𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶

Cyber Enthusiast

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@token_night & then just sharing insight i imagine? if so i'm down

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@ChartNerdTA is this full chart history available on tradingview? or somewhere else?

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@PixelRainbowNFT @Sharpmatik hindsight has been comically difficult to configure. do you have any tips

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@witcheer excluding the personal setup: i have a research assistant agent assigned to a telegram group chat of less than 10 people. it helps with deep dives and documents items, makes connections, supports or breaks claims, etc. an extra set of hands for us

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@witcheer is there a similar process one could do where they have hermes go through its sessions history on a specific topic to pinpoint the issue they've been having with it

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

I like the glm 5+ family because it doesn't pollute the sessions with overly verbose detail , it just goes to work

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@justinbrock @witcheer using qwen3.8 27b. fairly decent, but needs tuned properly with llama.cpp flags or its a no-op

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@justinbrock @witcheer solar pro4(worst one easily, terrible loops, sees almost nothing available to it after a few turns), stepfun 3.7 flash, not very good. hy3 is somewhat reliable but circles a lot. my own local model did a lot better but iffy at times

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@witcheer @justinbrock i'm going to be honest, mine doesn't seem to have that, & i'm literally constantly debugging. I can't tell if its a model or hermes issue. i'm going nuts !!!!

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@Teknium @NousResearch ��� HERMES AGENT 🔥 🗣️🔊
🔥 NOUS RESEARCH 🔥 🗣️🔊

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@Teknium very cool. what were you learning or already knew leading up to it to go from not working in tech -> tech? 🔥

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@kaish3n that's generous. lol

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@zriael @sudoingX hey thanks. i'm using buun llamas template atm which builds off of froggerics if i'm not mistaken. i suspect some of my issues are related to the memory system i'm using tho i haven't fully isolated/proven it yet

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@Teknium nice. discovered recently one profile was using a skill as a journal. lol

[Teknium 🪽 @Teknium]

What's crazy to me is that Fable is blocked from life sciences broadly, nerfed even if you get passed the classifiers and filter level blocks.

The whole point of AGI/ASI is to cure all diseases. Everything else is just nice to haves. But Anthropic wants to close off that path.

I think Anthropic might be the worst company on the planet.

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@Teknium Thanks a lot teknium. Just have to compete with the other users occasionally 😁😁😹

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@sudoingX not sure how anyone is using this setup reliably with the current full prompt re-processing & checkpoint erasure issue for this arch

[𝚙𝚊𝚝𝚛𝚘𝚕.𝚋𝚊𝚗𝚝𝚜 🌶 @patrolbants]

@populartourist https://t.co/9tkMMq6AnD

not offloading, just the specific qwen3.6 & other similar archs. the GH link details it. i'm suspecting it happens under certain conditions considering i've seen zero mention of it except in the GH