Corrupt download?
The 8-bit version of the download seems to stop at 31.7GB, I tried from 2 public-wifi sites, to complete. I am on linux-mint 20, using, waterfox & firefox, its all up-to-date. In other words, it wont complete in the last 500MB. I have been trying to get this file for over 5 weeks, and it is a critical download, that I will not be able to attempt again. I have paused and resumed multiple times, it just stops then fails, retrying does exactly the same. I have 13b, 70b wont fit in 8bit, and to note gpt4 said as a rough estimation, that 30b @ 8bit would be better than 70b @ 4bit, for my needs.
Well gpt4 is pretty wrong here. First put 30b is llama 1 while 70b is llama 2.
Llama2 was trained on much more and higher quality data compared to llama 1.
Secondly a quantized larger model will most likely perform better then a unquantized smaller model. Gptq, awq, exl2, gguf quants also have been higher quality now.
You can look at this chart and it generally shows that a quantized high param model is better then a low param full model.
(Lower is better)
If you still really want to use platypus model, consider using wget command or something similar like nurb432 said
Oh, ok, I was assuming it was incorrectly named, and was actually llama2 based, in that case, wont be using it anyhow, and thanks for the tips. I think Platypus2 70b 5/6bit, is best for 64GB RAM then. I'm guessing this is because there was no 30b llama2, hugely disappointing, that one for users with 64GB.