YOO WHAT?
and i just realized my zeroGPU usage is back to 5 minutes instead of the 40
Same 😭😭
RIP Hugging science😭😭😭
RIP hugging-science indeed 🪦💀
@8BitStudio p100 has no tensor cores, and the v100 has 1st gen cores and supports only fp32 and fp16 for training im pretty sure. plus you have to jerry rig something so they stay cool too
Yeah not the best for training or even inference on some models but cheap vram. I think at a time they were like $4 per gb or something
First step finished and earth dosent exist anymore lol
AMD isn't bad if you could survive the emotional/kernel shock of CUDA libraries
For me , best AI model for coding task is Qwen3.8 Flash Next.... 0.5$/1 Million token output is peak .... And if task is harder then Mimo 2.6 Pro 1T 0.9$/M token.... These chinese labs are amazing when it comes to cost
Lets just train on CPU
@DedeProGames i actually tried that once, made the pc run at 3 frames a second and sometimes it would be unresponsive for 10+ minutes, after like 3 hours and it finishing step 3 of 3000000000000 i was able to stop training it was painful
Hi everybody,
We're working on improving the allocation of the org resources and to get the resources back to Hugging Science.
Feel free to share how Hugging Science helps you work on AI for science so we can prioritize. Thanks a lot for your help!
Emma
Hi everybody,
We're working on improving the allocation of the org resources and to get the resources back to Hugging Science.
Feel free to share how Hugging Science helps you work on AI for science so we can prioritize. Thanks a lot for your help!
Emma
The ability to post was a big one for me even if I never got to use it in time
I didn't use any of the buckets or extra features because my work is focused on agentic harness stuff and quantization
but im sure seeing how many buckets and models this org hosts that others did find them extremely useful
just my two cents, hope this helps.