English
coderofstuff
1.7K posts
coderofstuff
@coderofstuff_
There are 10 types of people in the world - those who know binary and those who don’t | coding DAGKnight https://t.co/WZfxjGEEgq
Joined Nisan 2023
168 Following3.8K Followers
@coderofstuff_ @supertypo_kas @lAmeR_1337 @IzioDev Right, is the fix usually
Hardware or just some software filtering?
English
@PoW_Odie @supertypo_kas @lAmeR_1337 @IzioDev Kraken though in particular imo is the best, and they have great devs keeping their service running well
English
@PoW_Odie The questions are pouring in. You don’t see them but @supertypo_kas @lAmeR_1337 @IzioDev and I do. There are many in the ecosystem that need to adapt.
English
@alwaysaimbig @realvijayk @thekaspaonion Might be helpful: @coderofstuff/kaspa-where-do-i-begin-14ee038cb7e8" target="_blank" rel="nofollow noopener">medium.com/@coderofstuff/…
English
@alwaysaimbig @realvijayk @thekaspaonion The idea is to use a consensus algorithm (ghostdag in this case) to induce a linear ordering over the blocks in the dag then process transactions according to that order. That’s the intuition behind it. Dive deeper with some of the links people have posted.
English
PROOF KASPA IS THE FINAL FORM OF MONEY New Video!!! Kaspa just did something insain!
PLEASE SHARE!!!!
English
@Kaspa_HypeMan @supertypo_kas It’s hero, but really @supertypo_kas deserves to be paid for carrying the burden of infra that these exchanges depend on.
English
@NikolaGalilei @elldeeone That took a while. What were you downloading it with, a carrier pigeon?
English
@coderofstuff_ @elldeeone Finally downloaded.
Update, this is how I code with this model (qwen27b/claude/opus)
GIF
English
coderofstuff retweeted
For those who aren't tuned into whats going on in the ai world, things are really heating up for self hosting.
You can run ai on a gpu from 5 years ago and it works a lot better than you'd think.
I'm fascinated by this, not because of what it's capable of doing right now (tbh it still doesn't suit my workflow although i might run it for non-primary tasks as an agent) but for what will be possible in the next 6-12 months.
Yes, to achieve frontier-level quality, with context limits, etc, nothing beats oai or anthropic (for now).. but it feels like the speed at which lower and lower resources are required to get results that just a year ago *we paid oai or anthropic for* are growing at a phenomenal rate.
Zooming out for a moment, the IT industry is experiencing growth at an exponential rate, similar to the dotcom boom but on crack. The troglodytes who oppose AI, with their handcrafted artisanal lines of code, will undoubtedly be left behind as new and experienced coders boost their capabilities.
Speaking with many coders, those who code as their profession and have genuinely embraced the change, they all tell me the same thing: AI has enabled them to achieve things they previously thought impossible, giving them more time to focus on the coding tasks that truly engage them.
If the bald coinbase man or the binance convict is to be listened to, they and many others in the industry are predicting that agents will make 1 million times more payments than humans on chain.
With the barrier for entry of personal self hosted AI agents rapidly decreasing, this may just become a reality.
Sudo su@sudoingX
hey if you have a 3060, or any GPU with 8GB or more sitting in a drawer right now, that thing can run 9 billion parameters of intelligence autonomously. and you don't know it yet. 2 hours ago i posted that 9B hit a ceiling. 2,699 lines across 11 files. blank screen. said the limit for autonomous multifile coding on 9 billion parameters is real. then i audited every file. found 11 bugs. exact file, exact line, exact fix. duplicate variable declarations killing the script loader. a canvas reference never connected to the DOM. enemies with no movement logic. particle systems called on the class instead of the instance. fed that list as a single prompt to the same Qwen 3.5 9B on the same RTX 3060 through Hermes Agent. it fixed all 11. surgically. patch level edits across 4 files. no rewrites. no hallucinated changes. game boots. enemies spawn, move, collide. background renders. particles fire. and here's what nobody is talking about. this is a 9 billion parameter model running a full agentic framework. Hermes Agent with 31 tools. file operations, terminal, browser, code execution. not a single tool call failed. the agent chain never broke. most people think you need 70B+ for reliable tool use. this is 9B on 12 gigs doing it clean. the model didn't fail. my prompting strategy did. the ceiling is not the parameter count. the ceiling is how you prompt it. this is not done. bullets don't fire yet. boss fights need wiring. but the screen that was black 2 hours ago now has a full game rendering in real time. iterating right now. anyone with a GPU from the last 5 years should be paying attention to what is happening right now.
English
@elldeeone You know how you can send messages to it through Discord, TG, Signal, etc? Maybe @kasiamessaging can work here too if integrated
English
Sudo su@sudoingX
if you're running Qwen 3.5 on any coding agent (OpenCode, Claude Code) you will hit a jinja template crash. the model rejects the developer role that every modern agent sends. people asked for the full template. here it is. two paths depending on which model you're running: path 1: patch base Qwen's template. add developer role handling + keep thinking mode alive. full command: llama-server -m Qwen3.5-27B-Q4_K_M.gguf -ngl 99 -c 262144 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 --chat-template-file qwen3.5_chat_template.jinja template file: gist.github.com/sudoingX/c2fac… without the patched template, --chat-template chatml silently kills thinking. server shows thinking = 0. no reasoning. no think blocks. check your logs. path 2: run Qwopus instead. Qwen3.5-27B with Claude Opus 4.6 reasoning distilled in. the jinja bug doesn't exist on this model. thinking mode works natively. no patched template needed. same speed, same VRAM, better autonomous behavior on coding agents. weights: huggingface.co/Jackrong/Qwen3… both fit on a single RTX 3090. 16.5 GB. 29-35 tok/s. 262K context.
English
@NikolaGalilei @elldeeone You need to run qwen3.5:27b in that 3090. I do that, it’ll get you 35tok/s
English
@crono_walker @IzioDev An interesting phenomenon could also happen when you have higher bps. Transactions become more propagated via blocks than they are via mempool, since time to inclusion decreases.
English
Transaction selection is weighted random (weighted by fee). A block picking txs to include will select N txs.
If mempool tx count is N, then the block just includes all of them. There is high redundancy here with parallel blocks (all of them try to include all txs)
As mempool tx count increases, the number of redundant txs decreases. Think of it this way, let’s say you have 10 people that need to pick 300 random objects from a bucket with 1000 items in it. In this case, many selections will collide. But if the bucket had 10,000 items in it, the likelihood of collision is smaller.
English
Because Kaspa's BlockDAG structure allows blocks to be mined in parallel, the same transaction naturally appears in multiple blocks. This provides inherent MEV resistance.
Ross 𐤊@crono_walker
@coderofstuff_ @emdin @IzioDev @averagecatdog Built a simple visualizer showing how the same TX gets included in multiple blocks. I might be thinking incorrectly. rossku.github.io/kaspa-tx-sprea…
English
@crono_walker @emdin @IzioDev @averagecatdog > many parallel blocks will include the same TXs
Why do you think that?
English
@coderofstuff_ @emdin @IzioDev @averagecatdog With Dagknight, many parallel blocks will include the same TXs. Currently keeps full TX data per block — no dedup. I wonder if normalizing TX storage (shared TX table + block→txid references) has been discussed? Feels like it could matter for storage and bandwidth.
English
Can confirm -- 25 minutes for the full sync and `GetVirtualChainFromBlockV2` is pure magic. ⚡
Kaspa@Kaspa_BlockDAG
Rusty Kaspa v1.1.0 is out. Faster syncing, less storage, and exchanges/wallets building on Kaspa just got a much easier time of it. The big one for integrators is a new API call that returns chain updates and transaction data together in one go, instead of having to juggle multiple parallel requests. If you've ever tried to integrate Kaspa and cursed at the DAG complexity, this is the fix. Node operators get up to 3x faster sync in the early header stage on some machines, plus lower disk usage. The stratum bridge also shipped as beta if you're running mining infrastructure. One heads-up - DB version bumped to 6. Upgrade is automatic, but you can't roll back to an older version without wiping the DB. github.com/kaspanet/rusty…
English
@sudoingX @mamboussa host shouldn’t need 0.0.0.0 if hermes is connecting from localhost right?
English
llama.cpp is the way. grab the Qwen3.5-9B Q4_K_M.gguf from huggingface, compile llama.cpp with CUDA, and launch with:
./llama-server -m model.gguf -ngl 99 -c 131072 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 --host 0.0.0.0
then install hermes agent and point it at localhost:8080. dm me if you get stuck.
English
Hi @sudoingX im going to install Agent Hermes on my desktop with a 3060 12gb, can you help me to configure it with Llama.cpp or Ollama ?
English
@dionyziz You could say the game has truly changed since then
English
When I was 8 years old and starting programming with QBasic and without Internet, I looked at two examples that came with the interpreter: nibbles (a snake game) and gorillas (angry-birds–like). The code was incomprehensible, but there were comments! I didn't understand English then, but I learned some to read the comments explaining the gameplay. Then I tried editing the comments to change the gameplay, and recompiled, but nothing changed in the game. How excited would little me be to find out that in 2026 you can finally just edit the comments and change the game!
English
Kaspa@kaspaunchained
🎉Good news! 📢 The long-awaited @Ledger integration is complete! Securely manage and store your $KAS on Nano S, Nano S+ and Nano X! Download the #Kaspa app via #LedgerLive and use kasvault.io to interact with your new app. Ledger Guide: support.ledger.com/hc/en-us/artic… (Link to KASVault user guide at the bottom) #L1 #ProofofWork #DigitalSilver #CryptoStorage
QME
@Chris_Hutch7 @study_kaspa @Ledger @coderofstuff_ Wow!! That's amazing! Thank you for letting me know!
English
@sudoingX thanks for this - testing it, looking legit so far
English
if you're running Qwen 3.5 on any coding agent (OpenCode, Claude Code) you will hit a jinja template crash. the model rejects the developer role that every modern agent sends.
people asked for the full template. here it is. two paths depending on which model you're running:
path 1: patch base Qwen's template.
add developer role handling + keep thinking mode alive.
full command:
llama-server -m Qwen3.5-27B-Q4_K_M.gguf -ngl 99 -c 262144 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 --chat-template-file qwen3.5_chat_template.jinja
template file: gist.github.com/sudoingX/c2fac…
without the patched template, --chat-template chatml silently kills thinking. server shows thinking = 0.
no reasoning. no think blocks. check your logs.
path 2: run Qwopus instead.
Qwen3.5-27B with Claude Opus 4.6 reasoning distilled in. the jinja bug doesn't exist on this model. thinking mode works natively. no patched template needed. same speed, same VRAM, better autonomous behavior on coding agents.
weights: huggingface.co/Jackrong/Qwen3…
both fit on a single RTX 3090. 16.5 GB. 29-35 tok/s. 262K context.
Sudo su@sudoingX
if you try to run qwen 3.5 27B with OpenCode it will crash on the first message. OpenCode sends a "developer" role. qwen's template only accepts 4 roles: system, user, assistant, tool. anything else hits raise_exception('Unexpected message role.') and your server returns 500s in a loop. unsloth's latest GGUFs still ship with the same template. the bug is in the jinja, not the weights. no quant update will fix it. the common fix floating around is --chat-template chatml. it stops the crash. it also silently kills thinking mode. your server logs will show thinking = 0 instead of thinking = 1. no think blocks. no chain of thought. you're running a reasoning model without reasoning and the server won't tell you. the real fix: patch the jinja template to handle developer role + preserve thinking mode. add this to the role handling block: elif role == "developer" -> map to system at position 0, user elsewhere else -> fallback to user instead of raise_exception full command with the fix: llama-server -m Qwen3.5-27B-Q4_K_M.gguf -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 --chat-template-file qwen3.5_chat_template.jinja thinking = 1 confirmed. full think blocks. no crashes. that's what's running in the video in the thread below. if you've been using chatml as a workaround, check your server logs for thinking = 0. you might be running half a model.
English
English
@coderofstuff_ @IzioDev @averagecatdog Yes, I managed to set it up locally and it worked. Would be really cool to have it merged!
English