@leonardodeheinz.bsky.social
Reposted
This is your December reminder that it's absolutely acceptable to try and get to a set annual book goal by cramming in as many novellas and cookbooks as you can find before Christmas swallows you up! #booksky 💙📚
December 2, 2025 at 7:16 PM
This is your December reminder that it's absolutely acceptable to try and get to a set annual book goal by cramming in as many novellas and cookbooks as you can find before Christmas swallows you up! #booksky 💙📚
Reposted
Whatever you do, don’t burn out. It’s terrible and it’s hard to fully recover, like a knee injury that makes you too scared to try jumping again
October 7, 2025 at 5:38 PM
Whatever you do, don’t burn out. It’s terrible and it’s hard to fully recover, like a knee injury that makes you too scared to try jumping again
Reposted
New Game: Mike Cook Presents SQORP! A Digital Puzzle Game
It's a quick little game about laying tiles in a grid, inspired by a mathematical puzzle.
Play free on itch, in the browser: illomens.itch.io/sqorp #pico8
It's a quick little game about laying tiles in a grid, inspired by a mathematical puzzle.
Play free on itch, in the browser: illomens.itch.io/sqorp #pico8
April 7, 2025 at 10:38 AM
New Game: Mike Cook Presents SQORP! A Digital Puzzle Game
It's a quick little game about laying tiles in a grid, inspired by a mathematical puzzle.
Play free on itch, in the browser: illomens.itch.io/sqorp #pico8
It's a quick little game about laying tiles in a grid, inspired by a mathematical puzzle.
Play free on itch, in the browser: illomens.itch.io/sqorp #pico8
Reposted
JD Vance in the oval office
March 1, 2025 at 11:05 AM
JD Vance in the oval office
Reposted
Reposted
February 17, 2025 at 3:32 AM
Reposted
🚀 VMAS Football just got a major upgrade! ⚽🔥
We've revamped it to make continuous control with multi-agent RL even more dynamic & competitive.
Link: github.com/proroklab/VectorizedMultiAgentSimulator/releases/tag/1.5.0
Here’s what’s new: 🧵👇
#AI #ReinforcementLearning #MARL #MachineLearning
We've revamped it to make continuous control with multi-agent RL even more dynamic & competitive.
Link: github.com/proroklab/VectorizedMultiAgentSimulator/releases/tag/1.5.0
Here’s what’s new: 🧵👇
#AI #ReinforcementLearning #MARL #MachineLearning
February 2, 2025 at 5:34 PM
🚀 VMAS Football just got a major upgrade! ⚽🔥
We've revamped it to make continuous control with multi-agent RL even more dynamic & competitive.
Link: github.com/proroklab/VectorizedMultiAgentSimulator/releases/tag/1.5.0
Here’s what’s new: 🧵👇
#AI #ReinforcementLearning #MARL #MachineLearning
We've revamped it to make continuous control with multi-agent RL even more dynamic & competitive.
Link: github.com/proroklab/VectorizedMultiAgentSimulator/releases/tag/1.5.0
Here’s what’s new: 🧵👇
#AI #ReinforcementLearning #MARL #MachineLearning
Reposted
TAID: Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models
arxiv.org/abs/2501.16937
In our #ICLR2025 paper, we try to push the limits of model distillation! We also trained a SOTA ‘smol’ 🇯🇵LLM that works entirely on-device, and runs inside a web browser.
arxiv.org/abs/2501.16937
In our #ICLR2025 paper, we try to push the limits of model distillation! We also trained a SOTA ‘smol’ 🇯🇵LLM that works entirely on-device, and runs inside a web browser.
この度、新手法「TAID」を用いて学習された小規模日本語言語モデル「TinySwallow-1.5B」を公開しました。
sakana.ai/taid-jp
私たちは、大規模言語モデル（LLM）の知識を効率的に小規模モデルへ転移させる新しい知識蒸留手法「TAID」を開発しました。この手法では、小規模モデルの学習進度に合わせて大規模モデルの知識を転移させることで、効果的な知識転移を実現します。この研究は機械学習分野の国際会議ICLR 2025に採択されました。
論文: arxiv.org/abs/2501.16937
デモ: pub.sakana.ai/tinyswallow/
sakana.ai/taid-jp
私たちは、大規模言語モデル（LLM）の知識を効率的に小規模モデルへ転移させる新しい知識蒸留手法「TAID」を開発しました。この手法では、小規模モデルの学習進度に合わせて大規模モデルの知識を転移させることで、効果的な知識転移を実現します。この研究は機械学習分野の国際会議ICLR 2025に採択されました。
論文: arxiv.org/abs/2501.16937
デモ: pub.sakana.ai/tinyswallow/
January 30, 2025 at 1:21 AM
TAID: Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models
arxiv.org/abs/2501.16937
In our #ICLR2025 paper, we try to push the limits of model distillation! We also trained a SOTA ‘smol’ 🇯🇵LLM that works entirely on-device, and runs inside a web browser.
arxiv.org/abs/2501.16937
In our #ICLR2025 paper, we try to push the limits of model distillation! We also trained a SOTA ‘smol’ 🇯🇵LLM that works entirely on-device, and runs inside a web browser.