4 ms·
Show HN: A GPU/VRAM filter for finding LLMs that will run on your hardware
I kept seeing people ask "Which model i can run on my gpu", "will model X fit on my GPU". Thats why I built a filter on whichllmmodel that lets you search models by what will actually fit on your hardware (8GB, 16GB, 24GB, etc.) at a given quantization level.
- necovek 4mo agoVery broken: "live minimums" do not allow me to remove 512 token limit and put a bigger number easily. No unified or shared memory scenarios (like Apple's M platform or AMD's integrated GPU platform).
- johng 4mo agoWas going to mention this. I'm on an M1 Max and wanted to see what the site suggested.
- mzubairtahir 4mo agoactually that input is broken. and sorry for that. and I am adding shared memory features in next iterations.
- mzubairtahir 4mo agothat broken input is fixed
- CRSilkworth 4mo agovery nice idea. Would be nice if you could also keep desired context as a free parameter and let the models tell you what maximum context you could have.
- mzubairtahir 4mo agoactually that's free by design, it is just broken. fixing it in next sprint. And really thanks for your feedback!!!
- mzubairtahir 4mo agonow that is fixed, please try it
- xlr8_track 4mo agoAwesome, how do I contribute to this? A gihub link or smthg?
- mzubairtahir 4mo agoactually, currently it is not open source, but I am thinking about making it open source so that other developers can also contribute in it(espeically data layer). what do you think?
- xlr8_track 4mo agoYes, many people like myself are willing to contribute. It'll take the load off you and give you time to work on other features or projects.
- GreyOcten 4mo agohandy, but the gap most of these filters have is that "fits in VRAM" doesn't mean usable. context length blows up the KV cache fast, a 7B that fits at 2k tokens will OOM at 32k. factoring context len + quant into the estimate is where it'd actually save people from getting burned.
- mzubairtahir 4mo agoi think you did not check app properly, it is actually taking required context window from the user and then caluclate kv cache size and then count it along with size of model itself. it also reserves some more memory to avoid oom....
- heliskyr2 4mo ago[flagged]