I have recently also played a bit with llama.cpp
Best performance for your hardware is by running a model that uses most of the VRAM. A good match for me has been Qwen 3.5 9B Uncensored. I just tried chatbot things, and it has been pretty good.
Walt Munny
npub17l…3ew74
2026-10-26 20:30:27 UTC
in reply to nevent1q…qxah
Author Public Key
npub17ldlkylwcnyn3wjdmtse9320xpalgg8nzpkj5mg5q2axlc0uy6as63ew74Published at
2026-10-26 20:30:27 UTCEvent JSON
{
"id": "92cfe8728197b7dd0f7812898b15e3c320ce67dee92469965e7cf50adb9b99ac",
"pubkey": "f7dbfb13eec4c938ba4ddae192c54f307bf420f3106d2a6d1402ba6fe1fc26bb",
"created_at": 1793046627,
"kind": 1,
"tags": [
[
"e",
"7bf948ced855729a29bb885b89ab8745c6570d3055dfd2ea9d0b1d770bb43de4",
"wss://wm.feeds.relay.tools/",
"root",
"4d7842051782e0d3feb034d150adc2b6bae4ee3b49786793bffa468b6f5b96b3"
],
[
"p",
"4d7842051782e0d3feb034d150adc2b6bae4ee3b49786793bffa468b6f5b96b3",
"wss://relay.damus.io/"
],
[
"client",
"Amethyst"
]
],
"content": "I have recently also played a bit with llama.cpp\n\nBest performance for your hardware is by running a model that uses most of the VRAM. A good match for me has been Qwen 3.5 9B Uncensored. I just tried chatbot things, and it has been pretty good.",
"sig": "c7a99a160fec8d212f183271df701f51aa6f6ad3c671342ad67fe348cc867b9494f827ae8dd9fc5c71d07e1808070ed0dd638741d0c2a675e0b8e7e48ef4e6b8"
}