Anyone have any experience with or opinion about “really open source” models?
Wanted to try Apertus because they say it know lot’s of languages, or at least trained on lot’s of languages. But I got this
so I had to make huginface acc, I don’t consider it open in that sense. It’s just their license is like that you have to request access and they store data you provided them with.
Big dislike for that, but if it know serbian and i’ts 8B model that I can run on a CPU with llama.cpp(gguf format) then great. DL is easy part xD There is no gguf yet, and per comments it cannot be made yet, so need to wait some time.
But anyway, I’m confused what “open source” in model world means ? Is it just the weights or you need entire training set to be available to download and train yoursefl, or something completley else ?
I didn’t yet took a look at allenai.org
2 Likes
I’m not really sure about first link, but the second one state that their models build “with open and accessible training data, open-source training code, reproducible training recipes”.
As far as I know that’s the difference between “Open Weight” and “Open Source” =)
2 Likes
1 Like
GitHub - RWKV/rwkv.cpp: INT4/INT5/INT8 and FP16 inference on CPU for RWKV language model Jel probao neko ovo ? Trebalo bi da je performatnije za lokalni llm
1 Like
za python ti je najjaci ChatGTP-PRO (cca 20€/mj) prvi mj. free… probaj…
!Note: Nvidia closes mega deal with company behind ChatGPT: investment of $100 billion
sve vam je jasno nasta ChatGTP ide…
najjači LLM vam je Gemini Advanced (isto cca 20€/mj) prvi mj. free…
sve ostali su ispod s kapacitetom…
2 Likes
2 Likes
1 Like
Qwen3.5 released
1 Like