I have been heavily relying on Qwen3.6-27B-UD-Q4_K_XL.gguf -model and Pi agent (https://pi.dev/) for local tasks and coding. I have used llama-cpp-turboquant fork with some custom cherrypicked MTP patches from another fork.
I'm running this on V100 32GB (~900GB/s memory bandwidth) with 200,000 context window, --spec-type mpt --spec-draft-n-max 3 --spec-draft-n-min 0 --cache-type-k turbo3 --cache-type-v turbo3 to mention most relevant parts.
I usually get somewhere 45-60 t/s. I believe that speed could be improved slightly by switching to ik_llama.cpp fork and Qwen3.6-27B-IQ4_NL.gguf -model but there's no turboquant support and it's with some other tradeoffs too.
This might not be exactly what you meant or wanted, but I use OpenSnitch, which alerts me about all new connections. I also use temporary disposable virtual machines in QubesOS to isolate those kinds of activities from my more vulnerable systems.
You have to choose from one of two strategies, either you go with tor-browser (also includes Mulvad-browser) route and try make your browser indistinguishable from others or you randomize values to make stable fingerprinting impossible.
When trying to be similar to everyone else, even small changes to the browser, like changing window size, can make you easily identifiable from everyone else. Randomizing will allow you to modify your browser. None of the fingerprinting protections matter if you use your browser and session to login to some sites.
I use multiple browsers. One is for login to sites and tor-browser is for most of my browsing.
This is easily the best fingerprinting extension that I have found so far:
https://jshelter.org/
I'm having lots of problems with fingerprinting protection on Librewolf and ungoogled-chromium. I use uBlock Origin and JShelter extensions on both. I'm always getting "your browser is out of date" despite always having the most newest versions.
Some sites like Stackexchange will work after just reloading the page. And rest of the sites usually work when I remove Javascript protection and Fingerprint detection from JShelter. Sill not all of them. So, they maybe/probably want to reliably fingerprint my browser to let me continue.
If I use crappy fingerprint protection, I'm not having problems but if I actually randomize some values then sites wont work. JShelter deterministicly randomizes some values using session identifier and eTLD+1 domain as a key to avoid breaking site functionality but apparently Cloudflare is beeing really picky. Tor browser is not having these problems but it uses different strategy to protect itself from fingerprinting and doesn't randomize values but tries to have unified values across different users making identification impossible.
I'm running this on V100 32GB (~900GB/s memory bandwidth) with 200,000 context window, --spec-type mpt --spec-draft-n-max 3 --spec-draft-n-min 0 --cache-type-k turbo3 --cache-type-v turbo3 to mention most relevant parts.
I usually get somewhere 45-60 t/s. I believe that speed could be improved slightly by switching to ik_llama.cpp fork and Qwen3.6-27B-IQ4_NL.gguf -model but there's no turboquant support and it's with some other tradeoffs too.