I find that on my M2 Mac that number is a rough approximation to how much memory the model needs (usually plus about 10%) - which matters because I want to know how much RAM I will have left for running other applications.
Anything below 20GB tends not to interfere with the other stuff I'm running too much. This model looks promising!
That time is just for the very first prompt. It is basically the startup time for the model. Once it is loaded, it is much much faster in responding to your queries. Depending on your hardware of course.
The models feel pretty snappy when interacting with them directly via ollama, not sure about the TPS
However I've also ran into 2 things: 1) most models don't support tools, sometimes it's hard to find a version of the model that correctly uses tools, 2) even with good TPS, since the agents are usually doing chain-of-thought and running multiple chained prompts, the experience feels slow - this is even true with Cursor using their models/apis
People have all sorts of hardware, TPS is meaningless without the full spec of the hardware, and GPU is not the only thing, CPU, ram speed, memory channel, PCIe speed, inference software, partial CPU offload? RPC? even OS, all of these things add up. So if someone tells you TPS for a given model, it's meaningless unless you understand their entire setup.
I’ve been playing around with Zed, supports local and cloud models, really fast, nice UX. It does lack some of the deeper features of VSCode/Cursor but very capable.
In ollama, how do you set up the larger context, and figure out what settings to use? I've yet to find a good guide. I'm also not quite sure how I should figure out what those settings should be for each model.
There's context length, but then, how does that relate to input length and output length? Should I just make the numbers match? 32k is 32k? Any pointers?
Ollama breaks for me. If I manually set the context higher. The next api call from clone resets it back.
And ollama keeps taking it out of memory every 4 minutes.
LM studio with MLX on Mac is performing perfectly and I can keep it in my ram indefinitely.
Ollama keep alive is broken as a new rest api call resets it after. I’m surprised it’s this glitched with longer running calls and custom context length.
I was able to run it on my M2 Air with 24GB. Startup was very slow but less than 10 minutes. After that responses were reasonably quick.
Edit: I should point out that I had many other things open at the time. Mail, Safari, Messages, and more. I imagine startup would be quicker otherwise but it does mean you can run with less than 32GB.
Almost all models listed in the ollama model library have a version that's under 20GB. But whether that's a 4-bit quantization (as in this case) or more/fewer bits varies.
AFAICT they usually set the default tag to sa version around 15GB.
I find that on my M2 Mac that number is a rough approximation to how much memory the model needs (usually plus about 10%) - which matters because I want to know how much RAM I will have left for running other applications.
Anything below 20GB tends not to interfere with the other stuff I'm running too much. This model looks promising!