Mathew Shen

三尺微命, 一介书生。

[Mathew Shen @MathewShen42]

“科学不是专家学者的专利，而是谁都能做到。不论由谁操作，都能获得相同的结果。” —— 石神千空

[Mathew Shen @MathewShen42]

在北京租房租到麻了...找个安静无甲醛的房子都是奢侈了Orz

[Mathew Shen @MathewShen42]

最近最开心的莫过于看到A➗和cc变成路边一条，并丝滑开通了codex

[Mathew Shen @MathewShen42]

https://t.co/XvmQzbOYje

[Mathew Shen @MathewShen42]

在Codex的帮助下，终于把博客从MkDocs迁移过来。另外就是，有了AI之后，原本显得折腾的东西，现在一句话就搞定了。

[Mathew Shen @MathewShen42]

@xitin842468091 上了班之后，能读大部头的书都是奢侈的享受了Orz

[Mathew Shen @MathewShen42]

个人文学书单

https://t.co/B8ti6nZ3Iu

[Mathew Shen @MathewShen42]

@paradisor092 你这都不给批工伤病假吗

[Mathew Shen @MathewShen42]

@hzqsns Trae还没有那么好用吧:)

[Mathew Shen @MathewShen42]

之前写Agent App时，有个Chat界面流式输出滚动和用户滚动的冲突问题调了好久才搞得足够丝滑，我当时觉着自己还是太菜了，这么简单的问题都不能one shot搞定——直到我用了被吹上天的内部某应用...

[Mathew Shen @MathewShen42]

@hzqsns 杭州是这样的...在地址上写了不送货上门投诉也不给送:)

[Mathew Shen @MathewShen42]

《论如何将初中生物课本的知识迁移到模型调优》：
"你可以参考假说演绎法，来不断推进你的实验"

[Mathew Shen @MathewShen42]

@tualatrix 我觉着这语法实在是丑...拼接怪的感觉

[Mathew Shen @MathewShen42]

根据模型特性来优化硬件的实例。BTW，这个实验室老板是Transformer七子之一。

[hardmaru @hardmaru]

The human brain🧠 is incredibly efficient because it only activates the specific neurons needed for a thought. Modern LLMs naturally try to do this too (> 95% of neurons in feedforward layers stay silent for any given word), but our hardware punishes them for it.

One of the most frustrating paradoxes in deep learning: making a model do less math often makes it run slower. Why? Because unstructured sparsity introduces irregular memory access, and GPUs are built for predictable, dense blocks of math.

We teamed up with @NVIDIA to try to fix this hardware mismatch. Instead of forcing the GPU to adapt to the sparsity, we built a "Hybrid" format that reshapes the sparsity to fit the GPU. Our sparsity format (TwELL) dynamically routes the 99% of highly sparse tokens through a fast path, and uses a dense backup matrix as a safety valve for the rare, heavy tokens.

Through TwELL and a new set of custom CUDA kernels for both LLM inference and training, we translated theoretical sparsity into actual wall-clock speedups: >20% faster training and inference on H100 GPUs, while also cutting energy consumption and memory requirements.

Paper: https://t.co/rqIY9SYBDe
Blog: https://t.co/oRjNbpJKha
Code: https://t.co/FAFaJwpxAJ
⚡️

[Mathew Shen @MathewShen42]

@yusan1776 确实如此

[Mathew Shen @MathewShen42]

@paradisor092 “我这一生，如履薄冰”🤡

[Mathew Shen @MathewShen42]

我觉着财富自由了之后还想做的事就是真的想做的事。

那么，有没有可能在财富自由之前就去做？如果可以，那么什么时候可以呢？

想来想去，感觉还是要这种事要能赚钱才可以 ->赚钱就有压力 –> 有压力可能就没那么喜欢 –> 还是得回去打工赚钱:)

[Mathew Shen @MathewShen42]

假期把维护的两个Rust项目做了一次升级。有了AI之后写起来快多了:)

[Mathew Shen @MathewShen42]

@Orange41324306 恭喜恭喜

[Mathew Shen @MathewShen42]

爱是一项必须加以练习才能习得的能力。真正的爱并不常见，甚至罕有。

[Mathew Shen @MathewShen42]

诗歌回顾:)

[tison @tison1096]

经典回顾。