell them to deliberately restrict their audience?
I don't have any motivation to say any such thing, and I wouldn't either. Is that really your take away from reading that issue?
Stating something like "Ollama is a daemon/cli for running LLMs in your terminal" on your website isn't a restriction whatsoever, it's just being clear up front what the tool is. Currently, the website literally doesn't say what Ollama actually is.
> Is that really your take away from reading that issue?
Yes. You went to them with a definition of who they're trying to serve and they wrote back that they didn't agree with your relatively narrow scope. Now you're out in random threads about Ollama complaining that they didn't like your definition of their target audience.
Yes, I'm not asking them to adopt what I think is the target audience, I'm asking them to define any target audience, then add at least one sentence on their website describing what Ollama is, not sure why that's controversial.
Basically the only information on the website right now is "Get up and running with large language models.", do you think that's helping people? Could mean anything.
It’s so hard to decipher the complaints about ollama in this comment section. I keep reading comments from people saying they don’t trust it, but then they don’t explain why they don’t trust it and don’t answer any follow up questions.
As someone who doesn’t follow this space, it’s hard to tell if there’s actually something sketchy going on with ollama or if it’s the usual reactionary negativity that happens when a tool comes along and makes someone’s niche hobby easier and more accessible to a broad audience.
>So where's the non-sketchy, non-for-profit equivalent?
Serving models is currently expensive. I'd argue that some big cloud providers have conspired to make egress bandwidth expensive.
That, coupled with the increasing scale of the internet, make it harder and harder for smaller groups to do these kinds of things. At least until we get some good content addressed distributed storage system.
As has been pointed out in this thread in a comment that you replied to (so I know you saw it) [0], Ollama goes to a lot of contortions to support multiple llama.cpp backends. Yes, their solution is a bit of a hack, but it means that the effort to adding a new back end is substantial.
And again, they're doing those contortions to make it easy for people. Making it easy involves trade-offs.
Yes, Ollama has flaws. They could communicate better about why they're ignoring PRs. All I'm saying is let's not pretend they're not doing anything complicated or difficult when no one has been able to recreate what they're doing.
This is incorrect. The effort it took to enable Vulkan was relatively minor. The PR is short and to be honest it doesn't do much, because it doesn't need to.
that PR doesn't actually work though -- it finds the Vulkan libraries and has some memory accounting logic, but the bits to actually build a Vulkan llama.cpp runner are not there. I'm not sure why its author deems it ready for inclusion.
(I mean, the missing work should not be much, but it still has to be done)
This is a change from your response to the comment that I linked to, where you said it was a good point. Why the difference?
Maybe I should clarify that I'm not saying that the effort to enable a new backend is substantial, I'm saying that my understanding of that comment (the one you acknowledged made a good argument) is that the maintenance burden of having a new backend is substantial.
Okay, now we're playing semantics. "Reasonable argument" were your words. What changed between then and now to where the same argument is now "incorrect"?
But you never said why, and you never said it was incorrect, you said it was a reasonable argument and then appealed to the popularity of the PR as the reason why you disagree.
But now suddenly what I said is not just an argument you disagree with but is also incorrect. I've been genuinely asking for several turns of conversation at this point why what I said is incorrect.
Why is it incorrect that the maintenance burden of maintaining a Vulkan backend would be a sufficient explanation for why they don't want to merge it without having to appeal to some sort of conspiracy with Nvidia?
llama.cpp already supports Vulkan. This is where all the hard work is at. Ollama hardly does anything on top of it to support Vulkan. You just check if the libraries are available, and get the available VRAM. That is all. It is very simple.
Llamafile is great but solves a slightly different problem very well: how do I easily download and run a single model without having any infrastructure in place first?
Ollama solves the problem of how I run many models without having to deal with many instances of infrastructure.
I think you are missing the point. To get things straight: llama.cpp is not hard to setup and get running. It was a bit of a hassle in 2023 but even then it was not catastrophically complicated if you were willing to read the errors you were getting. People are dissatisfied for two, very valid reasons: ollama gives little to no credit to llama.cpp. The second one is the point of the post: a PR has been open for over 6 months and not a huge PR at that has been completely ignored. Perhaps the ollama maintainers personally don't have use for it so they shrugged it off but this is the equivalent of "it works on my computer". Imagine if all kernel devs used Intel CPUs and ignored every non-intel CPU-related PR. I am not saying that the kernel mailing list is not a large scale version of a countryside pub on a Friday night - it is. But the maintainers do acknowledge the efforts of people making PRs and do a decent job at addressing them. While small, the PR here is not trivial and should have been, at the very least, discussed. Yes, the workstation/server I use for running models uses two Nvidia GPU's. But my desktop computer uses an Intel Arc and in some scenarios, hypothetically, this pr might have been useful.
> To get things straight: llama.cpp is not hard to setup and get running. It was a bit of a hassle in 2023 but even then it was not catastrophically complicated if you were willing to read the errors you were getting.
It's made a lot of progress in that the README [0] now at least has instructions for how to download pre-built releases or docker images, but that requires actually reading the section entitled "Building the Project" to realize that it provides more than just building instructions. That is not accessible to the masses, and it's hard for me to not see that placement and prioritization as an intentional choice to be inaccessible (which is a perfectly valid choice for them!)
And that's aside from the fact that Ollama provides a ton of convenience features that are simply missing, starting with the fact that it looks like with llama.cpp I still have to pick a model at startup time, which means switching models requires SSHing into my server and restarting it.
None of this is meant to disparage llama.cpp: what they're doing is great and they have chosen to not prioritize user convenience as their primary goal. That's a perfectly valid choice. And I'm also not defending Ollama's lack of acknowledgment. I'm responding to a very specific set of ideas that have been prevalent in this thread: that not only does Ollama not give credit, they're not even really doing very much "real work". To me that is patently nonsense—the last mile to package something in a way that is user friendly is often at least as much work, it's just not the kind of work that hackers who hang out on forums like this appreciate.
llama.ccp is hard to set up - I develop software for a living and it wasn’t trivial for me. ollama I can give to my non-technical family members and they know how to use it.
As for not merging the PR - why are you entitled to have a PR merged? This attitude of entitlement around contributions is very disheartening as oss maintainer - it’s usually more work to review/merge/maintain a feature etc than to open a PR. Also no one is entitled to comments / discussion or literally one second of my time as an OSS maintainer. This is imo the cancer that is eating open source.
> As for not merging the PR - why are you entitled to have a PR merged?
I didn’t get entitlement vibes from the comment; I think the author believes the PR could have wide benefit, and believes that others support his position, thus the post to HN.
I don’t mean to be preach-y; I’m learning to interpret others by using a kinder mental model of society. Wish me luck!
llama.cpp has supported vulkan for more than a year now. For more than 6 months now there has been an open PR to add vulkan backend support for Ollama. However, Ollama team has not even looked at it or commented on it.
Vulkan backends are existential for running LLMs on consumer hardware (iGPUs especially). It's sad to see Ollama miss this opportunity.
This is great, I did not know about RamaLama and I'll be using and recommending that in future and if I see people using Ollama in instructions I'll recommend they move to RamaLama in the future. Cheers.
This is fascinating. I’ve been using ollama with no knowledge of this because it just works without a ton of knobs I don’t feel like spending the time to mess with.
As usual, the real work seems to be appropriated by people who do the last little bit — put an acceptable user experience and some polish on it — and they take all the money and credit.
It’s shitty but it also happens because the vast majority of devs, especially in the FOSS world, do not understand or appreciate user experience. It is bar none the most important thing in the success of most things in computing.
My rule is: every step a user has to do to install or set up something halves adoption. So if 100 people enter and there are two steps, 25 complete the process.
For a long time Apple was the most valuable corporation on Earth on the basis of user experience alone. Apple doesn’t invent much. They polish it, and that’s where like 99% of the value is as far as the market is concerned.
The reason is that computers are very confusing and hard to use. Computer people, which most of us are, don’t see that because it’s second nature to us. But even for computer people you get to the point where you’re busy and don’t have time to nerd out on every single thing you use, so it even matters to computer people in the end.
The problem is that whatever esoteric motivations technical people have to join the FOSS movement (scratching an itch, seeking fame, saving the world, doing what everybody else is doing etc.), does not translate well to the domain of designing user experiences. People with the education and talent to have an impact here have neither incentives nor practical means to "FOSS-it". You could Creative Commons some artwork (and there are beautiful examples) but thats about it. The art and science of making software usable thus remains a proprietary pursuit. Indeed if that bottleneck could somehow be relaxed, adoption of FOSS software would skyrocket because the technical core is so good and keeps getting better.
Yeah, I would love an actual alternative to Ollama, but RamaLama is not it unfortunately. As the other commenter said, onboarding is important. I just want one operation install and it needs to work and the simple fact RamaLama is written in Python, assures it will never be that easy, and this is even more true with LLM stuff when using AMD gpu.
I know there will be people that disagree with this, that's ok. This is my personal experience with Python in general, and 10x worse when I need to figure out all compatible packages with specifc ROCm support for my GPU. This is madness, even C and C++ setup and build is easier than this Python hell.
RamaLama's use of Python is different: it appears to just be using Python for scripting its container management. It doesn't need ROCm to work with Python or anything else. It has no difficult dependencies or anything else: I just installed it with `uv tool install ramalama` and it worked fine.
I'd agree that Python packaging is generally bad, and that within an LLM context it's a disastrous mess (especially for ROCm), but that doesn't appear to be how RamaLama is using it at all.
@cge you have this right, the main python script has no dependancies, it just uses python3 stdlib stuff. So if you have a python3 executable on your system you are good to go. All the stuff with dependancies runs in a container. On macOS, using no containers works well also, as we basically just install brew llama.cpp
There's really no major python dependancy problems people have been running this on many Linux distros, macOS, etc.
We deliberately don't use python libraries because of the packaging problems.
I gave Ramalama shot today. I'm very impressed. `uvx ramalama run deepseek-r1:1.5b` just works™ for me. And that's saying A LOT, because I'm running Fedora Kinoite (KDE spin of Silverblue) with nothing layered on the ostree. That means no ROCm or extra AMDGPU stuff on the base layer. Prior to this, I was running llamafile in a podman/toolbox container with ROCm installed inside. Looks like the container ramalama is using has that stuff in there and amdgpu_top tells me the gpu is cooking when I run a query.
Side note: `uv` is a new package manager for python that replaces the pips, the virtualenvs and more. It's quite good. https://github.com/astral-sh/uv
One of the main goals of RamaLama at the start was to be easy to install and run for Silverblue and Kinoite users (and funnily enough that machine had an AMD GPU, so we had almost identical setups). I quickly realized contributing to Ollama wasn't possible without being an Ollama employee:
I just realized that ramalama is actually part of the whole Container Tools ecosystem (Podman, Buildah, etc). This is excellent! Thanks for doing this.
Thanks, just yesterday I discovered that Ollama could not use iGPU on my AMD machine, and was going through a long issue for solutions/workarounds (https://github.com/ollama/ollama/issues/2637). Existing instructions are based on Linux, and some people found it utterly surprising that anyone wants to run LLMs on Windows (really?). While I would have no trouble installing Linux and compile from source, I wasn't ready to do that to my main, daily-use computer.
Great to see this.
PS. Have you got feedback on whether this works on Windows? If not, I can try to create a build today.
The PR has been legitimately out-of-date and unmergeable for many months. It was forward-ported a few weeks ago, and is now still awaiting formal review and merging. (To be sure, Vulkan support in Ollama will likely stay experimental for some time even if the existing PR is merged, and many setups will need manual adjustment of the number of GPU layers and such. It's far from 100% foolproof even in the best-case scenario!)
For that matter, some people are still having issues building and running it, as seen from the latest comments on the linked GitHub page. It's not clear that it's even in a fully reviewable state just yet.
this pr was reviewable multiple times, rebased multiple times. all because ollama team kept ignoring it. it has been open for almost 7 months now without a single comment from the ollama folks.
It's gets out of date with conflicts, etc. Because it's ignored, if this was the upstream project of Ollama, llama.cpp the maintainers would have got this merged months ago.
ollama was good initially in that it made LLMs more accessible for non-technical people while everyone was figuring things out.
Lately they seem to be contributing mostly confusion to the conversation.
The #1 model the entire world is talking about is literally mislabeled their side. There is no such thing as R1-1.5b. Quantization without telling users also confuses noobs as to what is possible. Setting up an api different from the thing they're wrapping adds chaos. And claiming each feature added llama.cpp as something "ollama now supports" is exceedingly questionable especially when combined with the very sparse acknowledgement that it's a wrapper at all.
What do you mean there is no such thing as R1-1.5b? DeepSeek released a distilled version based on a 1.5B Qwen model with the full name DeepSeek-R1-Distill-Qwen-1.5B, see chapter 3.2 on page 14 of their research article [0].
That may have been ok if it was just same model at different sizes but they're completely different things here & it's created confusion out of thin air for absolutely no reason other than ollama being careless.
Ollama needs competition. I’m not sure what drives the people that maintain it but some of their actions imply that there are ulterior motives at play that do not have the benefit of their users in mind.
However such projects require a lot of time and effort and it’s not clear if this project can be forked and kept alive.
The most recent one of the top of my head is their horrendous aliasing of DeepSeek R1 on their model hub, misleading users into thinking they are running the full model but really anything but the 671b alias is one of the distilled models. This has already led to lots of people claiming that they are running R1 locally when they are not.
The whole DeepSeek-R1 situation gets extra confusing because:
- The distilled models are also provided by DeepSeek;
- There's also dynamic quants of (non-distilled) R1 - see [0]. Those, as I understand it, are more "real R1" than the distilled models, and you can get as low as ~140GB file size with the 1.58-bit quant.
I actually managed to get the 1.58-bit dynamic quant running on my personal PC, with 32GB RAM, at about 0.11 tokens per second. That is, roughly six tokens per minute. That was with llama.cpp via LM Studio; using Vulkan for GPU offload (up to 4 layers for my RTX 4070 Ti with 12GB VRAM :/) actually slowed things down relative to running purely on the CPU, but either way, it's too slow to be useful with such specs.
Only if you insist on realtime output: if you're OK with posting your question to the model and letting it run overnight (or, for some shorter questions, over your lunch break) it's great. I believe that this use case can fit local-AI especially well.
I'm not sure that's fair, given that the distilled models are almost as good. Do you really think Deepseek's web interface is giving you access to 671b? They're going to be running distilled models there too.
I’m not sure I understand what this comment is responding to. Wouldn’t a distilled Deepseek still use the same tokenizer? I’m not claiming they are using llama in their backend. I’m just saying they are likely using a lower-parameter model too.
The small models that have been published as part of the DeepSeek release are not a "distilled DeepSeek", they're fine-tuned varieties of Llama and Qwen. DeepSeek may have smaller models internally that are not Llama- or Qwen-based but if so they haven't released them.
Thank you. I’m still learning as I’m sure everyone else is, and that’s a distinction I wasn’t aware of. (I assumed “distilled” meant a compressed parameter size, not necessarily the use of another model in its construction.)
Given that the 671B model is reportedly MoE-based, it definitely could be powering the web interface and API. MoE slashes the per-inference compute cost - and when serving the model for multiple users you only have to host a single copy of the model params in memory, so the bulk doesn't hurt you as much.
LM Studio has been around for a long time and does a lot of similar things but with a more UI-based approach. I used to use it before Ollama, and seems it's still going strong. https://lmstudio.ai/
First I got the feeling because of how they store things on disk and try to get all models rehosted in their own closed library.
Second time I got the feeling is when it's not obvious at all about what their motives are, and that it's a for-profit venture.
Third time is trying to discuss things in their Discord and the moderators there constantly shut down a lot of conversation citing "Misinformation" and rewrites your messages. You can ask a honest question, it gets deleted and you get blocked for a day.
Just today I asked why the R1 models they're shipping that are the distilled ones, doesn't have "distilled" in the name, or even any way of knowing which tag is which model, and got the answer "if you don't like how things are done on Ollama, you can run your own object registry" which doesn't exactly inspire confidence.
Another thing I noticed after a while is that there are bunch of people with zero knowledge of terminals that want to run Ollama, even though Ollama is a project for developers (since you do need to know how to run a terminal). Just making the messaging clearer would help a lot in this regarding, but somehow the Ollama team thinks thats gatekeeping and it's better to teach people basic terminal operations.
For what it's worth, HuggingFace provides documentation on how you can run any GGUF model inside Ollama[0]. You're not locked into their closed library or have to wait for them to add new models.
Granted, they could be a lot more helpful in providing information on how you do this. But this feature exists, at least.
Ollama team's response (verbatim) when asking what they think of the comments about Ollama in this HN submission: "Who cares? It's the internet... everybody has an opinion... and they're usually bad". Not exactly the response you'd expect from people who should ideally learn from what others think (correct or not) about your project.
Ollama doesn't really need competition. Llama.cpp just needs a few usability updates to the gguf format so that you can specify a hugging face repository like you can do in vLLM already.
I totally agree that ollama needs competition. They have been doing very sketchy things lately. I wish llama.cpp had an alternative wrapper client like ollama.
> Letting people download 400GB just to find that out is also .. not optimal.
Letting people download any amount of bytes just to find out they got something else isn't optimal. So what to do? Highlight the differences when you reference them so people understand.
> DeepSeek's first-generation reasoning models are achieving performance comparable to OpenAI's o1 across math, code, and reasoning tasks! Give it a try! 7B distilled: ollama run deepseek-r1:7b
Are really misleading. Reading the first part, you think the second part is that model that gives "performance comparable to OpenAI's o1" but it's not, it's a distilled model with way worse performance. Yes, they do say it's the distilled model, but I hope I'm not alone in seeing how people less careful would confuse the two.
If they're doing this on purpose, I'd leave a very bad taste in my mouth. If they're doing this accidentally, it also gives me reason to pause and re-evaluate what they're doing.
Well, it definitely runs faster on external dGPU's. With iGPU's and possibly future NPU's, the pre-processing/"thinking" phase is much faster (because that one is compute-bound) but text generation tends to be faster on CPU because it makes better use of available memory bandwidth (which is the relevant constraint there). iGPU's and NPU's will still be a win wrt. energy use, however.
I made an argument for performance, not for compatibility.
If you run your local llm in the least performant way possible on tour overly expensive GPU, then you are not making value of your purchase.
Vulkan is a fallback option is all.
I even see people running on their CPU because some apps dont support their hardware and llama.cpp made it even possible. It is still a really bad idea.
I'm willing to bet that Vulkan will outperform OpenVINO.
Vulkan is the API right now in the graphics world. It's very well supported and actively being improved on. Everyone is pouring resources into making Vulkan better.
OpenVINO feels barely developed. Intel never made it a proper backend for Pytorch like AMD did with ROCm. It's hard to see where it is going, or if it is going anywhere at all. Between Sycl and OneApi it's hard to see how much interest Intel has developing it.
> I'm willing to bet that Vulkan will outperform OpenVINO.
> Vulkan is the API right now in the graphics world.
YUP, Vulkan is all the rage in the graphics world, and for good reasons. But we arent discussing graphics now are we?
Vulkan is a general graphics API with some computing capabilities.
OpenVINO is a toolkit for inference neural networks, by intel built to make use of their GPUs and NPUs for this specific task.
Using vulkan, first you need to translate your payload to shaders, then they need to be compiled to SPIR-V, then they can use a subset of the cards capabilities.
How could this even remotely match something written specifically for the task?
Also, it is dead easy to benchmark if you still think otherwise.
Ollama is sketchy enough that I run it in a VM. Which is odd because it would probably take less effort to just run Llama.cpp directly, but VMs are pretty easy so just went that route.
When I see people bring up the sketchiness most of the time the creator responds with the equivalent of shrugs, which imo increases the sketchiness.
Can you please elaborate? How are you running ollama? I just build it from source and have written a shell script to start/stop it. It runs under my local user account (I should probably have its own user) and is of course not exposed outside localhost.
That’s the curse and blessing of open source I guess?
I have billion dollar companies running my oss software without giving me anything - but do I gripe about it in public forums? Yea maybe sometimes but it never helps to improve the situation.
It's the curse of permissively licensed open source. Copyleft is not the answer to everything but against companies leeching and not giving back it is effective.
Well we had this conversation but once you change the license to something like that you lose the trust of the oss community instantly and as elasticsearch etc. show it also doesn’t really help with monetisation
No it’s totally different from this case - but the software is a super important part of their stack - if it were to stop working they can shut down their business
Ollama tries to appeal to a lowest common denominator user base, who does not want to worry about stuff like configuration and quants, or which binary to download.
I think they want their project to be smart enough to just 'figure out what to do' on behalf of the user.
That appeals to a lot of people, but I think them stuffing all backends into one binary and auto-detecting at runtime which to use and is actually a step too far towards simplicity.
What they did to support both CUDA and ROCm using the same binary looked quite cursed last time I checked (because they needed to link or invoke two different builds of llama.cpp of course).
I have only glanced at that PR, but I'm guessing that this plays a role in how many backends they can reasonably try to support.
In nixpkgs it's a huge pain that we configure quite deliberately what we want Ollama to do at build time, and then Ollama runs off and does whatever anyways, and users have to look at log output and performance regressions to know what it's actually doing, every time they update their heuristics for detecting ROCm. It's brittle as hell.
I disagree with this, but it's a reasonable argument. The problem is that the Ollama team has basically ignored the PR, instead of engaging the community. The least they can do is to explain their reasoning.
This PR is #1 on their repo based on multiple metrics (comments, iterations, what have you)
I don't know why one would use Ollama instead of llama.cpp. llama.cpp is so easy to use and the maintainer is pretty famous and active in the community.
I tried using ollama because I couldn't get ROCm working on my system with llama-cpp. Ollama bundles the ROCm libraries for you. I got around 50 tokens per second with that setup.
I tried llama-cpp with the Vulkan backend and doubled the amount of tokens per second. I was under the impression ROCm is superior to Vulkan, so I was confused about the result.
It depends on your GPU. Vulkan is well-supported by essentially all GPUs. AMD support ROCm well for their datacenter GPUs, but support for consumer hardware has not been as good.
Could it be that supporting multiple platforms open up more support tickets and adds more work to keep the software working on those new platforms?
As someone who built apps for Windows, Linux, macOS, iOS and Android, it is not trivial to ensure your new features or updates work on all platforms, and you have to deal with deprecations.
This is not quite correct. Ollama must assess the state of Vulkan support and amount of available memory, then pick the fraction of the model to be hosted on GPU. This is not totally foolproof and will likely always need manual adjustment in some cases.
If it's "just" a friendly front and, why doesn't llama.cpp just drop one themselves? Do they actually care about the situation, or are random people just mad on their behalf?
At the risk of being pedantic (I don't know much about C++ and I'm genuinely curious), if Ollama is really just a wrapper around Llama.cpp, why would it need the Vulkan specific flags?
Shouldn't it just call Llama.cpp and let Llama.cpp handle the flags internally within Llama.cpp? I'm thinking from an abstraction layer perspective.
The Vulkan-specific flags are needed (1) to set up the llama.cpp build options when building Ollama w/ Vulkan support - which apparently is still a challenge with the current PR, if the latest comments on the GitHub page are accurate; also (2) to pick how many model layers should be run on the GPU, depending on available GPU memory. Llama.cpp doesn't do that for you, you have to set that option yourself or just tell it to move "everything", which often fails with an error. (Finding the right amount is actually a trial-and-error process which depends on the model, quantization and also varies depending on how much context you have in the current conversation. If you have too many layers loaded and too little GPU memory, a large context can result in unpredictable breakage.)
This is going to sound like a troll, but it's an honest question: Why do people use Ollama over llama.cpp? llama.cpp has added a ton of features, is about as user-friendly as Ollama, and is higher-performance. Is there some key differentiator for Ollama that I'm missing?
llama.cpp - Read the docs, with loads of information and unclear use cases. Question if it has API compatibility and secondary features that a bunch of tools expect. Decide it's not worth your effort when `ollama` is already running by the time you've read the docs
Additionally, Ollama makes model installation a single command. With llama.cpp, you have to download the raw models from Huggingface and handle storage for them yourself.
I can only speak for myself but to me llama.ccp looks kind of hard to use (tbh never tried to use it), whereas ollama was just one cli command away.
Also I had no idea that its equivalent, I thought llama.ccp is some experimental tool for hardcore llm cracks, not something that I can teach my for example my non-technical mom to use.
Looking at the repo of llama.ccp it’s still not obvious to me how to use it without digging in - I need to download models from huggingface it seems and configure stuff etc - with ollama I type ollama get or something and it works.
Tbh I don’t just that stuff a lot or even seriously, maybe once per month to try out new local models.
I think having an easy to use quickstart would go a long way for llama.ccp - but maybe it’s not intended for casual (stupid?) users like me…
In my mind, it doesn't help that llama.cpp's name is that of a source file. Intuitively, that name screams "library for further integration," not "tool for end-user use."
Basically, if you know how to use a computer, you can use Ollama (almost). You can't say the same thing about llama.cpp. Not everyone knows how to build from source, or even what "build" means.
> - It doesn't have a download page, you have to build it yourself
I'd wager that anyone capable enough to run a command line tool like Ollama should also be able to download prebuilt binaries from the llama.cpp releases page[1]. Also, prebuilt binaries are available on things like homebrew[2].
I am very technically inclined and use Ollama (in a VM, but still) because of all the steps and non-obviousness of how to run Llama.cpp. This framing feels a bit like the “Dropbox won’t succeed because rsync is easy” thinking.
> This framing feels a bit like the “Dropbox won’t succeed because rsync is easy” thinking.
No this isn't. There are plenty of end user GUI apps that make it far easier than Ollama to download and run local LLMs (disclaimer: I build one of them). That's an entirely different market.
IMO, the intersection between the set of people who use a command line tool, and the set of people who are incapable of running `brew install llama.cpp` (or its Windows or Linux equivalents) is exceedingly small.
I can't install any .app on my fairly locked down work computer, but I can `brew install ollama`.
When I read the llama.cpp repo and see I have to build it, vs ollama where I just have to get it, the choice is already made.
I just want something I can quickly run and use with aider or mess around with. When I need to do real work I just use whatever OpenAI model we have running on Azure PTUs
And you still need to find and download the model files yourself, among other steps, which is intimidating enough to drive away most users, including skilled software engineers. Most people just want it to work and start using it for something else as soon as possible.
The same reason I use apt install instead of compiling from source. I can definitely do that, but I don't, because it's just a way to get things installed.
Ok I was looking at the repo from mobile and missed the releases.
Still it's not immediate obvious from README that there is an option to download it. There are instructions on how to build it, but not how to download it. Or maybe I'm blind, please correct me.
I used both. I had a terrible time with llama, and did not realise it until I used ollama.
I owned an RTX2070, and followed the llama instructions to make sure it was compiling with GPU enabled. I then hand-tweaked settings (numgpulayers) to try to make it offload as much as possible to the GPU. I verified that it was using a good chunk of my GPU ram (via nvidia-smi), and confirmed that with-gpu was faster than cpu-only. It was still pretty slow, and influenced my decision to upgrade to an RTX3070. It was faster, but still pretty meh...
The first time I used ollama, everything just worked straight out of the box, with one command and zero configuration. It was lightning fast. Honestly if I'd had ollama earlier, I probably wouldn't have felt the need to upgrade GPU.
Maybe it was lightning fast because the model names are misleading? I installed it to try out deepseek, I was surprised how small the download artifact was and how easily it ran on my simple 3 years old Mac. I was a bit disappointed as deepseek gave bad responses and I heard it should be better than what I used on OpenAI… only to then realize after reading it on Twitter that I got a very small version of deepseek r1.
If it was faster with ollama, then you most probably just downloaded a different model (hard to recognize with ollama). Ollama only adds UX to llama.cpp, and nothing compute-wise.
While not rocketscience, a lot of its features requires to know how to recompile the project with passing certain variables. Also you need to properly format prompts for each instructor model.
ramalama still needs users to be able to install docker first, no? That’s a barrier to entry for many users esp. Windows where I have had my struggles running Docker not to mention a massive resource hog.
It took very long for them to support KV cache quantisation too (which drastically reduces the amount of VRAM needed for context!). Even though the underlying llama.cpp had offered it for ages. And they had it handed to them on a platter, someone had developed everything and submitted a patch.
The developer of that patch even was about to give up as he had to constantly keep it up to date with upstream even though he was constantly being ignored. So he had no idea if it would ever be merged.
They just seem to be really hesitant to offer new features.
Eventually it was merged and it made a huge difference to people with low VRAM cards.
They also decided to rehost the model files in their own (closed) library/repository + store the files split into layers on disk, so you cannot easily reuse model-files between applications. I think the point is that models can share layers, I'm not sure how much space you actually save, I just know that if you use both LM Studio + Ollama you cannot share models but if you use LM Studio + llama.cpp you can share the same files between them, no need to download duplicate model weights.
The way Ollama has basically been laundering llama.cpp’s features as its own felt dodgy, this appears to confirm there’s something underhanded going on.
This is where I want to work. But I feel like an AI swe is more likely to go "down" than an AI company is likely to hire me, a guy who loves optimizing pipelines for parallelism.
I think it's important to bring up the face that llama.cpp has an MIT license[0]. Notably, the MIT license "permits reuse within proprietary software, provided that all copies of the software or its substantial portions include a copy of the terms of the MIT License and also a copyright notice.[1]"
You'll find that Ollama is also distributed under an MIT license[2]. It's fine to disagree with their priorities and lack of transparency. But trying to argue how they use code from other repositories that permit such a thing is tilting at windmills, IMHO.
Well yes, though I was thinking more that they have no clear way to get income besides VCs and need to figure out a way to monetize in some weird way eventually. I would not have predicted them taking Nvidia money to axe AMD compatibility though lol.
can someone please give a quick summary of the criticism towards ollama?
as far as my intel goes it's a mozilla project shouldered mostly by one 10x programmer. i found ollama through hn and last time i didn't notice any lack of trust or suspected sketchiness ... so what changed?
IMO ggerganov is a 10x programmer in the same way Fabrice Bellard is: doing the actual hard infrastructure work that most developers would not be able to do in a reasonable amount of time and at a high performance.
In contrast, the ollama dev team is doing useful work (creating an easy interface) but otherwise mostly piggybacking off the already existing infrastructure
Open-weights LLMs provide a dizzying array of options.
You'd have Llama, Mistral, Gemma, Phi, Yi.
You'd have Llama, Llama 2, Llama 3, Llama 3.2...
And those offer with 8B, 13B or 70B parameters
And you can get it quantised to GGUF, AWQ, exl2...
And quantised to 2, 3, 4, 6 or 8 bits.
And that 4-bit quant is available as Q4_0, Q4_K_S, Q4_K_M...
And on top of that there are a load of fine-tunes that score better on some benchmarks.
Sometimes a model is split into 30 files and you need all 30, other times there's 15 different quants in the same release and you only need a single one. And you have to download from huggingface and put the files in the right place yourself.
ollama takes a lot of that complexity and hides it. You run "ollama run llama3.1" and the selection and download all gets taken care of.
Not sure this is a good analogy. LM Studio is closer to Dropbox as both takes X and makes it easier for users who don't necessarily are very technical. Ollama is a developer-oriented tool (used via terminal + a daemon), so wouldn't compare it to what Dropbox is/did for file syncing.
With ollama I type brew install ollama and then ollama get something, and I have it already running.
With llama.ccp it’s seems i have to build it first, then manually download models somewhere - this is an instant turnoff, i maybe have 5 minutes of my life to waste on this
So where's the non-sketchy, non-for-profit equivalent? Where's the nice frontend for llama.cpp that makes it trivial for anyone who wants to play around with local LLMs without having to know much about their internals? If Ollama isn't doing anything difficult, why isn't llama.cpp as easy to use?
Making local LLMs accessible to the masses is an essential job right now—it's important to normalize owning your data as much as it can be normalized. For all of its faults, Ollama does that, and it does it far better than any alternative. Maybe wait to trash it for being "just" a wrapper until someone actually creates a viable alternative.