ishandotsh

[ishandotsh @ishandotsh]

@sidbing It’s a me

[ishandotsh @ishandotsh]

@NousResearch con had lasers

[ishandotsh @ishandotsh]

@witcheer I’m ready to be in you

[ishandotsh @ishandotsh]

@_lyraaaa_ Idk man

[ishandotsh @ishandotsh]

@radmadvlad do you also see it go dumb after like 2 or 3 turns?

[ishandotsh @ishandotsh]

@luckyrobots @Sentdex Will there be a cowboy version

[ishandotsh @ishandotsh]

@pkuhar @dwiamfurry Is there a reason you’re not using the push-push mechanism built into SD card readers?

[ishandotsh @ishandotsh]

@sentientcar W

[ishandotsh @ishandotsh]

@witcheer @sudoingX Tps aside, I’m not really happy w the actual performance of this (almost exactly the same) setup. Posted entire convos here: https://t.co/ibAHKClH0Y

[ishandotsh @ishandotsh]

@sudoingX 3090 24gb. pretty sure it was a bitcoin miner in its past life. I'm running unsloth/Qwen3.6-27B-GGUF:Q4_K_M -c 131072 -ctk q8_0 -ctv turbo3 ~30tok/s as a worker subagent with pi/hermes and planning w gpt-5.5

[ishandotsh @ishandotsh]

@akseljoonas I’m about to have so much fun at my job

[ishandotsh @ishandotsh]

My setup couldn’t resolve a fairly basic typescript import issue running pi with llama.cpp

[ishandotsh @ishandotsh]

Re: People pushing insane tok/s numbers on a 3090, are the models you’re running smart enough to get actual work done?

[Shen Huang @ShenHuang]

3090 Qwen 3.6 27B 本地跑：你需要知道的一切

花一下午把 X 上推荐的 Qwen 3.6 27B 本地部署方案全试了一遍——3090 24GB 上没一个跑得通。

跑通的全是 VRAM 80GB 起步的 setup。

跑出来的人很多都没给完整setup。我花一下午从头踩到尾：

1. DFlash 暂时配不上任何 27B 4bit 优化 z-lab 的 Qwen3.6-27B-DFlash 按Qwen3 训练和16bit。要自己手动去convert，暂时放弃。
2. vLLM 只有nightly 支持DFlash

最简单的动作（3090 24GB + Qwen 3.6 27B + OpenAI 兼容 API）：

编译 llama.cpp 带 CUDA（一行 cmake，sm_86）

下 unsloth/Qwen3.6-27B-GGUF 的 Q4_K_M（17GB，单文件，hf-transfer 十几分钟）

llama-server 一行起服务，端口暴露 OpenAI 兼容协议
LAN 内 Mac、生产机用标准 OpenAI SDK 直连，零改造
实测 35-50 tok/s，单用户日常够用，再快只是吹牛

RTX 3090 (24GB) + Qwen3.6-27B Q4_K_M + llama-server

Decode: 37 tok/s (stable)

Prompt eval: 342-430 tok/s

VRAM: 17.8 / 24 GB

GPU util: 96%

Power: 385W

目前测试能用的：llama-server + unsloth Qwen 3.6 27B GGUF Q4_K_M。

等 Dflash + DDTree 适配 Q4_K_M 版本应该速度能再上一个台阶。

[ishandotsh @ishandotsh]

I love this so much
Beautiful robots in beautiful environments

[Yinghao Xu @YinghaoXu1]

🎉 After one year of teamwork, we are excited to release our 3D foundation model — LingBot-Map!

Unlike DA3/VGGT, LingBot-Map is a purely autoregressive model for streaming 3D reconstruction ⚡
It achieves ~20 FPS on 518×378 resolution over sequences exceeding 10,000 frames — and beyond 🚀

Two key insights behind LingBot-Map:
🔑 Keep SLAM's structural wisdom: build Geometric Context Attention with long-context modeling while maintaining a compact streaming state
🔑 Make everything end-to-end learnable — no optimization, no post-processing

Let's check out our demos 👇

[ishandotsh @ishandotsh]

Oh they’re deleting everything off it

[ishandotsh @ishandotsh]

Official NVIDIA Omniverse discord got hacked lol

[ishandotsh @ishandotsh]

What did they mean by this

[ishandotsh @ishandotsh]

Opus 4.5 suffers from the same issue that 4 faced in that it loves adding its own micro features-that I never asked for-to account for cases that could never arise
Like an intern with adhd

[Justin Murphy @jmrphy]

Claude Code is personal AGI. You can't use this thing for more than a weekend without realizing it's completely over. At first you make a GUI app, OK cool. Then you're like wait, GUIs are a waste of time, let's just make a terminal app. Then you're like wait APPS are a drag, what if I just ask Claude Code to do the thing directly? Works immediately. Then you're like damn, now asking Claude Code to do stuff feels like a drag, can I... have Claude make a system that says this stuff for me, in the order I've been saying it, for the reasons I've been saying it? Maybe it can do all my tab changing and clicking too? And you really think you're going to break it by asking for something ridiculous, and then even this works, and that's when you realize.... It's over. Claude is now building you an agent system, and it works. Recursively self-improving machine intelligence, today, on your laptop. An agent system building you a custom agent system which gets better simply by using it (and lightly nudging it to improve based on what was learned in the session). I was bearish on Agents for a while. Not anymore. I'm talking home-cooked retard agents. My wife surely thinks I've lost it once and for all. You have to actually pull yourself away because every attempt to find its limit fails to hit any limit. It's not even 'addictive', it's just not believable that the next crazy idea could also work. And then it works and you're like f*ck... It's over.

[ishandotsh @ishandotsh]

@DilumSanjaya Very windows media player visualization coded

[ishandotsh @ishandotsh]

@LeRobotHF please let me train policies off of local datasets PLEASE