DDraceNetwork - developer

00:40

trust me bro it works

what am I looking at

vscode diff

hmm thats a lot of diff

yeah

SollyBunny

[ddnet/ddnet] Pull request opened: #9601 Rewrite nameplates again

!I'll fucking do it again

Id can be put onto new line with cl_nameplates_ids_line
No more flicker of nameplates
Hearts now an icon
Added dummy nameplate preview
Its modular now

Checklist

[x] Tested the change ingame
[x] Provided screenshots if it is a visual change
[x] Tested in combination with possibly related configuration options
[ ] Written a unit test (espe...

why does the text in the screenshots look blurry? was it always like that?

idk

01:00

01:01

old (edited)

01:01

new

01:01

one of them was taken at 1920/2 x1080

01:02

standard zooms gotta be rough i guess

01:02

could turn off that if not in game

01:03

i aded an ingame for you ><

Solly

Click to see attachment 🖼️

Is the 2 at the top of this the ID?

yes

Was below hook strength previously

do you want it that way around

Ya, strength is better at the top imo

kk

SollyBunny

[ddnet/ddnet] Pull request opened: #9602 Clearer arch linux clang format

I completley missed that there was an aur package the first time I read it ages ago and just now. Make it more obvious it exists and isn't just for mac (Yeah yay sucks but everyone will recognise it)

How sv_input_fifo works, and how to use it on VDS

https://youtu.be/yh1pF1zaauc

Rob Braxman Tech

End-to-End Encryption Now a Historical Footnote. They Won.

@Discord Mod

1

02:33

every channel

Solly

Click to see attachment 🖼️

the heart icon is too detailed

Learath2

https://youtu.be/yh1pF1zaauc

end-to-end encryption situation is crazy

Learath2

https://youtu.be/yh1pF1zaauc

i agree with these sort of stances on privacy but it's kinda silly to call the average tech user a 'normie' when these sort of messages should be presented to the average tech user

02:42

if this guy wants to make videos for his techie echo chamber community he can, but its silly to act like the average person is a tech illiterate normie who's ruining privacy for everyone else

louis

if this guy wants to make videos for his techie echo chamber community he can, but its silly to act like the average person is a tech illiterate normie who's ruining privacy for everyone else

no offence to anyone here but thats literally how most techies act

02:53

they trash on everything that isnt linux

you generalizing like that only makes it less easy to get along though

03:05

also i personally trash on everything including linux

Learath2

https://youtu.be/yh1pF1zaauc

is there anything new in this video?

03:08

from the parts I skimmed it seems like they have not removed e2ee

03:09

it's pretty obvious people care more about convenience than privacy anyway (we're all here using discord)

03:10

there's nothing the government can do to stop you from using math if you so decide

discord apparently has e2ee for voice and video at least

really? that's impressive

03:12

e2ee between you and discord or between you and the other participants?

03:13

new viral github PR https://github.com/ggerganov/llama.cpp/pull/11453

ggml : x2 speed for WASM by optimizing SIMD by ngxson · Pull Reques...

Motivation This PR provides a big jump in speed for WASM by leveraging SIMD instructions for qX_K_q8_K and qX_0_q8_0 dot product functions. Surprisingly, 99% of the code in this PR is written by De...

Tater

e2ee between you and discord or between you and the other participants?

seems like between users only https://discord.com/blog/meet-dave-e2ee-for-audio-video#heading-3

Tater

new viral github PR https://github.com/ggerganov/llama.cpp/pull/11453

i was almost impressed

the ai wrote code that improves its own codebase

yea the ai wrote it

03:46

not a human

if ai is so good when is it going to go through all the issues and make prs for all of them while still maintaining backwards compat

Hecta

yea the ai wrote it

But a human verified it and probably tested it

04:09

development and optimization haven't been any way more easier than now

the ai did the annoying part of changing the syntax to be a special flavor of SIMD, the human just had to write tests and benchmark it.

04:10

it's like compiler optimization but better

Yeah, wonders

04:10

A net positive overall no?

04:11

Maybe not the current time for an AI to do it fully, but the accomplishment of AI helping with certain things for developers is exponentially revolutionary

the reason this PR is viral is because codebase is the same one that could be used to run the ai that wrote the PR, the fact that ai can write code is nothing new or notable. This is just a novelty

Oh, so AIception? xd

Hecta

if ai is so good when is it going to go through all the issues and make prs for all of them while still maintaining backwards compat

Tater

Click to see attachment 🖼️

the rocket emoji cammostripes

it's only a question of when

probably not anytime near in corperate america

quite the opposite actually

04:15

coperate america can't help but pay hand over fist for ai solutions to their problems

yea but its probably an expensive closed spurcw thing

deepseek is open source as well as metas models

Bamcane

[ddnet/ddnet] Issue opened: #9603 [EDITOR IDEA] Auto generate quads from SVG file

In order to decrease the map footprint, I currently achieve some map effects by quads without any image. Here are some of them: !Image !Image Also, in some maps, some things are only used once, such as Tee, map logos. I think if we can auto generate quads from svg file, the mappers can create more wonderful map effects with low...

louis

deepseek is open source as well as metas models

how long do u think itll take until deepseek is bought out

i mean you can't buy out the open sourced weights and code and paper published

04:37

their team might change their licensing but it still won't stop open source models getting better

Open source models don't matter all that much if the training set isn't freely available and the training process is too expensive for any small group of individuals

you can self host open source models while you need to pay for closed source apis

and you can fine tune them too, sure

04:53

but your freedoms are still severely limited

not really

risu

Open source models don't matter all that much if the training set isn't freely available and the training process is too expensive for any small group of individuals

a small group of individuals cannot create cpus, does that make them not matter all that much?

if you have the entire training set and 5000 gpus it doesn't matter what model we're even talking about

Tater

a small group of individuals cannot create cpus, does that make them not matter all that much?

I mean, the supposed openness doesn't matter all that much Just like even if you were given the patterns used for the masks of a CPU it wouldn't matter all that much

their techniques are already being replicated by other labs RIGHT NOW

(though that would matter a whole lot more because you could look for interesting features)

they reduced training costs for all models by 95% not just theirs

you can't really self host a cpu with just a cnc machine though

04:59

open source models ensure privacy and offlineability etc.

04:59

and they also have research benefits ^

You can self host a CPU, that's what you do when you run it in your own machine instead of using hardware as a service

Tater

their techniques are already being replicated by other labs RIGHT NOW

Yeah, if a CPU manufacturer released their masks it would matter a whole lot for other CPU manufacturers

Tater

they reduced training costs for all models by 95% not just theirs

to be clear I don't mean the model itself reduces training costs, the techniques in the paper they published are more than detailed enough to replicate their work, many orders of magnitude better than other ai labs

05:00

they basically did release the masks (edited)

what I'm saying is that it doesn't help the little guys like open source software usually does

05:01

and that's what irks me

risu

You can self host a CPU, that's what you do when you run it in your own machine instead of using hardware as a service

that's different

who are the little guys?

We?

if you're given the specs and instructions to build a cpu you're not going to be able to come out with a working cpu

05:02

if i have a 3080 sitting at home i'm able to run small 7b models without worrying about whether chatgpt will be up or not

there is never going to be a time now or in the future again where the little guys get to run the best LLM model because methods that use more compute (and more money) will always be better http://www.incompleteideas.net/IncIdeas/BitterLesson.html

and i don't have to worry about my data being collected in some cloud data center

this is the closest thing you can possibly get to helping the little guys

Tater

there is never going to be a time now or in the future again where the little guys get to run the best LLM model because methods that use more compute (and more money) will always be better http://www.incompleteideas.net/IncIdeas/BitterLesson.html

Yeah I know, it's the same as people never being able to fab their own chips...

05:03

life do be sad

obviously it's not going to be able to solve the hardest coding problems but it can do the job for simple llmable tasks like organizing a todo list or whatever

people are running the model on their own hardware in their homes (the full model)

05:04

this will probably never happen again

05:04

openai will integrate their techniques and the open models will never catch up

05:05

this is like the first time you have ever been able to run a model that is equal to the best model at home since GPT2

I guess I'm just being a doomer

they're giving away all their effort for free, imagine intel spent $500 million making a new cpu and then just sold it for $0

I'm being here sitting grumpy like "how dare they be excited when we're all doomed anyways"

eh i mean they're probably making money since they have a public api

I don't think they're in it for the money

05:08

their api is very underpriced

05:09

maybe they put in a nasty short on nvidia stocks before releasing their paper

they probably have the same motivation as meta/facebook

05:12

they publish their work to prevent the closed research labs from getting a huge lead over everyone else. Even if the individuals can't make direct use of it the people contributing to the open knowledge research can.

actually yeah, there's value in keeping anyone from running too far ahead

"Due to large-scale malicious attacks on DeepSeek's services, registration may be busy."

LOLL

sjrc6

[ddnet/ddnet] Issue opened: #9604 Improve server voting menu

The whole system could use a refresh. Even on official servers we still need to add fake vote options for titles and do things like (Reason=Stars) which is very confusing. You also need to vote a different category before being able to see maps in that category. It would be nice if we could bring it to a better UX standard.

github-merge-queue[bot]

[ddnet:master] 2 new commits

28b7203 clearer arch linux clang format - SollyBunny f9156d5 Merge pull request #9602 from SollyBunny/clearer_arch_linux_clang_format - def-

05:53

def-

[ddnet/ddnet] Pull request closed: #9602 Clearer arch linux clang format

GitHub

Click to see attachment 🖼️

but backwards compatibility

06:15

~~just stack more protocol~~

def-

[ddnet/ddnet] Pull request opened: #9605 Try to upgrade gradle

Seeing build failures currently on nightly.

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test (especially base/) or added coverage to integration test
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memcheck](htt...

morning peeps

07:16

and peeppetes

Koll Potato

but backwards compatibility

I still think it's worth discussing

time to review more prs

edg-l

[ddnet/ddnet] Pull request closed: #5313 material layer

07:44

edg-l

[ddnet/ddnet] Pull request closed: #5403 Draft: Implement Ice Material

07:44

edg-l

[ddnet/ddnet] Pull request closed: #5413 Draft: Implement bouncy slime material

@Learath2 do u wanna continue working on https://github.com/ddnet/ddnet/pull/2754 ?

Check for the integrity of the data directory by Learath2 · Pull Re...

Isn't finished yet but it'd be nice if you could take a look before I spend more time on it. Fixes #2748

github-merge-queue[bot]

[ddnet:master] 2 new commits

32cad0f Show position of paused character currently spectating - Pioooooo fc45c80 Merge pull request #9598 from Pioooooo/spec-pos - edg-l

07:45

edg-l

[ddnet/ddnet] Pull request closed: #9598 Show position of paused character currently spectating

07:45

edg-l

[ddnet/ddnet] Pull request closed: #6724 Update to DDNet HUD and OPD (On Player Displays)

07:46

edg-l

[ddnet/ddnet] Pull request closed: #6959 Redesign quad/env/quadenv points

07:46

edg-l

[ddnet/ddnet] Pull request closed: #6797 Use C++20 (fixes #6552)

jesus, that's a lot of stale PRs

07:48

glad ur doing this, but at the same time it's sad seeing all the cool updates just sitting there

its better to close em imho they are 2+yr old and i plan on redoing some

07:48

ill try the c++20 now

hell yea

also those had like 10+ conflicts xd

07:49

we also getting a lot of new cool prs lately

Ryozuki

also those had like 10+ conflicts xd

well duh, they probably collected more dust than a vacuum cleaner

im gonna make a label: has-conflicts

07:53

makes knowing what to review easierr

07:53

the sad thing is the pr author cant remove it

07:53

mayb not worth it

github-merge-queue[bot]

[ddnet:master] 2 new commits

fe14bd0 Fix calling convention error - CherryEx f2bd19c Merge pull request #9595 from CherryEx/pr_discord_msvc_fix - edg-l

07:53

edg-l

[ddnet/ddnet] Pull request closed: #9595 Fix win32 msvc build with discord

waiting on author maybe better

Looking back at the reviews on my pr, I do want to discuss touching up the editor in the future. Mainly small things, like:

How in src/game/editor/editor.cpp:4228, the padding for Ui()->DoPopupMenu() doesn't change when it's selecting a game layer, so it's massive when it shouldn't be
How, with these new automapper updates, they could be put on top, where the Destructive button is
How, like you said, a lot of the variables should be renamed to be more clear.

Improving the usability of automappers by VoxelDoesCode · Pull Requ...

Summary As of: Jan. 12, 2025: This PR focuses on making the automapping feature much more manageable and flexible for mappers and map designers. The features included/planned are as follows: Addin...

edg-l

[ddnet/ddnet] Issue opened: #9606 Use "enum class" more instead of plain (anonymous) enums

They pollute less the namespace and are safer (tm) https://stackoverflow.com/questions/18335861/why-is-enum-class-considered-safer-to-use-than-plain-enum Also c++20 deprecates arithmetic between different enum types, which we use a bit, we should look into a alternative like constants maybe

08:12

edg-l

[ddnet/ddnet] Pull request opened: #9607 Use c++20

Checklist

[x] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test (especially base/) or added coverage to integration test
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memcheck](https://github.com/ddnet/ddnet/#using-ad...

They can even be in one packet. Just slap both in there. Old clients ignore the messages they do not know. Should be simpler than 0.7 which you did at 3am

Replying to @Davide I may try to send both redirect & reconnect, but it will be a little bit…

Yay blessed by def instamerge

How sv_input_fifo works, and how to use it on VDS

def-

[ddnet-web:master] 1 new commit

4e25d10 Master2 - def-

any opinions on this https://github.com/ddnet/ddnet/pull/1264

Fix (partially) skipping through teles and death tiles by heinrich5...

Previously you could partially skip through teleporters: Line two teleporters up in a line, if you through both of them in a single tick, first, the game would teleport you to the first teleporter&...

close this

09:27

its not a bug its a feature

edg-l

[ddnet/ddnet] Issue opened: #9608 Discussion: Change references from make to ninja, and require ninja

nproc may not exist on some platforms (macos, windows?) and ninja always uses the most resources, i also think its widespread enough that its not a problem. It would be cool also if ninja could be made default on unix in cmake but i guess thats non trivial maybe

Don't think it should be fixed right? It's used on many maps and is mostly common knowledge for experienced players. It would fix many skips which would result in many ranks being invalid. Removing those ranks would be pretty unfair since people spend a lot of time trying those skips since skipping death tiles is mostly inconsistent and you can get unlucky. Skipping tele is also a nice trick.

@murpi should we poll this? im unsure what to do so i wont do anything for now xd

Teero

Don't think it should be fixed right? It's used on many maps and is mostly common knowledge for experienced players. It would fix many skips which would result in many ranks being invalid. Removing those ranks would be pretty unfair since people spend a lot of time trying those skips since skipping death tiles is mostly inconsistent and you can get unlucky. Skipping tele is also a nice trick.

we have a feature that we can whitelist maps that use old bugs tho

*also would invalidate my deadrun tas

Ryozuki

we have a feature that we can whitelist maps that use old bugs tho

Wouldn't that be a little confusing? xd

we do that already with a grenade bug in some map

09:31

but i guess xd

09:31

tbh i have no opinion on this issue i dont rly care

09:31

but its the oldest pr open now so we should decide (edited)

I mean ddnet should move forward I think this should be fixed with your map setting

are there server options to enable/disable bugs? like sv_shotgun_bug

i think it should stay as is, it's part of ddnet's history and identity

would be useful if ddnet wants to get rid of bugs but keep them for old maps

github-merge-queue[bot]

[ddnet:master] 3 new commits

d55dd9f Consistently indent bullet points - heinrich5991 cdaa9c8 README.md: Don't require the user to mkdir/cd - heinrich5991 9f0f75e Merge pull request #8365 from heinrich5991/pr_ddnet_readme_cmake_b - edg-l

09:46

edg-l

[ddnet/ddnet] Pull request closed: #8365 README.md: Don't require the user to mkdir/cd

There is any way to differentiate a normal connect packet from a dummy connect packet? (I guess no?) @ChillerDragon @heinrich5991

nah, there is no difference afaik

rip me

why rip

because I need to write a lot of stuff haha

def-

[ddnet-web:master] 1 new commit

0c1912f master2 funding - def-

@Skeith

10:07

someone online?

Thanks

nice

How the DDNet version is encoded in the sys.client_version packet?

you can check the NETMSG_CLIENTVER message

Davide

How the DDNet version is encoded in the sys.client_version packet?

https://github.com/ddnet/ddnet/blob/9f0f75e38d3842d8eef38ce778e21745a568e8a7/src/engine/client/client.cpp#L214

ddnet/src/engine/client/client.cpp at 9f0f75e38d3842d8eef38ce778e21...

DDraceNetwork, a free cooperative platformer game. Contribute to ddnet/ddnet development by creating an account on GitHub.

Thanks

hmmm i wonder if MsgVer.AddRaw(&m_ConnectionId, sizeof(m_ConnectionId)); changes when a dummy is being connected

10:29

ig it stays the same

Davide

How the DDNet version is encoded in the sys.client_version packet?

if you are working with the protocol i suggest setting up wireshark with libtw2 dissector

1

he is most certainly familiar with those things

1

alright

10:34

Any idea why it is ignoring the redirect packet?

10:35

10:36

It may be sent too late? Or maybe an "unknown" (which didn't exist yet on that version, the sys.5f4d....etc) packet can somehow force the client to ignore it? (edited)

10:40

Ok confirmed, the issue is the unknown packet

10:41

So yes, I must do a per-version check

Ryozuki

@Learath2 do u wanna continue working on https://github.com/ddnet/ddnet/pull/2754 ?

No. I think it belongs outside the client, in a launcher

edg-l

[ddnet/ddnet] Pull request closed: #2754 Check for the integrity of the data directory

Davide

Ok confirmed, the issue is the unknown packet

This is curious. The unknown packet should get ignored just fine

Learath2

This is curious. The unknown packet should get ignored just fine

It is not

10:50

Every next packet is ignored too after it

10:50

Redirect included

10:51

Actually, not every packet, I managed to be able to close the connection

That's very weird. Would maybe reordering them help?

Learath2

That's very weird. Would maybe reordering them help?

What do you mean?

10:52

Sending the unknown packet (for this version) after the redirect?

Yes. But then I guess the new client might behave weird?

Exactly

10:52

10:53

I'm basically catching the old clients which don't have a new function (the reconnect), and if that's the case I send the old function (the redirect)

10:53

But after I send the reconnect (which is a new function), the old clients just ignore the redirect

That's very weird. Why would a previous packet that's unknown influence future packets parsing? :/

I have no idea

10:57

I tried without the unknown packet and it worked

10:57

With the unknown packet -> not working

Tater

I don't think they're in it for the money

DeepSeek R1 is also MIT licensed

Davide

I tried without the unknown packet and it worked

I might investigate it later. Can you tell me the uuids for the two messages and whether you were sending them in one packet or two?

Sure, single packets

10:59

Reconnect = 5f4d5db739473711b04e07a1ff23c970 (the unknown packet) Redirect = 4efe406a777433f1bfde1806ff6d1528

Tater

is there anything new in this video?

It's talking about all these ai assistants everyone is adding to their devices that have full access to the screen. No amount of e2e will fix that

I could see individual terminal sessions being fed to an assistant with that assistant running locally

Idk I still haven't found much use for those but they're being pushed pretty hard. I only use the assistant if I'm driving

But I'm not going to feed everything I'm doing even when the model is on my own computer

11:11

I don't use them either, I'm just thinking about how much of my data I'd be willing to pour into a magic black box

Learath2

I might investigate it later. Can you tell me the uuids for the two messages and whether you were sending them in one packet or two?

Well, for now, I ended by doing a simple silent disconnect, if you can find the reason why it doesn't work poke me

1

My desktop

Time to clean up

I like the wtf.pcap

11:27

wtf is always reasonable

Avolicious

I like the wtf.pcap

HAHAHAHA

I have a fuckyou.pcap from 2016

Learath2

I have a fuckyou.pcap from 2016

Do you still know what its about?

Avolicious

Do you still know what its about?

Yeah. It's the dump from a dos attack, used it to craft iptables rules

I found a cryhard.pcap from 2018

FIFO is a file. You can set the config to a file name or path such as sample.fifo and then just write to that file. All lines written to that file will be executed as rcon commands.

Replying to @IgnoringLynx How sv_input_fifo works, and how to use it on VDS

Learath2

Yeah. It's the dump from a dos attack, used it to craft iptables rules

Oh, nice. Good old days where DoS was a problem

I also think there is not. But we could add some if you need it. Sounds like interesting info.

Replying to @Davide There is any way to differentiate a normal connect packet from a dummy c…

11:38

That should not cause issues. Just make sure you use the correct sequence numbers. Both of the messages are vital so if you send two of them you need to increment the sequence number.

Replying to @Davide Ok confirmed, the issue is the unknown packet

chillerdragon

That should not cause issues. Just make sure you use the correct sequence numbers. Both of the messages are vital so if you send two of them you need to increment the sequence number.

The sequence number being wrong would make more sense

11:40

With the next packet being ignored too

Since innocent Davide only ever sent one non control message he got away with not worrying about those damn sequence numbers hehe

Ryozuki

any opinions on this https://github.com/ddnet/ddnet/pull/1264

How sv_input_fifo works, and how to use it on VDS

11:47

?

I already told you @IgnoringLynx

chillerdragon

I already told you @IgnoringLynx

Sorry, I didn't notice.

chillerdragon

I already told you @IgnoringLynx

Can I record them in real time? if so, I have to delete all the lines that were entered before

chillerdragon

That should not cause issues. Just make sure you use the correct sequence numbers. Both of the messages are vital so if you send two of them you need to increment the sequence number.

Oh, thanks, I'll check that!

def-

[ddnet-web:master] 1 new commit

01e63b6 Add 100 € donation by TsFreddie - def-

@TsPiggy Thanks! ^

1

IgnoringLynx

How sv_input_fifo works, and how to use it on VDS

ppl who reply to a question with a totally unrelated thing are amongst the things i hate most

12:02

npc behaviour xd

12:03

not even "please"

Ryozuki

ppl who reply to a question with a totally unrelated thing are amongst the things i hate most

Ryozuki

ppl who reply to a question with a totally unrelated thing are amongst the things i hate most

Can you send me the recipe for making a stew? ~~Please~~ (edited)

12:06

/jk

Davide

/jk

Beef Stew Recipe 1. Heat oil in a pot, brown 500g beef cubes. 2. Add 1 diced onion, 2 minced garlic cloves; sauté. 3. Stir in 2 tbsp flour, then add 4 cups beef stock. 4. Add 3 chopped carrots, 2 diced potatoes, 2 celery stalks, salt, and pepper. 5. Simmer covered for 1.5-2 hours, stirring occasionally. 6. Adjust seasoning and serve.

Oh no, AI is here

Idk what you are talking about

Learath2

Beef Stew Recipe 1. Heat oil in a pot, brown 500g beef cubes. 2. Add 1 diced onion, 2 minced garlic cloves; sauté. 3. Stir in 2 tbsp flour, then add 4 cups beef stock. 4. Add 3 chopped carrots, 2 diced potatoes, 2 celery stalks, salt, and pepper. 5. Simmer covered for 1.5-2 hours, stirring occasionally. 6. Adjust seasoning and serve.

To thank you I'll also give you a receipe

12:09

Davide

Click to see attachment 🖼️

I do this every morning

Learath2

I do this every morning

The same steps?

Yes

Oh wow nice

The second step 2 is essential

I agree!

Learath2

Beef Stew Recipe 1. Heat oil in a pot, brown 500g beef cubes. 2. Add 1 diced onion, 2 minced garlic cloves; sauté. 3. Stir in 2 tbsp flour, then add 4 cups beef stock. 4. Add 3 chopped carrots, 2 diced potatoes, 2 celery stalks, salt, and pepper. 5. Simmer covered for 1.5-2 hours, stirring occasionally. 6. Adjust seasoning and serve.

just put raw beef in a furnace for 10 seconds

12:11

works like a charm

Koll Potato

just put raw beef in a furnace for 10 seconds

That doesnt make stew. It makes Steak

ima try deepsek locally wiht ollama

Learath2

That doesnt make stew. It makes Steak

ah yeah

https://ollama.com/library/deepseek-r1

deepseek-r1

DeepSeek's first-generation of reasoning models with comparable performance to OpenAI-o1, including six dense models distilled from DeepSeek-R1 based on Llama and Qwen.

Ryozuki

ima try deepsek locally wiht ollama

Do you have enough memory for it even?

Learath2

Do you have enough memory for it even?

can i run the 32b with a 3080?

12:12

xd

12:13

https://www.reddit.com/r/selfhosted/comments/1iblms1/running_deepseek_r1_locally_is_not_possible/

From the selfhosted community on Reddit

Explore this post and more from the selfhosted community

12:13

hm idk

12:13

I don't know why Ollama decides to call these models Deepseek R1, but it's problematic. Running the actual Deepseek R1 in q4 requires more than 400GB of VRAM or RAM depending on how long are are willing to sit there waiting for an answe

Idk if there is a way to swap pages in an out while inferring

12:13

But you definitely can't fit the model on your single 3080

I have a few servers with 1TB RAM, I may try it!

man

Davide

I have a few servers with 1TB RAM, I may try it!

Cpu inference is usually muuch slower. It might be unfeasible

when consumer hardware ai that is okayish

12:15

i always had the idea to add a local ai translation service to ddnet

Learath2

Cpu inference is usually muuch slower. It might be unfeasible

EPYC 9474F, should be decent

but i gues its not possible yet

The people who ran it at home successfully mostly all used a swarm of Apple Silicon macs

Davide

EPYC 9474F, should be decent

I maaged to run a 100B+ and it was pretty fast

Davide

EPYC 9474F, should be decent

doesnt epyc have like a fpga

12:15

https://www.tomshardware.com/news/amd-to-fuse-fpga-ai-engines-onto-epyc-processors-arrives-in-2023

AMD to Fuse FPGA AI Engines Onto EPYC Processors, Arrives in 2023

AMD to arm its chips with FPGAs

Nice

Learath2

The people who ran it at home successfully mostly all used a swarm of Apple Silicon macs

(With the gpu having access to all of the unified memory on those)

Ryozuki

i always had the idea to add a local ai translation service to ddnet

I have a draft of non local translate in tater

yeah non local being the keyword

12:18

ffxiv has a plugin that translated the game to spanish using deepl

Idk about the legality but that also means you could add Google translate and all the others

(oddly enough ffxiv is not in spanish and doesnt plan to, ignoring the wide userbase that exists in spain and south america)

Ryozuki

i always had the idea to add a local ai translation service to ddnet

I never felt like translation belonged in the client

(but it has french and german translations)

12:19

square enix: lets ignore the second most spoken language

Ryozuki

(oddly enough ffxiv is not in spanish and doesnt plan to, ignoring the wide userbase that exists in spain and south america)

That's bizarre. Spanish speaking market is massive (edited)

Learath2

I never felt like translation belonged in the client

Copying from console is exceedingly slow and not worth when half the messages are derogatory

12:19

So I end up just not understanding russian which I could

xd

Ryozuki

i always had the idea to add a local ai translation service to ddnet

yes make ddnet require 1.3tb of vram xD New steam requirements

xd

12:22

well

12:22

maybe with the 3k$ nvidia pc

You can get good enough or better than nothing with small models

12:22

Which even run in reasonable times on CPUs

Smaller models are easy to run

12:23

But ddnet would depend on ollama which is idk (edited)

1

Well it can be an optional dependency

12:24

Like plz add ollama to path to access this feature

Solly

Like plz add ollama to path to access this feature

So much effort for mods explaining to ppl

12:24

We need a wiki page then

Teero

So much effort for mods explaining to ppl

You can steal the ollama install guide for which probably exists

Solly

You can steal the ollama install guide for which probably exists

I guess so

But tech illiterate people are so with ot without a guide

$ollama

This command was not found.

Ryozuki

Click to see attachment 🖼️

orphan rule violation justatest

https://ollama.com/library/mistral-small

mistral-small

Mistral Small is a lightweight model designed for cost-effective use in tasks like translation and summarization.

People are going to jailbreak chat so fast xD

ima try this

Imagine ppl write smth in russian and it gives you instructions on how to build a bomb

12:33

can someone say smth in german

12:33

to verify

Ryozuki

Click to see attachment 🖼️

what even is that

oops wrong person

xd

jxsl13

what even is that

mistral-small

12:35

lol

we need better manga translation AIs

have u tried libretranslate?

12:36

or is it too bad (edited)

Ryozuki

Click to see attachment 🖼️

i can translate these for free if u want

12:36

12:36

omg what a troll

I tried some other one, forgot the name. cleaning was cool, the translation and typesetting was meh.

i told it to not say anything but translation

cleaning up speech bibbles and stuff

12:38

bibbles the clown

ollama looks nice

12:42

       |                   ^
/home/runner/work/ddnet/ddnet/src/base/log.cpp:526:17: error: 'atomic_load_explicit<ILogger>' is deprecated: use 'std::atomic<std::shared_ptr<T>>' instead [clang-diagnostic-deprecated-declarations,-warnings-as-errors]
  526 |         pLogger = std::atomic_load_explicit(&m_pLogger, std::memory_order_relaxed);

12:42

c++20 deprecates lot of stuff lol

@deen do u know if c++20 works in old build servers? or maybe its time to upgrade the old build servers?

ddnet/src/game/server/entities/character.cpp:23:1: error: pointer arithmetic on polymorphic object of type 'CCharacter' can result in undefined behavior if the dynamic type differs from the pointer type [bugprone-pointer-arithmetic-on-polymorphic-object,-warnings-as-errors]
   23 | MACRO_ALLOC_POOL_ID_IMPL(CCharacter, MAX_CLIENTS)
      | ^
/data2/edgar/ddnet/src/game/alloc.h:66:20: note: expanded from macro 'MACRO_ALLOC_POOL_ID_IMPL'
   66 |                 dbg_assert(Id == (POOLTYPE *)pObj - (POOLTYPE *)gs_PoolData##POOLTYPE, "invalid id"); \
      |                 ~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/data2/edgar/ddnet/src/base/system.h:82:38: note: expanded from macro 'dbg_assert'
   82 | #define dbg_assert(test, msg) assert(test)
      |                               ~~~~~~~^~~~~
/usr/include/assert.h:100:27: note: expanded from macro 'assert'
  100 |      (static_cast <bool> (expr)                                         \
      |                           ^~~~

Ryozuki

@deen do u know if c++20 works in old build servers? or maybe its time to upgrade the old build servers?

looks like ubuntu-20.04 works and its 5 years old

Ryozuki

https://ollama.com/library/deepseek-r1

14:20

gm devs

14:20

and others

who will host this ai for this chat, to auto answer questions in #questions (edited)

14:23

and train it with the data of this server ofc xd

@MilkeeyCat i think im gonna make a writer api for qbe

14:23

in rust

14:24

to output qbe ir

14:24

https://c9x.me/compile/doc/il.html

14:24

oh it was milkey or another person

14:24

i forgor

this already exists https://github.com/garritfra/qbe-rs

GitHub - garritfra/qbe-rs: QBE IR in natural Rust data structures

QBE IR in natural Rust data structures. Contribute to garritfra/qbe-rs development by creating an account on GitHub.

ohh

Ryozuki

@MilkeeyCat i think im gonna make a writer api for qbe

I'm gonna contribute to i3 greenthing

Jupstar ✪

who will host this ai for this chat, to auto answer questions in #questions (edited)

if this happens im gonna try and jailbreak it

MilkeeyCat

I'm gonna contribute to i3 greenthing

epyc

14:27

what will u add

probably will try to fix some bug

can u make it tile better

pf, i can make it tile worse

Koll Potato

this already exists https://github.com/garritfra/qbe-rs

doesnt look that complete

yea, when i did some stuff with qbe i just copied the source of qbe rs and adjusted for my own needs

What? You don’t have to delete lines. Just give it a try

Replying to @IgnoringLynx Can I record them in real time? if so, I have to delete all the lines th…

@Robyt3 https://github.com/ddnet/ddnet/commit/8e90bc443556591c9ba05b896ffc461769e9fce2 seems to have broken something

Fix crash with sv_dnsbl_vote 1 and debug dummies · ddnet/ddnet@8e...

In the CServer::DistinctClientCount function a check for the address being nullptr was missing in the inner loop. Before #9591 this was comparing addresses from uninitialized memory instead. Du...

Learath2

@Robyt3 https://github.com/ddnet/ddnet/commit/8e90bc443556591c9ba05b896ffc461769e9fce2 seems to have broken something

You mean that commit in particular or the series of fixes? This would only affect sv_dnsbl_vote 1 since DistinctClientCount is only used once

I haven't diagnosed it entirely but this is the most recent change in the backtrace

16:38

It's crashing all the servers, there is a core in RUS2:/home/teeworlds/servers/core_crash

And servers are already on this commit? Could otherwise be the opposite that the commit fixed this crash

I'm guessing it has something to do with the ClientIngame(i) being gone

Robyt3

And servers are already on this commit? Could otherwise be the opposite that the commit fixed this crash

That's a good question, idk, I assumed it was a recent deploy since it wasn't crashing yesterday

Probably also possible that is crashed without debug dummies (without the fix), probably happened also when there are large gaps in the range of valid clients

In the backtrace it seems it was trying to compare a client with m_State == STATE_EMPTY

16:41

cid 9 was STATE_EMPTY, cid 10 was STATE_INGAME, they got passed to net_addr_comp_noport

The ClientIngame check should ensure CClient::STATE_INGAME, otherwise the address entry in apAddresses is nullptr which is checked in the second loop

(gdb) print GIT_SHORTREV_HASH
$10 = 0x564e54c02b5d "9f306fcc712de032"

Looks like it doesn't include the fix yet

Yeah, so maybe a redeploy might fix it

16:43

I wonder why it wasn't crashing yesterday though

Was sv_dnsbl_vote always enabled?

Probable

16:49

Let me compile a new one and try deploy the latest commit

16:51

We really need to start monitoring these better too, we should know about dozens of crashes in a row before people can report it to us

did a server just crash?

16:52

i clicked away the ip but some back in the days server just went connection problems on me

16:53

pretty sure it was this one 45.141.57.22:8328

Automate rollbacks!

deployed the current master, restarted all empty servers

Learath2

We really need to start monitoring these better too, we should know about dozens of crashes in a row before people can report it to us

yeah, would be really useful to get notifications about crashing servers, with crash dump and version

RUS2 was also set up to just keep overwriting a single corefile

@Robyt3 but there are crash dumps right?

16:56

notifcations are nice ish but logs of all crashes and someone who checks it once and then should be enough

yeah, but it would make the process a lot smoother if all the crashes would just be logged in a web console or something

give me ssh ill do it xd

ok

or fix ddnet-scripts repo to correctly reflect prod and ill pr it

ssh@jupeyy

We should just rip out all these messy script mish-mashes and just add some proper monitoring with modern tools

no

I would not even know where to add what as it is currently running

17:04

I don't get how this wasn't happening all the time

17:04

It literally just requires any slot to be empty as far as I can see

what so server full is crash?

17:08

gotta start kicking players then

I guess it was getting safe addresses that pointed somewhere readable atleast (edited)

17:08

but I don't get why that changed recently

Learath2

It literally just requires any slot to be empty as far as I can see

I don't know, I tested the specific case of client in slots 0 and 2 before the fix without debug dummies because that's how I read the code as well, but couldn't get it to crash

This makes no sense to me which is scary

I suppose there was something valid on the stack that net_addr_comp_noport doesn't crash on

17:17

so it was just writing a 0 to a random memory location monkaS

and then calling mem_comp on it

It does look like just luck that we were always getting a readable garbage address

17:24

mh it doesn't seem like the new version is crashing, the servers that didn't restart yet keep crashing though, that's a shame

Shouldn't they use the new version after crashing once?

do ddnet test servers use the latest release?

17:27

if they use latest master, can they have sv_practice_by_default enabled? or at least a toggle in vote menu

@Avolicious when fix BETA fra emote delay

when use discord app?

when bridge kog

no bridge, use app

17:29

easy fix

no app, use bridge

17:29

easy fix

no, bridge is spammy

17:30

can be exploited

no, discord is spammy

banned users can still join

can be exploited

not good

users can be banned

17:30

not good

agree irc best

whatever

go join avo pls

i’ve never seen any repeated spam

and remove emoticon delay

17:30

i might fail run because i think its lag when emote doesnt work

guh

github-merge-queue[bot]

[ddnet:master] 2 new commits

79206b0 Make it easier to download a file and have If-Modified-Since support - heinrich5991 923859f Merge pull request #9574 from heinrich5991/pr_ddnet_http_easier_file_download - Robyt3

18:21

Robyt3

[ddnet/ddnet] Pull request closed: #9574 Make it easier to download a file and have If-Modified-Since support

MilkeeyCat

probably will try to fix some bug

After an hour of looking at different issues, I didn't find one to work on feelsbadman

MilkeeyCat

After an hour of looking at different issues, I didn't find one to work on feelsbadman

Editor bucket fill tool

19:19

Or https://discord.com/channels/252358080522747904/1325143766360653855

MilkeeyCat

After an hour of looking at different issues, I didn't find one to work on feelsbadman

https://discord.com/channels/252358080522747904/757720336274948198/1332334328029646859

i got a crash in DistinctClientCount() in ddnet-insta is that coming from ddnet? https://github.com/ddnet-insta/ddnet-insta/issues/247

crash in NumConnectedIps · Issue #247 · ddnet-insta/ddnet-insta

#0 0x0000555555856a28 in net_addr_comp_noport (a=0x7ffff59c89ac, b=0x5c7000005c5) at /home/chiller/git/ddnet-insta/src/base/system.cpp:1074 #1 0x00005555556d946a in CServer::DistinctClientCount (th...

ChillerDragon

i got a crash in DistinctClientCount() in ddnet-insta is that coming from ddnet? https://github.com/ddnet-insta/ddnet-insta/issues/247

yeah, should be fixed again on master

ah nice ok lemme merge

19:54

fixed by 8e90bc443556591c9ba05b896ffc461769e9fce2 ?

https://github.com/ddnet/ddnet/pull/9600

Fix crash with sv_dnsbl_vote 1 and debug dummies by Robyt3 · Pull...

In the CServer::DistinctClientCount function a check for the address being nullptr was missing in the inner loop. Before #9591 this was comparing addresses from uninitialized memory instead. Duplic...

ah yes thanks

https://www.tomshardware.com/software/linux/facebook-flags-linux-topics-as-cybersecurity-threats-posts-and-users-being-blocked

Facebook flags Linux topics as 'cybersecurity threats' — posts and ...

DistroWatch is one of the largest affected organizations.

19:56

soon probably, Rust also banned as cybersecurity breaks

@Jupstar ✪ my sister is getting the 5090 and im getting her 4090 (edited)

20:01

Ryozuki

@Jupstar ✪ my sister is getting the 5090 and im getting her 4090 (edited)

nice, then u can run 32b model

20:01

but i ofc also take it

20:01

np

20:01

send your old gpu to @Teero pls

Ryozuki

@Jupstar ✪ my sister is getting the 5090 and im getting her 4090 (edited)

whats the difference justatest

if you are rich, any % more perf is what matters

20:04

xd

Ryozuki

@Jupstar ✪ my sister is getting the 5090 and im getting her 4090 (edited)

What is she even doing with a 5090 dam

she earns 15k month

20:13

she can buy it

20:13

i get the 4090

20:13

im not gonna ask

Ryozuki

she earns 15k month

that is some insane income, she should start investing a lot of it

she... doesnt care XD

If I was making 15k a month after tax I'd be investing like 10k

20:14

Then in only 30 months you have enough to get 2k from interest every month for the rest of your life

Learath2

What is she even doing with a 5090 dam

sounds good for playing the sims

needs low latency brush strokes in krita justatest

xd

20:17

she plays lot of aaa games

unironically the only aaa game that looks interesting to play is RDR2

Elden Ring was fun

anyone tried ff7 rebirth

Jupeyy

[ddnet-rs:main] 2 new commits

d0c92f8 Fix console error tilde not showing if error range is outside of msg text. - Jupeyy 0fa9d0f Merge pull request #92 from Jupeyy/pr_fix_error_tilde_console - Jupeyy

Robyt3

[ddnet/ddnet] Pull request opened: #9609 Consistently use SDL hint constants from SDL_hints.h

Checklist

[X] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test (especially base/) or added coverage to integration test
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memcheck](https://github.com/ddnet/ddnet/#using-addresss...

I'd rather go to cinema than play some aaa

Invest in what?

Replying to @Learath2 Then in only 30 months you have enough to get 2k from interest every mon…

Jupstar ✪

send your old gpu to @Teero pls

Yes pls @Ryozuki

@Teero: hi gores pro

no i give it to my gf

21:51

xd

Teero

Yes pls @Ryozuki

can your pc handle a 4090?

Tater

can your pc handle a 4090?

My cpu is shit xd

i have a 3080 rn

Tater

can your pc handle a 4090?

I can cut out some of the case xd

Just changed to an 3090 :)

Teero

I can cut out some of the case xd

Valid

chillerdragon

@Teero: hi gores pro

Chiller: Hi gores enjoyer

21:53

If not for you the t0 would have not made it.

21:54

Bcs I followed u xd

21:54

jupstar accused me of cheating ;(

no, i only said unfair advantage xD

@Teero: I mean your moves are insane. And you did a fair amount of sus teeworlds bot development so …

his unfair advantage is that he is 17

??

+1

Agism

That's not true!

22:05

xd

chillerdragon

@Teero: I mean your moves are insane. And you did a fair amount of sus teeworlds bot development so …

I stole all the code from deens secret ddnet cheat repos i would never discuss or develop bots!

github-merge-queue[bot]

[ddnet:master] 2 new commits

38aea14 Convert language scripts readme.txt to Markdown - rffontenelle 64e45d5 Merge pull request #9599 from rffontenelle/patch-4 - def-

22:08

def-

[ddnet/ddnet] Pull request closed: #9599 Convert scripts/languages/readme.txt to Markdown

22:08

github-merge-queue[bot]

[ddnet:master] 2 new commits

11b312e Consistently use SDL hint constants from SDL_hints.h - Robyt3 ca1c597 Merge pull request #9609 from Robyt3/Client-SDL-Hint-Cleanup - def-

22:08

def-

[ddnet/ddnet] Pull request closed: #9609 Consistently use SDL hint constants from SDL_hints.h

does anyone feel like reviewing #8432? I would like to implement more input reliability and rewrite #8372 so the game isn't unplayable if you have bad internet.

https://github.com/ddnet/ddnet/pull/8432 https://github.com/ddnet/ddnet/pull/8372

Refactor Server Input Handling by sjrc6 · Pull Request #8432 · ddne...

The motivation for this PR is to make the server side input code easier to understand, and improve the way the input buffer for each client is handled. First change: This PR should allow clients to...

Presend inputs by JSaurusRex · Pull Request #8372 · ddnet/ddnet

The goal of this pr is to improve antiping player prediction 2024-05-17.12-55-33.mp4 This change sends inputs that already arrived at the server to the clients ...

currently if you have packet loss adding more prediction margin makes your game worse https://discord.com/channels/252358080522747904/757720336274948198/1333495919299002449 (edited)

Learath2

Cpu inference is usually muuch slower. It might be unfeasible

It's actually getting quite fast now, you can get speeds comparable or better than GPU if you have DDR5 memory and an over powered EPYC cpu.

chillerdragon

Invest in what?

literally anything, you just need like 7% return

but not really, watch inflation halve your yields

22:49

remove a third of* (edited)

22:49

same

22:50

i was more trying to be witty now i look like a jackass

How much inflation are you even expecting? This isn't Turkey or Venezuela, the average is about 3.3% in the US

22:51

7 - 3.3 = 3.7%, which is 11k per year that you can safely withdraw from your investments without shrinking them

yea

Above 500k invested you are pretty much safe forever unless some global catastrophe happens

yeah

22:54

im not as apt with money as i'd like to be

22:55

so far i have made $29 from investing.

i want 1500 TB of RAM

22:56

* 16 (edited)

you want 24PB of ram

this is the real boss fight

what does it mean

thats how many neurons and synapses the average adult human has

yea

22:58

oh i see you are comparing it to the ram lol

xD

that was lost on me

gotta beat the human brain in everything

how can a neuron/synapse be counted in bytes tho

22:59

is it about the number

yeah i mean the real brain has inifinite precision

22:59

that is kinda unfair

22:59

but lets assume 16 bit fp perform good enough

23:00

tho synapses are bit more complicated actually

23:00

where are our biologists in this chat btw

23:00

we have everything, we need those too

lol

23:02

brb going to medical school

23:06

actually nah

23:06

i think that only ends in disaster for me

Since when is literally anything just 7% return :D

Replying to @Learath2 literally anything, you just need like 7% return

23:34

You should see my portfolio haha that’s full of anything and not very 7%

VWCE has been returning about 7-6% historically, SXR8 9-10%, SPYI 13-14%. Sure past performance does not indicate future performance but getting to around 7% shouldn't be rocket surgery

This is not investment advice, I'm no investor

7% yield for anything is a dream afaik

23:54

any reliable investment is going to be low yield