Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

New work on Training Foundation Models with a Single Algorithm: Paper: Code, Models: This work introduces Unified Training, a framework that unifies pretraining, midtraining, supervised fine-tuning and RL under a single maximum-likelihood objective. Building on MaxRL, it enables static data and RL-generated data to be trained together within the... same system. Empirically, Unified Training improves accuracy and compute efficiency, generalizes better to harder reasoning problems, and reduces forgetting. Interestingly, reusing data across training stages can improve performance even when conventional staged training fails. It also produces models that adapt more effectively to new tasks, leading to stronger continual learning. See an excellent thread by Yuda Song.show more

Russ Salakhutdinov

129,440 subscribers

42,317 views • 7 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Can “chicken” make a base model reason? 🐔 Yes! Why? With the right first tokens, base models can match RL in reasoning. We trace this to learned training data associations and use a simple data edit to make “chicken” elicit reasoning too. blog:

Can “chicken” make a base model reason? 🐔 Yes! Why? With the right first tokens, base models can match RL in reasoning. We trace this to learned training data associations and use a simple data edit to make “chicken” elicit reasoning too. blog:

Sophie Wang

198,863 views • 1 month ago

A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵

A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵

Thomas Fel

41,904 views • 1 day ago

We found a new way to get language models to reason. 🤯 No RL, no training, no verifiers, no prompting. ❌ With better sampling, base models can achieve single-shot reasoning on par with (or better than!) GRPO while avoiding its characteristic loss in generation diversity.

We found a new way to get language models to reason. 🤯 No RL, no training, no verifiers, no prompting. ❌ With better sampling, base models can achieve single-shot reasoning on par with (or better than!) GRPO while avoiding its characteristic loss in generation diversity.

Aayush Karan

280,889 views • 1 year ago

Our progress on computer vision tech to democratize and annotate sports video data for accurate predictions, continues thanks to SkalskiP. Next step is sourcing and building an improved dataset to continue training the cv models for better accuracy and number (object) recognition.

Our progress on computer vision tech to democratize and annotate sports video data for accurate predictions, continues thanks to SkalskiP. Next step is sourcing and building an improved dataset to continue training the cv models for better accuracy and number (object) recognition.

Sportstensor

16,508 views • 1 year ago

Introducing CQN: Coarse-to-fine Q-Network, a value-based RL algorithm for continuous control🦾Initialized with 20~50 demonstrations, it learns to solve real-world robotic tasks within 10 mins of training, without any pre-training and shaped rewards! (1/4)

Introducing CQN: Coarse-to-fine Q-Network, a value-based RL algorithm for continuous control🦾Initialized with 20~50 demonstrations, it learns to solve real-world robotic tasks within 10 mins of training, without any pre-training and shaped rewards! (1/4)

Younggyo Seo

16,438 views • 2 years ago

🚀Thrilled to share what we’ve been building at TRI over the past several months: our first Large Behavior Models (LBMs) are here! I’m proud to have been a core contributor to the multi-task policy learning and post-training efforts. At TRI, we’ve been researching how LBMs can help robots learn faster, better, and more efficiently. The key takeaways: ✅ We built an evaluation pipeline to benchmark LBM performance with real 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 ✅ Pre-training on hundreds of tasks makes models more robust—plus, we can teach new, complex tasks with 80% 𝐥𝐞𝐬𝐬 𝐝𝐚𝐭𝐚 ✅ The bigger and more diverse the pre-training, the better the results Check out our overview video, webpage and paper for more details: ✨ 🌎 📄 We hope this work helps move the field of robotics forward!

🚀Thrilled to share what we’ve been building at TRI over the past several months: our first Large Behavior Models (LBMs) are here! I’m proud to have been a core contributor to the multi-task policy learning and post-training efforts. At TRI, we’ve been researching how LBMs can help robots learn faster, better, and more efficiently. The key takeaways: ✅ We built an evaluation pipeline to benchmark LBM performance with real 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 ✅ Pre-training on hundreds of tasks makes models more robust—plus, we can teach new, complex tasks with 80% 𝐥𝐞𝐬𝐬 𝐝𝐚𝐭𝐚 ✅ The bigger and more diverse the pre-training, the better the results Check out our overview video, webpage and paper for more details: ✨ 🌎 📄 We hope this work helps move the field of robotics forward!

Zubair Irshad

20,518 views • 1 year ago

Announcing our commercial partnership with Booster Robotics Booster builds humanoid robot hardware, OS, and developer tools to make humanoid robots more affordable, reliable, and practical. The partnership centers on using simulation to multiply the value of real-robot data—expanding teleoperation demonstrations into scalable training data across diverse tasks and scenes. This joint effort powers sim-real co-training and foundation-model development, accelerating progress from hardware iteration to deployable robot policies. Together, we're building simulation‑powered data infrastructure for Physical AI — making scalable training data accessible to model developers and the broader robotics ecosystem.

Announcing our commercial partnership with Booster Robotics Booster builds humanoid robot hardware, OS, and developer tools to make humanoid robots more affordable, reliable, and practical. The partnership centers on using simulation to multiply the value of real-robot data—expanding teleoperation demonstrations into scalable training data across diverse tasks and scenes. This joint effort powers sim-real co-training and foundation-model development, accelerating progress from hardware iteration to deployable robot policies. Together, we're building simulation‑powered data infrastructure for Physical AI — making scalable training data accessible to model developers and the broader robotics ecosystem.

Axis Robotics

72,960 views • 3 months ago

Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive. So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.

Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive. So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.

Modal

167,236 views • 24 days ago

Introducing RL Environment Creator Skill Now any one can create RL environments $ npx skills add adithya-s-k/RL_Envs_101 > You can create environments across multiple frameworks like OpenEnv, OpenReward, Verifiers, NemoGym ... > the repo has live working examples of environments that your coding agent can reference > The skill is design to first understand what type of model you are training and create an environment while keeping that in mind ps. There’s a lot more to building RL environments that can be used for training. One major aspect is the data, which this skill can’t directly solve. However, the skill will help with implementing tools, rewards, and other components of an RL environment, making it easier to go from idea to implementation quickly across different frameworks. Let me know if you’d be interested in a detailed, end-to-end blog/tutorial on building an environment and actually training a model for a useful use case.

Introducing RL Environment Creator Skill Now any one can create RL environments $ npx skills add adithya-s-k/RL_Envs_101 > You can create environments across multiple frameworks like OpenEnv, OpenReward, Verifiers, NemoGym ... > the repo has live working examples of environments that your coding agent can reference > The skill is design to first understand what type of model you are training and create an environment while keeping that in mind ps. There’s a lot more to building RL environments that can be used for training. One major aspect is the data, which this skill can’t directly solve. However, the skill will help with implementing tools, rewards, and other components of an RL environment, making it easier to go from idea to implementation quickly across different frameworks. Let me know if you’d be interested in a detailed, end-to-end blog/tutorial on building an environment and actually training a model for a useful use case.

Adithya S K

47,002 views • 5 months ago

Announced by Mark Zuckerberg this morning — today we're releasing DINOv2, the first method for training computer vision models that uses self-supervised learning to achieve results matching or exceeding industry standards. More on this new work ➡️

Announced by Mark Zuckerberg this morning — today we're releasing DINOv2, the first method for training computer vision models that uses self-supervised learning to achieve results matching or exceeding industry standards. More on this new work ➡️

AI at Meta

1,224,667 views • 3 years ago

Canter 2B is a photography-oriented text-to-image model being trained on just a single GPU by data-archetype 🐎 are we entering the indie ai models era? still training, the v0002 checkpoint is available, more to come ▶️ play on spaces

Canter 2B is a photography-oriented text-to-image model being trained on just a single GPU by data-archetype 🐎 are we entering the indie ai models era? still training, the v0002 checkpoint is available, more to come ▶️ play on spaces

Hugging Apps

24,291 views • 2 months ago

Massive update for AI Engineers! Training diffusion models just got a lot easier. dLLM is an open-source library that does for diffusion models what Hugging Face did for transformers. Here's why this matters: Traditional autoregressive models generate text left-to-right, one token at a time. Diffusion models work differently - they refine the entire sequence iteratively, giving you better control over generation quality and more flexible editing capabilities. The problem? Building and training these models required stitching together scattered tools and reimplementing research papers from scratch. Most of the tooling has been scattered and is hard to reproduce. dLLM changes this: It unifies everything you need to train, evaluate, and deploy diffusion language models: ↳ Scalable training with LoRA, DeepSpeed, and FSDP support ↳ Unified evaluation that abstracts away inference complexity ↳ Ready-to-use recipes for pretraining, finetuning, and evaluation The library includes implementations of models like LLaDA and Dream, plus training algorithms like Edit Flows that enable insertion, deletion, and substitution operations. The team just released ModernBERT-Chat models showing you can turn BERT into lightweight chatbots through masked instruction tuning. This is practical and worth exploring. The setup is straightforward. Run locally with Accelerate or scale to multi-node clusters with Slurm. If you're working with language models and want to explore diffusion-based approaches without rebuilding infrastructure, dLLM gives you a production-ready starting point. Link to the repo in the next tweet.

Massive update for AI Engineers! Training diffusion models just got a lot easier. dLLM is an open-source library that does for diffusion models what Hugging Face did for transformers. Here's why this matters: Traditional autoregressive models generate text left-to-right, one token at a time. Diffusion models work differently - they refine the entire sequence iteratively, giving you better control over generation quality and more flexible editing capabilities. The problem? Building and training these models required stitching together scattered tools and reimplementing research papers from scratch. Most of the tooling has been scattered and is hard to reproduce. dLLM changes this: It unifies everything you need to train, evaluate, and deploy diffusion language models: ↳ Scalable training with LoRA, DeepSpeed, and FSDP support ↳ Unified evaluation that abstracts away inference complexity ↳ Ready-to-use recipes for pretraining, finetuning, and evaluation The library includes implementations of models like LLaDA and Dream, plus training algorithms like Edit Flows that enable insertion, deletion, and substitution operations. The team just released ModernBERT-Chat models showing you can turn BERT into lightweight chatbots through masked instruction tuning. This is practical and worth exploring. The setup is straightforward. Run locally with Accelerate or scale to multi-node clusters with Slurm. If you're working with language models and want to explore diffusion-based approaches without rebuilding infrastructure, dLLM gives you a production-ready starting point. Link to the repo in the next tweet.

Akshay 🚀

64,011 views • 11 months ago

Decentralized Diffusion Models power stronger models trained on more accessible infrastructure. DDMs mitigate the networking bottleneck that locks training into expensive and power-hungry centralized clusters. They scale gracefully to billions of parameters and generate photorealistic images with just a week of training on eight independent GPU nodes. They’re easy to implement, adopt DiT hyperparameters directly and outperform standard models FLOP-for-FLOP.

Decentralized Diffusion Models power stronger models trained on more accessible infrastructure. DDMs mitigate the networking bottleneck that locks training into expensive and power-hungry centralized clusters. They scale gracefully to billions of parameters and generate photorealistic images with just a week of training on eight independent GPU nodes. They’re easy to implement, adopt DiT hyperparameters directly and outperform standard models FLOP-for-FLOP.

David McAllister

46,480 views • 1 year ago

As a supporter of the open weights ecosystem, we're proud to be a post-training partner for NVIDIA Nemotron. We post-train Nemotron models for customer use cases, de-risk mainline RL runs on our AC2 platform and training stack, and contribute aggregate workload statistics for inference benchmarking. This is how open models get better, and we're excited to keep working closely with NVIDIA AI.

As a supporter of the open weights ecosystem, we're proud to be a post-training partner for NVIDIA Nemotron. We post-train Nemotron models for customer use cases, de-risk mainline RL runs on our AC2 platform and training stack, and contribute aggregate workload statistics for inference benchmarking. This is how open models get better, and we're excited to keep working closely with NVIDIA AI.

Applied Compute

25,683 views • 3 months ago

What if you kept asking an LLM to "make it better"? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The naive approach to training for K self-improvement steps leads to K times the number of rollout steps per episode. We introduce Exploratory Iteration (ExIt), an RL-based automatic curriculum method that bootstraps diverse training distributions of self-improvement tasks by upcycling the LLM's own responses at previous turns as the starting points for both self-improvement and *self-divergence.* In order to decide what task to train on next, the curriculum prioritizes sampling of partial turn histories that led to higher return variance in its GRPO group (a learnability score that comes for free). This automatic curriculum over the bootstrapped task space teaches the model how to perform iterated self-improvement while only ever training the model on single-step self-improvement tasks. We look at ExIt's impact in both single-turn (contest math problems) and multi-turn (BFCLv3 multi-turn tasks), as well as MLE-bench, where the LLM is run in a search scaffold to produce solutions to real Kaggle competitions. Across these eval settings, we find ExIt produces models with greater capacity for inference-time self-improvement compared to GRPO. Notably, ExIt models can self-improve on test tasks for many more steps than the typical solution depth encountered during training, including a 22% improvement in MLE-bench performance compared to GRPO.

What if you kept asking an LLM to "make it better"? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The naive approach to training for K self-improvement steps leads to K times the number of rollout steps per episode. We introduce Exploratory Iteration (ExIt), an RL-based automatic curriculum method that bootstraps diverse training distributions of self-improvement tasks by upcycling the LLM's own responses at previous turns as the starting points for both self-improvement and self-divergence. In order to decide what task to train on next, the curriculum prioritizes sampling of partial turn histories that led to higher return variance in its GRPO group (a learnability score that comes for free). This automatic curriculum over the bootstrapped task space teaches the model how to perform iterated self-improvement while only ever training the model on single-step self-improvement tasks. We look at ExIt's impact in both single-turn (contest math problems) and multi-turn (BFCLv3 multi-turn tasks), as well as MLE-bench, where the LLM is run in a search scaffold to produce solutions to real Kaggle competitions. Across these eval settings, we find ExIt produces models with greater capacity for inference-time self-improvement compared to GRPO. Notably, ExIt models can self-improve on test tasks for many more steps than the typical solution depth encountered during training, including a 22% improvement in MLE-bench performance compared to GRPO.

Minqi Jiang

41,154 views • 1 year ago

Reinforcement learning should be able to improve upon behaviors seen when training. In practice, RL agents often struggle to generalize to new long-horizon behaviors. Our new paper studies *horizon generalization*, the degree RL algorithms generalize to reaching distant goals. 1/

Reinforcement learning should be able to improve upon behaviors seen when training. In practice, RL agents often struggle to generalize to new long-horizon behaviors. Our new paper studies horizon generalization, the degree RL algorithms generalize to reaching distant goals. 1/

Vivek Myers

79,523 views • 1 year ago

World modeling and imitation learning have largely been considered two disparate worlds. In our recent work, Unified World Models, just accepted to #RSS2025, Chuning Zhu provides a dead-simple unifying solution: just train a joint diffusion model over actions and future states, but with *decoupled* diffusion time steps across these modalities. Manipulating these decoupled time steps then allows for marginalization or conditioning on actions or states; a single model can serve as a policy, forward dynamics model, video prediction model, or inverse dynamics model by simply setting diffusion timesteps carefully. The resulting model can leverage video datasets along with robot training data much more effectively, and shows improved robustness, generalization, and flexibility. This is exciting because it is frustratingly simple, scalable, and shows strong improvement on real-world robotics problems. Please refer to Chuning Zhu 's excellent thread for more details! More details/code can be found on our website and in the paper -

World modeling and imitation learning have largely been considered two disparate worlds. In our recent work, Unified World Models, just accepted to #RSS2025, Chuning Zhu provides a dead-simple unifying solution: just train a joint diffusion model over actions and future states, but with decoupled diffusion time steps across these modalities. Manipulating these decoupled time steps then allows for marginalization or conditioning on actions or states; a single model can serve as a policy, forward dynamics model, video prediction model, or inverse dynamics model by simply setting diffusion timesteps carefully. The resulting model can leverage video datasets along with robot training data much more effectively, and shows improved robustness, generalization, and flexibility. This is exciting because it is frustratingly simple, scalable, and shows strong improvement on real-world robotics problems. Please refer to Chuning Zhu 's excellent thread for more details! More details/code can be found on our website and in the paper -

Abhishek Gupta

11,430 views • 1 year ago