Reposted by Matteo Robino
Choudhury and Kim et al., "Accelerating Vision Transformers With Adaptive Patch Sizes"
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
October 22, 2025 at 8:08 PM
Choudhury and Kim et al., "Accelerating Vision Transformers With Adaptive Patch Sizes"
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
Reposted by Matteo Robino
Is 32B-4bit equal to 16B-8bit? Depends on the task
* math: precision matters
* knowledge: effective param count is more important
* 4B-8bit threshold — for bigger prefer quant, smaller prefer more params
* parallel TTC only works above 4B-8bit
arxiv.org/abs/2510.10964
* math: precision matters
* knowledge: effective param count is more important
* 4B-8bit threshold — for bigger prefer quant, smaller prefer more params
* parallel TTC only works above 4B-8bit
arxiv.org/abs/2510.10964
![A scatter plot titled “AIME25 — Total Memory vs. Accuracy (Qwen3)” compares model accuracy (%) against total memory usage (weights + KV cache, in GB) for various Qwen3 model sizes and quantization levels.
Axes:
• X-axis: Total Memory (Weight + KV Cache) [GB] (log scale, ranging roughly from 1 to 100)
• Y-axis: Accuracy (%), ranging from 0 to 75
Legend:
• Colors: model sizes —
• 0.6B (yellow)
• 1.7B (orange)
• 4B (salmon)
• 8B (pink)
• 14B (purple)
• 32B (blue)
• Shapes: precision levels —
• Circle: 16-bit
• Triangle: 8-bit
• Square: 4-bit
• Marker size: context length —
• Small: 2k tokens
• Large: 30k tokens
Main trend:
Larger models (rightward and darker colors) achieve higher accuracy but require significantly more memory. Smaller models (left, yellow/orange) stay below 30% accuracy. Compression (8-bit or 4-bit) lowers memory usage but can reduce accuracy slightly.
Inset zoom (upper center):
A close-up box highlights the 8B (8-bit) and 14B (4-bit) models showing their proximity in accuracy despite differing memory footprints.
Overall, the chart demonstrates scaling behavior for Qwen3 models—accuracy grows with total memory and model size, with diminishing returns beyond the 14B range.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ckaz32jwl6t2cno6fmuw2nhn/bafkreiatqdqcrxckrhsruoka6llaggjdsr3cpgspql5lnn7kttwrx3q7ga@jpeg)
October 15, 2025 at 11:10 AM
Is 32B-4bit equal to 16B-8bit? Depends on the task
* math: precision matters
* knowledge: effective param count is more important
* 4B-8bit threshold — for bigger prefer quant, smaller prefer more params
* parallel TTC only works above 4B-8bit
arxiv.org/abs/2510.10964
* math: precision matters
* knowledge: effective param count is more important
* 4B-8bit threshold — for bigger prefer quant, smaller prefer more params
* parallel TTC only works above 4B-8bit
arxiv.org/abs/2510.10964
Reposted by Matteo Robino
NOUVELLE VIDEO ! Je décortique le cas Luc Julia, le réputé co-créateur de Siri et expert mondial de l'IA, encensé dans les médias et récemment auditionné au Sénat. Le résultat est salé mais c'était nécessaire.
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
Luc Julia au Sénat : autopsie d'un grand N'IMPORTE QUOI
YouTube video by Monsieur Phi
youtu.be
August 11, 2025 at 2:39 PM
NOUVELLE VIDEO ! Je décortique le cas Luc Julia, le réputé co-créateur de Siri et expert mondial de l'IA, encensé dans les médias et récemment auditionné au Sénat. Le résultat est salé mais c'était nécessaire.
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
youtu.be/e5kDHL-nnh4
Reposted by Matteo Robino
Je vous implore de lire ceci.
Eleven-minute race for food: how aid points in Gaza became ‘death traps’ – a visual story
Hundreds of people have died while seeking food since delivery was taken over by the Gaza Humanitarian Foundation in May. But Palestinians facing extreme hunger have no choice but to take the risk
www.theguardian.com
July 23, 2025 at 2:16 PM
Je vous implore de lire ceci.
Reposted by Matteo Robino
Why would you ride in a car driven by a human? Do you have some sort of death wish?
June 28, 2025 at 8:16 PM
Why would you ride in a car driven by a human? Do you have some sort of death wish?
Reposted by Matteo Robino
Last month I did a little experiment.
I wanted to see how the exact same post would perform on both X (Twitter) and Bluesky.
The results were...interesting...
[Thread]
I wanted to see how the exact same post would perform on both X (Twitter) and Bluesky.
The results were...interesting...
[Thread]
March 16, 2025 at 6:08 PM
Last month I did a little experiment.
I wanted to see how the exact same post would perform on both X (Twitter) and Bluesky.
The results were...interesting...
[Thread]
I wanted to see how the exact same post would perform on both X (Twitter) and Bluesky.
The results were...interesting...
[Thread]
Reposted by Matteo Robino
Want strong SSL, but not the complexity of DINOv2?
CAPI: Cluster and Predict Latents Patches for Improved Masked Image Modeling.
CAPI: Cluster and Predict Latents Patches for Improved Masked Image Modeling.
February 14, 2025 at 6:05 PM
Want strong SSL, but not the complexity of DINOv2?
CAPI: Cluster and Predict Latents Patches for Improved Masked Image Modeling.
CAPI: Cluster and Predict Latents Patches for Improved Masked Image Modeling.
Reposted by Matteo Robino
Outstanding Finalist 2: “DINOv2: Learning Robust Visual Features without Supervision," by Maxime Oquab, Timothée Darcet, Théo Moutakanni et al. 5/n openreview.net/forum?id=a68...
DINOv2: Learning Robust Visual Features without Supervision
The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could...
openreview.net
January 8, 2025 at 5:41 PM
Outstanding Finalist 2: “DINOv2: Learning Robust Visual Features without Supervision," by Maxime Oquab, Timothée Darcet, Théo Moutakanni et al. 5/n openreview.net/forum?id=a68...
Reposted by Matteo Robino
XFeat: Accelerated Features for Lightweight Image Matching
code: github.com/verlab/accel...
paper: arxiv.org/abs/2404.19174
project: www.verlab.dcc.ufmg.br/descriptors/...
code: github.com/verlab/accel...
paper: arxiv.org/abs/2404.19174
project: www.verlab.dcc.ufmg.br/descriptors/...
GitHub - verlab/accelerated_features: Implementation of XFeat (CVPR 2024). Do you need robust and fast local feature extraction? You are in the right place!
Implementation of XFeat (CVPR 2024). Do you need robust and fast local feature extraction? You are in the right place! - verlab/accelerated_features
github.com
December 6, 2024 at 1:23 PM
XFeat: Accelerated Features for Lightweight Image Matching
code: github.com/verlab/accel...
paper: arxiv.org/abs/2404.19174
project: www.verlab.dcc.ufmg.br/descriptors/...
code: github.com/verlab/accel...
paper: arxiv.org/abs/2404.19174
project: www.verlab.dcc.ufmg.br/descriptors/...
Reposted by Matteo Robino
We've always been a fan of blueskies.
December 6, 2024 at 1:44 AM
We've always been a fan of blueskies.
Reposted by Matteo Robino
Free speech on twitter:
December 1, 2024 at 2:30 AM
Free speech on twitter:
Reposted by Matteo Robino
My deep learning course at the University of Geneva is available on-line. 1000+ slides, ~20h of screen-casts. Full of examples in PyTorch.
fleuret.org/dlc/
And my "Little Book of Deep Learning" is available as a phone-formatted pdf (nearing 700k downloads!)
fleuret.org/lbdl/
fleuret.org/dlc/
And my "Little Book of Deep Learning" is available as a phone-formatted pdf (nearing 700k downloads!)
fleuret.org/lbdl/
November 26, 2024 at 6:15 AM
My deep learning course at the University of Geneva is available on-line. 1000+ slides, ~20h of screen-casts. Full of examples in PyTorch.
fleuret.org/dlc/
And my "Little Book of Deep Learning" is available as a phone-formatted pdf (nearing 700k downloads!)
fleuret.org/lbdl/
fleuret.org/dlc/
And my "Little Book of Deep Learning" is available as a phone-formatted pdf (nearing 700k downloads!)
fleuret.org/lbdl/
Reposted by Matteo Robino
Inspired by @simonwillison.net llm-docs repo, I did a similar one compiling projects docs into single TXT files that can be fed to LLMs.
Right now, it only has atproto docs but already been useful to me to answer random questions about the project.
github.com/davidgasquez...
Right now, it only has atproto docs but already been useful to me to answer random questions about the project.
github.com/davidgasquez...
GitHub - davidgasquez/docs-to-llmstxt: 🤖 Compile docs into text files for LLMs
🤖 Compile docs into text files for LLMs. Contribute to davidgasquez/docs-to-llmstxt development by creating an account on GitHub.
github.com
November 19, 2024 at 9:48 AM
Inspired by @simonwillison.net llm-docs repo, I did a similar one compiling projects docs into single TXT files that can be fed to LLMs.
Right now, it only has atproto docs but already been useful to me to answer random questions about the project.
github.com/davidgasquez...
Right now, it only has atproto docs but already been useful to me to answer random questions about the project.
github.com/davidgasquez...