Thanks to Sam Altman... ;-)

Wow. Sam Altman did something fantastic. He motivated me. I’m using my AI workstation to work on Linux and experiment with AI!

Last winter I bought a new machine to edit my photos and learn AI on Linux. I’m still doing the photo editing part, but not the AI part. I wanted to keep going, but with my new work responsibilities I was too tired to spend energy on Learning. However, the statement about RAM by Sam Altman turned out to be a good motivation. No promises for regular new articles tough…

First of all, some RAM p*rn :-) 128 GB of RAM, from that 8 GB is set to be video RAM (but it could be set to anything from 512 MB to 96 GB…). The price of the little box doubled within a month of purchase simply due to the amount of RAM included. Sam & Co. have made RAM pretty much unaffordable…

root@fedora:~# free -h
               total        used        free      shared  buff/cache   available
Mem:           117Gi       2.2Gi       115Gi       8.0Mi       1.1Gi       115Gi
Swap:           39Gi          0B        39Gi

And finally, I’m working on Linux and AI again. Fedora is now updated from version 44 to 45 on the machine. And openSUSE Tumbleweed has also seen gigabytes worth of updates. Next up: getting AI workloads running again. I had to remove ROCm during upgrade: dependency problems, as far as I could see. I have just reinstalled the packages, and checking documentation…

I have Open WebUI installed from half a year ago, together with ollama. I tested both from the command line and from the Open WebUI: they work. Next time I will try to do something actually useful. :-)

root@fedora:~# ollama list
NAME               ID              SIZE      MODIFIED       
llama3.2:latest    a80c4f17acd5    2.0 GB    24 minutes ago    
gemma3:27b         a418f5838eaf    17 GB     7 months ago      
granite4:latest    4235724a127c    2.1 GB    7 months ago      
root@fedora:~# ollama show granite4
  Model
    architecture        granite    
    parameters          3.4B       
    context length      131072     
    embedding length    2560       
    quantization        Q4_K_M     

  Capabilities
    completion    
    tools         

  License
    Apache License               
    Version 2.0, January 2004    
    ...                          

root@fedora:~# ollama --verbose
>>> who are you?
I am Hermes, a language model created by IBM to assist with a variety of tasks and answer questions on many topics. I strive to provide helpful, informative, and engaging responses 
while being respectful and avoiding any harmful or unethical content.

total duration:       1.111554719s
load duration:        42.916831ms
prompt eval count:    12 token(s)
prompt eval duration: 23.637941ms
prompt eval rate:     507.66 tokens/s
eval count:           47 token(s)
eval duration:        1.020660833s
eval rate:            46.05 tokens/s
>>> print the first 10 prime numbers
The first 10 prime numbers are:

1. 2
2. 3
3. 5
4. 7
5. 11
6. 13
7. 17
8. 19
9. 23
10. 29

total duration:       1.434860225s
load duration:        47.052753ms
prompt eval count:    75 token(s)
prompt eval duration: 106.452539ms
prompt eval rate:     704.54 tokens/s
eval count:           58 token(s)
eval duration:        1.254814127s
eval rate:            46.22 tokens/s
>>> Send a message (/? for help)

You can read my older posts about my AI box under the toy tag.

• AI
• toy
• planets