@RN_
@T3s|4_
@eyedeekay
@orignal
@postman
@zzz
%Liorar
%acetone
%mareki2p
%snex
+FreefallHeavens
+Onn4l7h
+Onn4|7h
+altec_lansing
+leopold
+nyaa2pguy
+poriori
+profetikla
+qend-irc2p
+r00tobo
+uop23ip
Arch
Danny
Holmes
Irc2PGuest28384
Irc2PGuest63277
Irc2PGuest71791
Irc2PGuest84270
NiceBoat_
OfficialCIA_
Over1
SilentWave
U1F642
Wikk_0
ahiru
cims
dr|z3d
eyedeekay_
floatyfloatilla
mahlay
makoto
mesh
n2
nilbog
not_bob_afk2
pory
r00tobo[2]
red
sektorchef
thetia
user_
utp
vivid_reader56
nyaa2pguy
T3s|4_: I haven't used the Fed for that long, i'm basically a long time debian user hehe
T3s|4
nyaa2pguy: np, whatever works for you is best :D
RN_
who you callin' a fed nyaa2pguy? I'm a Bestie!
orignal
he is Japanese ))
RN_
LOL
RN_
oh... just realized
RN_
s/Bestie/Beastie/g
dr|z3d
> just working on a bash script that parses various ips from .i2p/sessionbans/sessionbans.txt and creates an applies a 24h iptables blocklist for all found ips that match the rules.
dr|z3d
iptables bans for anything over 4 hours, sybil analysis, old and slow and XG routers.
dr|z3d
run via crontab for automatic iptables bans.
dr|z3d
not bad, 33 microseconds job lag running on 2 runners (threads)
dr|z3d
new + dev build uploaded to skank. if you review 127.0.0.1:7657/jobqueue once you're updated (don't forget to ctrl+sshift+r) you'll notce the humber of finshed jobs will vary depending on your router's job queue backlog / latency.
dr|z3d
*number
dr|z3d
also on /jobs, we now by default track the last 10s for the jobqueue so you'll need to wait a few moments for it to refresh when you visit the page. you'll see just the stats for the jobs that have run in the last 10s, and the stats will be for that 10s window. full stats for all jobs also available via header link.
dr|z3d
I now have both a bash script and a python script for converting bans in .i2p/sessionbans/ to a tagged iptables banlist with removal after a configured period. once it's polished, I'll add it to the repo for anyone interested.
dr|z3d
if you want early access, ping me.
not_bob
Recent versions are working much better. Thank you!
dr|z3d
great, not_bob, thanks for the feedback. sorry for the extended shitstorm, hopefully worth the pain. :)
not_bob
It's how it goes sometimes.
not_bob
My scanner didn't detect any sites at all for a while (new). And then *WOOMF* a shit ton showed up.
dr|z3d
aside from various enhancements, both user-facing and on the backend, we have various fixes that should finally be robust after an extended wrangling session that spanned several weeks.
dr|z3d
the remove slow tunnels from local pools feature may well need additional work to get optimal, but with the fairly conservative defaults, it should work well enough (I hope).
dr|z3d
there are override config options listed on /help/advancedsettings for anyone who wants to play around with it.
dr|z3d
*woomf* is always a pleasant experience :)
not_bob
Yes
dr|z3d
re your scanner, now you just need to get it to use the recorded latency in the hostchecker file.
not_bob
I had a very busy day sorting new sites.
dr|z3d
RTP did a video on the new addressbook features, dunno if you saw that?
dr|z3d
I think he probably name-dropped you. :)
not_bob
Yes.
not_bob
I saw that.
not_bob
And, he did.
dr|z3d
good, good. so people *know* you're the source for categories, no nasty banner required :)
not_bob
lol
not_bob
So much AI slop out there :(
dr|z3d
there is? where's there?
dr|z3d
I hope you don't mean harry.i2p
not_bob
No, I just meen in general.
snex
gatheryourparty.i2p but i cant make stories anymore because ardullm disappeared
not_bob
I was playing with a local LLM a few days ago. It decided that z3pplin created the I2P network and even tried to explain to me how you can do some math to get the b64 from a b32 address...
dr|z3d
speaking of AI, you may be interested in this: phoronix.com/news/Open-Source-GreenBoost-NVIDIA .. I'm working on an AMD port.
dr|z3d
sounds like you're behind the curve.
not_bob
That looks like useful.
not_bob
I haven't had much free time at all as of late.
dr|z3d
can't make stories, snex?
not_bob
But, I'm impressed with having a local model that can reason.
dr|z3d
what I mean is, the best LLMs these days are agentic and will automatically research topics using available online data if they don't know. they might need a nudge to get there.
not_bob
Yes, that helps a lot. But, privacy is an issue.
snex
i can make stories but i cant make regular hard-boiled detective stories that take place in a d&d fantasy world
snex
> i can make stories but i cant make regular hard-boiled detective stories that take place in a d&d fantasy world
dr|z3d
why not?
snex
because thats hard work
dr|z3d
I mean, why not, what was ardullm providing?
not_bob
snex!
snex
an important aspect of the hard-boiled genre is that the stories dont exactly make sense and very little is resolved at the end. AI is super good at this
snex
ardullm made 100% of the content of the site
snex
the stories, the images, the titles, all of it
dr|z3d
plenty of sites out there that will do the same job(s), for free.
snex
i doubt anybody is gonna do for free that amount of content
dr|z3d
obviously you haven't looked hard enough.
dr|z3d
for text content, try perplexity.ai as a starter.
snex
also commercial LLMs will probably refuse to do forgotten realms stuff as thats copyrighted
dr|z3d
as for images, there are a ton of free services out there that don't require signup. freegen.app is but one.
dr|z3d
perplexity may throw a signin dialog in your face every so often, but otherwise it doesn't appear to impose daily limits like some.
dr|z3d
(and you don't need to sign in to use it)
snex
i told that to write a short story and it timed out
snex
2nd try. i get 3 paragraphs
snex
let me give it the actual prompt i used with ardullm
dr|z3d
fix your internet.
snex
yeah this thing cannot write a full story
dr|z3d
er, yes it can.
snex
i get 3-4 paragraphs consistently
snex
read the stories on the eepsite
dr|z3d
fix your internet :)
snex
it has nothing to do with internet
dr|z3d
different llms offer different context window sizes. perplexity isn't the best, but it doesn't do too badly. want more, try chat.qwen.ai
snex
qwen is terrible at writing these. ardullm had qwen and i tried all his models. gemma was used for the stories
dr|z3d
this is from perplexity, it'll do a chapter at a time (this is a single response): drop.i2p/p/c5af3b27-6f59-4ac6-a0c2-c27e3c82e7b6
snex
ill fuck around with it later. i have a semi-advanced prompt generator that creates a forgotten realms style date (based on the given earth date), the race of the protagonist, what city he lives in, and some possible quirks he can have chosen at random
snex
lol
snex
> ill fuck around with it later. i have a semi-advanced prompt generator that creates a forgotten realms style date (based on the given earth date), the race of the protagonist, what city he lives in, and some possible quirks he can have chosen at random
dr|z3d
whatever you need is out there, anyways.
nyaa2pguy
on SAM limits: i just tried 100 sessions and it worked pretty well lol. i was expecting routers to ban me but i guess it wasn't too bad (they all had a delayed startup too)
nyaa2pguy
this is with canon
snex
i actually went through the FR wiki and found the racial makeups of like 100 major cities and the protag race more or less has odds matching those
snex
theres always a small chance of a bizarre exotic monster race like a beholder or a dragon or something. one story on the site actually rolled a beholder protag and its hilarious
dr|z3d
just make sure you create a good harness for optimal results.
dr|z3d
use opencode, you can tell it to refer to websites/pages for source info etc.
dr|z3d
and you can always iterate on whatever it generates.
dr|z3d
it'll write the story, generate the html, various things.
dr|z3d
it can probably be persuaded to use an image generator and integrate the resulting image into your site with a bit of poking.
snex
the html is already how i want it, i just need the story, its title, and an image that integrates with it
snex
the nice thing about ardullm was that after the story was over, i could ask it to describe the macguffin or some other visual that i thought was important and then feed that description back into the image gen
dr|z3d
what I mean is, it'll happily use an exisiting html template and create a new page, or modify an existing page with the necessary html. but you'll find out. money back if you don't like it.
snex
ill just run my own shit at home and make an i2p portal to use it if you all buy the hardware for me =P
dr|z3d
yeah, right.
dr|z3d
*** laughs. ***
dr|z3d
we had this discussion before. you saw zero need for hardware. funny how times change :)
snex
zero need to run what i currently run
snex
i have no advanced gpus here
snex
i estimated the cost of what ardullm was doing and it was like 30k... that dude was totally a fed
dr|z3d
best bang for buck right now is probably the 9070XT if 16GB is sufficient.
dr|z3d
no.
dr|z3d
he had older hardware he bought for peanuts.
snex
he was doing video gen
dr|z3d
yup
dr|z3d
and?
dr|z3d
I do that here.
dr|z3d
I can generate 30s of video with an audio channel here without issue. longer if I wanted.
dr|z3d
he had something similar to this: newegg.com/nvidia-tesla-m40/p/1FT-0004-005S6 .. not recommended, and he probably paid around a 1/3 of the price on newegg.
dr|z3d
for around the same price, if 24GB is what you want, this is a much better bet: newegg.com/powercolor-hellhound-rx7900xtx-24g-l-oc-white-radeon-rx-7900-xtx-24gb-graphics-card-triple-fans/p/N82E16814131812T
dr|z3d
not necessarily that card, but that's around the price for new, less 2nd hand.
snex
i also have nothing to slot that into
snex
i was wondering if the bee-link docking thingy could handle these
dr|z3d
what ports do you have available?
snex
i have nothing
dr|z3d
usb4?
snex
all of my hardware is being used at capacity
dr|z3d
gpus will run over usb4 (thunderbolt, kind of)
dr|z3d
but you probably don't have any of those, by the sounds of things.
snex
nope
snex
my desktop is ancient, my xmr miner is used for xmr mining, and everything else is a laptop, bee-link, or le-potato
dr|z3d
well, anyways, the 7900XTX is supported by that adapter.
snex
i wouldnt mind spending to get an ai setup if i thought itd be worth it. i just dont think its worth it
snex
its a toy
dr|z3d
I don't see anything that isn't support other than the latest Nvidia gen which you don't want anyways.
snex
so if people want me to have an expensive toy that i let them use for free with free cycles then they can buy it for me =P
snex
unlike ardullm im not gonna randomly disappear with no notice
snex
unless im dead
snex
how do i make perplexity act like its writing a story rather than trying to have a conversation with me
dr|z3d
this is the prompt I started with: "write a story about a hard-boiled detective stories that take place in a d&d fantasy world in around 10K words."
dr|z3d
adjust to taste, I just cut and pasted your description with zero effort to refine it.
dr|z3d
for the followup, I used "sure, we want 10K words in chapters, please."
dr|z3d
I see you got busy on your site.
snex
what site
snex
the stories up there are old
snex
if you know forgotten realms at all you can convert the dates back into earth dates
snex
just a nice little easter egg for people who like FR
snex
i have another script that tells me how far apart each story should be. its 1 week on avg but random with a poisson distribution
snex
the database has a published_at field so i can roll stories in advance and they get revealed on the website as the date passes
snex
if perplexity works out ill start rolling em again
snex
what was really cool was that the model actually seemed to know about the FR date system so it would change the weather based on what the season was
snex
it knew that it couldnt rain in the underdark
snex
i did a manual override to make sure that you never had more than 2 human protag stories in a row, because humans are way overrepresented
snex
theres actually #gatheryourparty on here but nobody but me and cumlord ever showed up there lol
dr|z3d
maybe try perplexity for the initial outline, and opencode to refine.
dr|z3d
opencode will take an entire story and allow you to refine it, referencing web pages etc to bolser/correct.
snex
i care less about "correctness" whatever that means. the hallucinations are part of the fun with these stories
snex
theres not really anything new going on in the forgotten realms, at least not anything id actually want. wizards of the coast has been only making shit woke lately
snex
i dont need it to devolve into a gay love story between orcs and drow
snex
and then tell me how actually orcs and drow are just victims in a world that shuns them for no reason
dr|z3d
snex: doesn't seem too bad, either: perchance.org/ai-vibrant-image-generator