rama (@RamaAdrien) profile picture

creating games

France
Joined May 2014
Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)Photo by rama (@RamaAdrien)
Rethinking Neural Thickets Without Weight Perturbations: Neural Thickets perturbs every weight of a language model, keeps the copies that score best on a small labeled set, and has them vote. We show that a random bias on the output logits, with all other weights unchanged, matches or exceeds these gains on a diverse set of tasks. Check out the blog and paper below
Neural Thickets (@yule_gan @phillip_isola): add random noise to an LLM's weights, keep the best copies, let them vote. Reasoning jumps, no training. Why does it work? It's sampling in disguise. Swap noise on billions of weights for one random vector on the logits: same gains 馃У
2
7
87
5,923
Neural Thickets (@yule_gan @phillip_isola): add random noise to an LLM's weights, keep the best copies, let them vote. Reasoning jumps, no training. Why does it work? It's sampling in disguise. Swap noise on billions of weights for one random vector on the logits: same gains 馃У
18
57
10
526
41,662
Instead of relying on pretrained video latent space we study which tradeoff of spatial and temporal resolution is optimal for real time WM. We open source all our models and training recipes so you can easily adapt it for your needs!
Many world models borrow their video autoencoder from text-to-video generators, and with it, a noticeable lag on every action. Today we release CADRE, open CAusal Diffusion REpresentation autoencoders that let real-time world models react fast, on a single GPU 馃У
1
8
31
983
After MIRA on Rocket League, I was very curious to know what it takes to bring interactive world models to the real world. It starts with the video autoencoder: how, where and how much to compress. CADRE is our answer, and it's all open 馃憞
Many world models borrow their video autoencoder from text-to-video generators, and with it, a noticeable lag on every action. Today we release CADRE, open CAusal Diffusion REpresentation autoencoders that let real-time world models react fast, on a single GPU 馃У
1
5
22
640
Many world models borrow their video autoencoder from text-to-video generators, and with it, a noticeable lag on every action. Today we release CADRE, open CAusal Diffusion REpresentation autoencoders that let real-time world models react fast, on a single GPU 馃У
6
32
4
160
9,251
New blog post about the Samuel text-to-speech, with 1,170 words and five (5) never-seen-before animations!
6
15
1
164
7,892
Introducing REKI, a 2.4B vision-language-action model pretrained on public data only. With a simple design, it rivals larger models trained on in-house data with complex multi-stage pipelines, setting a new Pareto frontier between performance and size. kyutai.org/reki
9
41
9
262
15,297
congratulations to @_Anthea___ for leading such a useful important robotics open-source project!!
Introducing REKI, a 2.4B vision-language-action model pretrained on public data only. With a simple design, it rivals larger models trained on in-house data with complex multi-stage pipelines, setting a new Pareto frontier between performance and size. kyutai.org/reki
5
150
Finally, a text-to-speech for the 80s.
42
197
26
3,153
157,377
Replying to @grok
@grok I'm confused what is happening in this clip?
Replying to @teslayoda
We鈥檙e making interesting progress here. Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human. Skill at Counter-Strike and other fast games is good too. League is also in training. Goal is to generalize across all games.
4
11
1
38
2,588
actual footage of how the 1v1s between GI鈥檚 models and whatever Elon is cooking will go (yes these are our vision based models)
Replying to @teslayoda
We鈥檙e making interesting progress here. Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human. Skill at Counter-Strike and other fast games is good too. League is also in training. Goal is to generalize across all games.
29
27
19
212
76,626
Tempted to start my Claude plushie business as demand skyrockets 馃
1
2
8
530
Checkout our latest work led by the great @RamaAdrien on representation alignment between unpaired modalities! You can align your autoencoder latent structure to the structure of a text model!
Many generative models work in an autoencoder's latent space; They train faster and generate better when this space is well organized. One way to organize it is distilling a pretrained model into it. But how can it work when the two spaces have different structures?
6
27
1,756
Many generative models work in an autoencoder's latent space; They train faster and generate better when this space is well organized. One way to organize it is distilling a pretrained model into it. But how can it work when the two spaces have different structures?
2
27
2
218
12,360
some autoencoders like mimi in audio or va-vae distill representations in the latent space to make it easier to generate. these works assume that the teacher has a structure that is somewhat shared by the student, which is not necessarily the case!
Many generative models work in an autoencoder's latent space; They train faster and generate better when this space is well organized. One way to organize it is distilling a pretrained model into it. But how can it work when the two spaces have different structures?
1
3
7
356
distilling dino into 1D spaces, or distilling text representations into an image autoencoder doesn't fit their approach. in this short paper with @nico_dufour, @ptrkprz and @david_picard, we explore this and propose a simple structure-agnostic way to tackle this distillation!
1
60
samuel now has a technical writeup from @vvolhejn go check out his wonderful project!
In many cases, Samuel learns to position its virtual mouth as a human would, even though its not explicitly trained to do so.
2
153
We've launched Nivalis Nights today!
128
531
75
6,355
357,158
Meta is making a tamagotchi for its AI? Cute, I've had a few on my desktop for a few weeks 馃悾 Tamaclaudchi is a (cute) Claude session manager to see and monitor all your Claude sessions at once Made in collaboration with @romfbr31 and more features to come with @RamaAdrien
1
5
1
10
620
not only is this very cute, but it is also extremely useful! i've been using some features of tamaclauidchi for a few weeks now and it's been really handy. stay tuned!!
Meta is making a tamagotchi for its AI? Cute, I've had a few on my desktop for a few weeks 馃悾 Tamaclaudchi is a (cute) Claude session manager to see and monitor all your Claude sessions at once Made in collaboration with @romfbr31 and more features to come with @RamaAdrien
2
110