Damus
Open in Damus
verita84 profile picture
verita84
@verita84

Nostr's #1 "Bad Actor" according to Derek Ross for trying to get him to remove the CSAM from ditto.pub and Alex Gleason for making my own Fediverse bridge not to his standards and somehow hammering DItto's relays that he never told me about.

"Have nothing to do with the fruitless deeds of darkness, but rather expose them."
Ephesians 5:11

Creator of PosterChan, the best Nostr Experience!

nprofile1qqsyk44m7swf9evxazyj0t9h3qmwkj0jkxzqs8hc2f39eautul2kh4s5ng9uqnpub
[email protected]nip05
https://github.com/loblawbob873-svg/posterchanaiwebsite
Relays (1)
  • wss://relay.poster.place – read & write

Recent Notes

Josephus · 18h
Wtf is algia?
verita84 profile picture
🧠 Self-hosted AI model test for PosterChan's Netrunner agent: Qwen 3.6 vs the new Qwen 3.8 27B (uncensored builds), on our own GPUs.

Netrunner tasks completed (one run each): Qwen 3.8 19/21, Qwen 3.6 13/21
Picked the right first tool: Qwen 3.6 8/8, Qwen 3.8 5-6/8
Refusals: none for either

Speed (generation, then wait before the reply starts on a 7k-token prompt):
• Intel Arc A770 16 GB: Qwen 3.6 34 t/s, 20 s | Qwen 3.8 10 t/s, 67-84 s
• RTX 3060 12 GB: Qwen 3.6 88 t/s, 4 s | Qwen 3.8 23 t/s, 13 s

Qwen 3.8 is smarter on long tasks and reads images on 12 GB cards, but it's 3.5-4x slower (it's dense). We're staying on Qwen 3.6 for now, shrinking the agent's prompt, and will re-test 3.8 after that.

cc @Tony "Lemme Scrape" Bacon

#ai #llm #qwen #selfhosted #posterchan