bakamomi

Please merge this because it's amazing on x86 with longer context. I tried generating 1500 tokens with the 7B model ( --ignore-eos -c 2048 -n 1500). On the master branch...

Alright, #775 clearly contributed to the results I got. I pulled master again with #775 already merged and now I'm getting: ``` llama_print_timings: load time = 928.29 ms llama_print_timings: sample...

Hi @gustavo-iniguez-goya I added this rule. I also tried enabling the built-in "Intercept forwarded connections (docker, etc)". It didn't work. Correct me if I'm wrong, but the rule you suggested...

The debug log turned out to be huge. I posted it on privatebin for convenience. https://bin.urla.no/?d4aba4eb25251af2#AYXK1uH97bJfcoZAjPJmC8dAC5iyWaPeYfPexm8PPTwD Regardless, the ip address of the wg interface (192.168.3.5) isn't recorded there. These are...

Thanks for looking into this! What way you choose to implement this feature is of course up to you, but here are my 2 cents: > they are routed through...

Hi, again! Since you said you already have PoC with nfqueues, can you post it maybe as a pull request or in a separate tree? I wouldn't mind compiling opensntich...