Ollama is great but I feel like Georgi Gerganov deserves way more credit for llama.cpp.
He (almost) single-handedly brought LLMs to the masses.
With the latest news of some AI engineers' compensation reaching up to a billion dollars, feels a bit unfair that Georgi is not getting a much larger slice of the pie.
`ggerganov` is one of the most under-rated and under-appreciated hackers maybe ever. His name belongs next to like Carmack and other people who made a new thing happen on PCs. And don't forget the shout out to `TheBloke` who like single-handedly bootstrapped the GGUF ecosystem of useful model quants (I think he had a grant from pmarca or something like that, so props to that too).
Nice release. Part of the problem right now with OSS models (at least for enterprise users) is the diversity of offerings in terms of:
- Speed
- Cost
- Reliability
- Feature Parity (eg: context caching)
- Performance (What quant level is being used...really?)
- Host region/data privacy guarantees
- LTS
And that's not even including the decision of what model you want to use!
Realistically if you want to use an OSS model instead of the big 3, you're faced with evalutating models/providers across all these axes, which can require a fair amount of expertise to discern. You may even have to write your own custom evaluations. Meanwhile Anthropic/OAI/Google "just work" and you get what it says on the tin, to the best of their ability. Even if they're more expensive (and they're not that much more expensive), you are basically paying for the priviledge of "we'll handle everything for you".
I think until providers start standardizing OSS offerings, we're going to continue to exist in this in-between world where OSS models theoretically are at performance parity with closed source, but in practice aren't really even in the running for serious large scale deployments.
Why does everything AI-related have to be $20? Why can't there be tiers? OpenAI setting the standard of $20/m for every AI application is one of the worst things to ever happen.
Watching ollama pivot from a somewhat scrappy yet amazingly important and well designed open source project to a regular "for-profit company" is going to be sad.
Thankfully, this may just leave more room for other open source local inference engines.
I don't blame them. As soon as they offer a few more models available with the Turbo mode I plan on subscribing to their Turbo plan for a couple of months - a buying them a coffee, or keeping the lights on kind of thing.
The Ollama app using the signed-in-only web search tool is really pretty good.
I see a lot of hate for ollama doing this kind of thing but also they remain one of the easiest to use solutions for developing and testing against a model locally.
Sure, llama.cpp is the real thing, ollama is a wrapper... I would never want to use something like ollama in a production setting. But if I want to quickly get someone less technical up to speed to develop an LLM-enabled system and run qwen or w/e locally, well then its pretty nice that they have a GUI and a .dmg to install.
It is weird but when I tried new gpt-oss:b20 model locally llama.cpp just failed instantly for me. At the same time under ollama it worked (very slow but anyway). I didn't find how to deal with llama.cpp but ollama definitely doing something under the hood to make models work.
Any more information on "Privacy first"? It seems pretty thin if just not retaining data.
For Draw Things provided "Cloud Compute", we don't retain any data too (everything is done in RAM per request). But that is still unsatisfactory personally. We will soon add "privacy pass" support, but still not to the satisfactory. Transparency log that can be attested on the hardware would be nice (since we run our open-source gRPCServerCLI too), but I just don't know where to start.
If any of the major inference engines - vLLM, Sglang, llama.cpp - incorporated api driven model switching, automatic model unload after idle and automatic CPU layer offloading to avoid OOM it would avoid the need for ollama.
What could be the benefit of paying $20 to Ollama to run inferior models instead of paying the same amount of money to e.g. OpenAI for access to sota models?
It says “usage-based pricing” is coming soon. I think that is the sweet spot for a service like this.
I pay $20 to Anthropic, so I don’t think I’d get enough use out of this for the $20 fee. But being able to spin up any of these models and use as needed (and compare) seems extremely useful to me.
A flat fee service for open-source LLMs is somewhat unique, even if I don't see myself paying for it.
Usage-based pricing would put them in competition with established services like deepinfra.com, novita.ai, and ultimately openrouter.ai. They would go in with more name-recognition, but the established competition is already very competitive on pricing
I am so so so confused as to why Ollama of all companies did this other than an emblematic stab at making money-perhaps to appease someone putting pressure on them to do so. Their stuff does a wonderful job of enabling local for those who want it. So many things to explore there but instead they stand up yet another cloud thing? Love Ollama and hope it stays awesome
For one of the top local open model inference engines of choice - only supporting OSS out of the gate feels like an angle to just ride the hype knowing OSS is announced today "oh OSS came out and you can use Ollama Turbo to use it"
The subscription based pricing is really interesting. Other players offer this but not for API type services. I always imagine that there will be a real pricing war with LLMs with time / as capabilities mature, and going monthly pricing on API services is possibly a symptom of that
What does this mean for the local inference engine? Does Ollama have enough resources to maintain both?
48 comments
[ 2.9 ms ] story [ 56.2 ms ] threadHe (almost) single-handedly brought LLMs to the masses.
With the latest news of some AI engineers' compensation reaching up to a billion dollars, feels a bit unfair that Georgi is not getting a much larger slice of the pie.
- Speed
- Cost
- Reliability
- Feature Parity (eg: context caching)
- Performance (What quant level is being used...really?)
- Host region/data privacy guarantees
- LTS
And that's not even including the decision of what model you want to use!
Realistically if you want to use an OSS model instead of the big 3, you're faced with evalutating models/providers across all these axes, which can require a fair amount of expertise to discern. You may even have to write your own custom evaluations. Meanwhile Anthropic/OAI/Google "just work" and you get what it says on the tin, to the best of their ability. Even if they're more expensive (and they're not that much more expensive), you are basically paying for the priviledge of "we'll handle everything for you".
I think until providers start standardizing OSS offerings, we're going to continue to exist in this in-between world where OSS models theoretically are at performance parity with closed source, but in practice aren't really even in the running for serious large scale deployments.
If I use local/OSS models it's specifically to avoid running in a country with no data protection laws. It's a big close miss here.
Thankfully, this may just leave more room for other open source local inference engines.
The Ollama app using the signed-in-only web search tool is really pretty good.
Sure, llama.cpp is the real thing, ollama is a wrapper... I would never want to use something like ollama in a production setting. But if I want to quickly get someone less technical up to speed to develop an LLM-enabled system and run qwen or w/e locally, well then its pretty nice that they have a GUI and a .dmg to install.
For Draw Things provided "Cloud Compute", we don't retain any data too (everything is done in RAM per request). But that is still unsatisfactory personally. We will soon add "privacy pass" support, but still not to the satisfactory. Transparency log that can be attested on the hardware would be nice (since we run our open-source gRPCServerCLI too), but I just don't know where to start.
It is completely compromised, especially if it is an AI company.
How do you think ollama was able to provide the open source AI models to everyone for free?
I am pretty sure ollama was losing money on every pull of those images from their infrastructure.
Those that are now angry at ollama charging money or not focusing on privacy should have been angry when they raised money from investors.
https://github.com/ollama/ollama/issues/5245
If any of the major inference engines - vLLM, Sglang, llama.cpp - incorporated api driven model switching, automatic model unload after idle and automatic CPU layer offloading to avoid OOM it would avoid the need for ollama.
I pay $20 to Anthropic, so I don’t think I’d get enough use out of this for the $20 fee. But being able to spin up any of these models and use as needed (and compare) seems extremely useful to me.
I hope this works out well for the team.
Usage-based pricing would put them in competition with established services like deepinfra.com, novita.ai, and ultimately openrouter.ai. They would go in with more name-recognition, but the established competition is already very competitive on pricing
It's very unfortunate that the local inference community has aggregated around Ollama when it's clear that's not their long term priority or strategy.
Its imperative we move away ASAP
For one of the top local open model inference engines of choice - only supporting OSS out of the gate feels like an angle to just ride the hype knowing OSS is announced today "oh OSS came out and you can use Ollama Turbo to use it"
The subscription based pricing is really interesting. Other players offer this but not for API type services. I always imagine that there will be a real pricing war with LLMs with time / as capabilities mature, and going monthly pricing on API services is possibly a symptom of that
What does this mean for the local inference engine? Does Ollama have enough resources to maintain both?