the only PBO Curve Optimizer Undervolting Guide you need in 2026
15:32
the only PBO Curve Optimizer Undervolting Guide you need in 2026
60.9K views1 month ago
YouTubenexify
Small Wins Turn into Big Losses: The Paradox of Micro-Optimizing
0:17
Small Wins Turn into Big Losses: The Paradox of Micro-Optimizing
7.2K views1 month ago
YouTubeAlig Skillen
How a Single int Can Crash Your C++ Code in 2038
4:55
How a Single int Can Crash Your C++ Code in 2038
6.3K views3 weeks ago
YouTubeSuboptimal Engineer
BtR11F5zo3JCJ8C8NipWbg7usfdv2E4gvcWkAJhbbM3vIt’s the first LLM built on a new sub-quadratic sparse-attention architecture (SSA) and the first top-tier model with a massive 12 million token context window.• 52× faster than FlashAttention at 1M tokens• Less than 5% the cost of OpusNormal transformers waste huge amounts of compute checking every possible relationship between words.SubQ only focuses on the ones that actually matter.That’s nearly 1,000× less compute — and a smarter way for AI to scal
1:24
BtR11F5zo3JCJ8C8NipWbg7usfdv2E4gvcWkAJhbbM3vIt’s the first LLM built on a new sub-quadratic sparse-attention architecture (SSA) and the first top-tier model with a massive 12 million token context window.• 52× faster than FlashAttention at 1M tokens• Less than 5% the cost of OpusNormal transformers waste huge amounts of compute checking every possible relationship between words.SubQ only focuses on the ones that actually matter.That’s nearly 1,000× less compute — and a smarter way for AI to scal
2.8M views1 week ago
x.com𝘉𝘪𝘨 𝘑𝘰𝘦𝘭 💚
> be SubQ > notice llms wasting compute on every token pair> realize only a small fraction matters> build SubQ with sub quadratic sparse attention> focus only on useful relationships> cut 1000x useless compute> unlock 12M token context> benchmark it> 52x faster than flashattention at 1m tokens> 5% the cost of opus> llm scaling just changed> the breakthrough isn’t bigger models,it’s smarter attention. #subq, @alex_whedon, @subquadratic
0:21
> be SubQ > notice llms wasting compute on every token pair> realize only a small fraction matters> build SubQ with sub quadratic sparse attention> focus only on useful relationships> cut 1000x useless compute> unlock 12M token context> benchmark it> 52x faster than flashattention at 1m tokens> 5% the cost of opus> llm scaling just changed> the breakthrough isn’t bigger models,it’s smarter attention. #subq, @alex_whedon, @subquadratic
18.6K views1 week ago
x.comCG
nvidia dropped a 120 billion parameter model with an architecture nobody has stress tested yet.so i did what any sane person would do. loaded the full thing unquantized on 2x H200 NVL and gave it one job. build me a GPU marketplace from scratch.coded 769 lines in a single file from one prompt on hermes agent. one shot. served it successfully to localhost.Nemotron Super 120B-A12B. Mamba-2 + MoE + Attention hybrid with LatentMoE routing. 12B active params out of 120B total. 60.2 tok/s at BF16 unqu
0:50
nvidia dropped a 120 billion parameter model with an architecture nobody has stress tested yet.so i did what any sane person would do. loaded the full thing unquantized on 2x H200 NVL and gave it one job. build me a GPU marketplace from scratch.coded 769 lines in a single file from one prompt on hermes agent. one shot. served it successfully to localhost.Nemotron Super 120B-A12B. Mamba-2 + MoE + Attention hybrid with LatentMoE routing. 12B active params out of 120B total. 60.2 tok/s at BF16 unqu
55.2K views1 month ago
x.comSudo su
this is exactly what Slowrun aims to solve. i think the main problem is not the loss function but the learning algorithm:- gradient descent massively underutilizes compute relative to the data it's trained on. i think brains use a lot of compute to search for solutions that generalize, rather than doing computationally cheap but unexploratory curve fitting.- we've already gotten 10x better sample efficiency with Slowrun by pulling on two levers that let us utilize a lot of compute: heavy overpar
2:10
this is exactly what Slowrun aims to solve. i think the main problem is not the loss function but the learning algorithm:- gradient descent massively underutilizes compute relative to the data it's trained on. i think brains use a lot of compute to search for solutions that generalize, rather than doing computationally cheap but unexploratory curve fitting.- we've already gotten 10x better sample efficiency with Slowrun by pulling on two levers that let us utilize a lot of compute: heavy overpar
784.4K views2 weeks ago
x.comSamip
15:22
1200万上下文新模型实测解析：稀疏注意力是突破还是噱头？ | Tim Carambat
1.3K views1 week ago
bilibili量子菠萝_
5:51
效率即智能！颠覆性架构SubQ：1200万token超长上下文，成本仅需1/5，AI Agent新范式
2.7K views1 week ago
bilibili鲲鹏Talk
7:23
Optimizers - EXPLAINED!
149.1K viewsFeb 10, 2020
YouTubeCodeEmporium
3:39
Subcooling Explained
96.8K viewsFeb 6, 2013
YouTubeDavid Jones
48:04
2. Optimization Problems
252.1K viewsMay 19, 2017
YouTubeMIT OpenCourseWare
8:28
Machine Independent Code Optimization | Compiler Design
87.8K viewsAug 9, 2020
YouTubeTHE GATEHUB
4:50
Dynamic Programming | Set 1 (Overlapping Subproblems Property) | GeeksforGeeks
447.6K viewsOct 26, 2016
YouTubeGeeksforGeeks
59:33
Hyperparameter Optimization: This Tutorial Is All You Need
114.6K viewsJul 19, 2020
YouTubeAbhishek Thakur
1:41:55
Deep Learning-All Optimizers In One Video-SGD with Momentum,Adagrad,Adadelta,RMSprop,Adam Optimizers
172.4K viewsNov 3, 2020
YouTubeKrish Naik
10:24
189 - Hyperparameter tuning for dropout, # neurons, batch size, # epochs, and weight constraint
23.5K viewsJan 5, 2021
YouTubeDigitalSreeni
14:56
【人工智能】强化学习Scaling | SemiAnalysis长文 | OpenAI新模型揭秘 | 下一代推理模型o4 | GRPO | 奖励函数 | Reward Hacking | 小模型蒸馏
10.8K views10 months ago
YouTube最佳拍档
0:57
Suboptimal Studies
17.3K viewsJul 25, 2023
YouTubeThoracic Radiology
0:24
Suboptimal Meaning
4.9K viewsApr 26, 2015
YouTubeADictionary
0:50
Superpolynomial & Subexponential?
305.2K views6 months ago
YouTubeEpsilonDelta
1:18:40
22. Graph Optimization
15.7K viewsSep 23, 2019
YouTubeMIT OpenCourseWare
12:39
Premature Optimization
981.8K viewsFeb 10, 2023
YouTubeCodeAesthetic
1:03:07
Lecture 4: Optimization
66.6K viewsAug 10, 2020
YouTubeMichigan Online
26:15
Unconstrained Optimization using SciPy.optimize.minimize package Part I
3.2K viewsJul 14, 2022
YouTubeJohn Wu
1:07:27
Lecture 2 | Convex Optimization II (Stanford)
22.9K viewsJul 9, 2008
YouTubeStanford
1:10:12
Lecture 9 | Convex Optimization II (Stanford)
17.1K viewsJul 9, 2008
YouTubeStanford
2:48
19 overloading a subprogram
6.1K viewsMay 29, 2019
YouTubeOU Education
14:55
Hyperparameter Optimization for Xgboost
126.8K viewsJun 26, 2019
YouTubeKrish Naik
11:26
CUDA Crash Course: Sum Reduction Part 1
22.1K viewsMar 4, 2019
YouTubeNick
See more