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.

1/ Au revoir, RLVR. New work: EBFT (Energy-Based Fine-Tuning), a post-training method that directly optimizes the long-horizon behavior of model generations, addressing SFT’s deployment-time error amplification without relying on sparse, task-specific rewards.

Sham Kakade

19,739 subscribers

267,650 views • 6 months ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

12 Comments

Sham Kakade6 months ago

2/ The problem: SFT trains on true prefixes, but the deployed model conditions on its own outputs. Minor mistakes can snowball. We measure this gap with a sequence level, feature-matching loss: how far apart in embedding space are model generations and ground-truth completions?

Sham Kakade6 months ago

3/ SFT (red) barely improves over the base model (grey), and RLVR (orange) actually makes it worse. EBFT (blue) directly optimizes this loss and achieves the lowest feature-matching loss at every completion length despite training with rollouts of only 8 tokens.

Sham Kakade6 months ago

4/ How EBFT works: match features of generations to the ground-truth completion. Sample short rollouts, embed them alongside ground-truth with a frozen feature network, then reward each sample based on how well it matches ground-truth features. No reward model, no verifier.

Sham Kakade6 months ago

5/ EBFT delivers strong OOD generalization. When trained on Python, SFT degrades transfer to other languages, while EBFT improves it. And on noisier translation domains, EBFT outperforms both SFT and RLVR– demonstrating its promise beyond clean, verifiable settings.

Sham Kakade6 months ago

6/ Work done with amazing collaborators: Samy Jelassi, @MujinKwun, @rosieyzh, Yuanzhi Li, @nfusi , @du_yilun, @cdomingoenrich

Sham Kakade6 months ago

7/ Tagging folks who might be interested: @ahatamiz1, @YejinChoinka, @natolambert, @_lewtun, @aviral_kumar2, @tengyuma, @Zanette_ai, @johnschulman2, @agarwl_, @ericzelikman

Sham Kakade6 months ago

8/ links: Blog post: Paper: Code: Website:

Jesse Farebrother6 months ago

Can you talk about the relationship to inverse RL? Eg, we had done exactly this in continuous control ( and makes me wonder how feature matching compares with IQ-learn for LLMs as seen here:

Boyuan (Nemo) Chen6 months ago

Error amplification in SFT is painfully real in production. Small distribution gaps snowball fast on long generations. The reward-free angle here is compelling, but how does the energy function handle tasks with multiple valid output paths? That's usually where these approaches get hairy.

0x_Vivek6 months ago

sft drift is real. optimizing long-horizon without relying on clunky reward models is the truth. rip rlvr, you won't be missed.

Ingmar6 months ago

really like this direction. reward design for long-horizon tasks is genuinely painful and most approaches just try to make the reward better. sidstepping it is a different bet entirely.

BlakeTheCoder6 months ago

Energy-based for long-horizon. Clever fix for SFT's error snowball. How's the training stability compared to RLVR? Graphs look solid but 1.5B's pretty small.

Related Videos

OpenAI engineer explained every form of fine-tuning in 1 hour 46 minutes - better than $3000 fine-tuning bootcamps. pick your task -> try prompting first -> drop to SFT for style -> run DPO for preference -> run RFT for verifiable rewards -> ship a model 10x cheaper than GPT-5. That loop is why teams that fine-tune beat teams that only prompt on cost and speed. Prompting + SFT + DPO + RFT + OpenAI fine-tuning API - that's the stack. Watch and save it, then pick the right fine-tune for your task this week.

OpenAI engineer explained every form of fine-tuning in 1 hour 46 minutes - better than $3000 fine-tuning bootcamps. pick your task -> try prompting first -> drop to SFT for style -> run DPO for preference -> run RFT for verifiable rewards -> ship a model 10x cheaper than GPT-5. That loop is why teams that fine-tune beat teams that only prompt on cost and speed. Prompting + SFT + DPO + RFT + OpenAI fine-tuning API - that's the stack. Watch and save it, then pick the right fine-tune for your task this week.

h100envy

41,245 views • 2 months ago

New short course on LLMOps! LLMOps (large language model operations) is a rapidly developing field that takes ideas from MLOps (machine learning operations) and specializes them to building and deploying LLM-based applications. In this course, taught by Google Cloud's Erwin Huizenga, you'll learn to use automation to make building, tuning and deploying an LLM-based application less manual and more efficient. You'll learn how to: - Apply supervised fine-tuning to tune an LLM to a specific task - Automate and orchestrate LLM-tuning and deployment by customizing a pre-built tuning pipeline - Apply best practices for preparing training data for supervised fine-tuning of an LLM - Create an LLMOps workflow you can adapt to other LLM-tuning jobs This course doesn't assume any prior MLOps or LLMOps experience. Sign up here to learn about this emerging field!

New short course on LLMOps! LLMOps (large language model operations) is a rapidly developing field that takes ideas from MLOps (machine learning operations) and specializes them to building and deploying LLM-based applications. In this course, taught by Google Cloud's Erwin Huizenga, you'll learn to use automation to make building, tuning and deploying an LLM-based application less manual and more efficient. You'll learn how to: - Apply supervised fine-tuning to tune an LLM to a specific task - Automate and orchestrate LLM-tuning and deployment by customizing a pre-built tuning pipeline - Apply best practices for preparing training data for supervised fine-tuning of an LLM - Create an LLMOps workflow you can adapt to other LLM-tuning jobs This course doesn't assume any prior MLOps or LLMOps experience. Sign up here to learn about this emerging field!

Andrew Ng

221,803 views • 2 years ago

RLVR has become the recipe for agentic post-training. But for Computer-Use Agents, the bottleneck is not the algorithm, it is the data. 🐌 🚀 We introduce CUA-Gym: a scalable, lightweight synthesis engine that turns arbitrary task queries into verifiable RLVR data for computer-use agents. The largest open CUA RLVR dataset to date: 🎯 32,122 verifiable RLVR tasks with programmatic setup scripts + rewards 🌐 110 environments: 16 desktop apps + 94 synthesized mock web apps 🏆 Qwen3.5-based CUA models trained with GSPO reach 72.6% on OSWorld-Verified and 56.6% on WebArena 📄 Paper: 🏠 Homepage: 🤗 Dataset: 💻 Codebase: 🧩 Environments: 🧵[1/6]

RLVR has become the recipe for agentic post-training. But for Computer-Use Agents, the bottleneck is not the algorithm, it is the data. 🐌 🚀 We introduce CUA-Gym: a scalable, lightweight synthesis engine that turns arbitrary task queries into verifiable RLVR data for computer-use agents. The largest open CUA RLVR dataset to date: 🎯 32,122 verifiable RLVR tasks with programmatic setup scripts + rewards 🌐 110 environments: 16 desktop apps + 94 synthesized mock web apps 🏆 Qwen3.5-based CUA models trained with GSPO reach 72.6% on OSWorld-Verified and 56.6% on WebArena 📄 Paper: 🏠 Homepage: 🤗 Dataset: 💻 Codebase: 🧩 Environments: 🧵[1/6]

Bowen Wang

104,564 views • 4 months ago

An exciting new course: Fine-tuning and Reinforcement Learning for LLMs: Intro to Post-training, taught by Sharon Zhou, VP of AI at AMD. Available now at Post-training is the key technique used by frontier labs to turn a base LLM--a model trained on massive unlabeled text to predict the next word/token--into a helpful, reliable assistant that can follow instructions. I've also seen many applications where post-training is what turns a demo application that works only 80% of the time into a reliable system that consistently performs. This course will teach you the most important post-training techniques! In this 5 module course, Sharon walks you through the complete post-training pipeline: supervised fine-tuning, reward modeling, RLHF, and techniques like PPO and GRPO. You'll also learn to use LoRA for efficient training, and to design evals that catch problems before and after deployment. Skills you'll gain: - Apply supervised fine-tuning and reinforcement learning (RLHF, PPO, GRPO) to align models to desired behaviors - Use LoRA for efficient fine-tuning without retraining entire models - Prepare datasets and generate synthetic data for post-training - Understand how to operate LLM production pipelines, with go/no-go decision points and feedback loops These advanced methods aren’t limited to frontier AI labs anymore, and you can now use them in your own applications. Learn here:

An exciting new course: Fine-tuning and Reinforcement Learning for LLMs: Intro to Post-training, taught by Sharon Zhou, VP of AI at AMD. Available now at Post-training is the key technique used by frontier labs to turn a base LLM--a model trained on massive unlabeled text to predict the next word/token--into a helpful, reliable assistant that can follow instructions. I've also seen many applications where post-training is what turns a demo application that works only 80% of the time into a reliable system that consistently performs. This course will teach you the most important post-training techniques! In this 5 module course, Sharon walks you through the complete post-training pipeline: supervised fine-tuning, reward modeling, RLHF, and techniques like PPO and GRPO. You'll also learn to use LoRA for efficient training, and to design evals that catch problems before and after deployment. Skills you'll gain: - Apply supervised fine-tuning and reinforcement learning (RLHF, PPO, GRPO) to align models to desired behaviors - Use LoRA for efficient fine-tuning without retraining entire models - Prepare datasets and generate synthetic data for post-training - Understand how to operate LLM production pipelines, with go/no-go decision points and feedback loops These advanced methods aren’t limited to frontier AI labs anymore, and you can now use them in your own applications. Learn here:

Andrew Ng

133,386 views • 11 months ago

StyleDrop: Text-to-Image Generation in Any Style introduce StyleDrop, a method that enables the synthesis of images that faithfully follow a specific style using a text-to-image model. The proposed method is extremely versatile and captures nuances and details of a user-provided style, such as color schemes, shading, design patterns, and local and global effects. It efficiently learns a new style by fine-tuning very few trainable parameters (less than 1% of total model parameters) and improving the quality via iterative training with either human or automated feedback. Better yet, StyleDrop is able to deliver impressive results even when the user supplies only a single image that specifies the desired style. An extensive study shows that, for the task of style tuning text-to-image models, StyleDrop implemented on Muse convincingly outperforms other methods, including DreamBooth and textual inversion on Imagen or Stable Diffusion. paper page:

StyleDrop: Text-to-Image Generation in Any Style introduce StyleDrop, a method that enables the synthesis of images that faithfully follow a specific style using a text-to-image model. The proposed method is extremely versatile and captures nuances and details of a user-provided style, such as color schemes, shading, design patterns, and local and global effects. It efficiently learns a new style by fine-tuning very few trainable parameters (less than 1% of total model parameters) and improving the quality via iterative training with either human or automated feedback. Better yet, StyleDrop is able to deliver impressive results even when the user supplies only a single image that specifies the desired style. An extensive study shows that, for the task of style tuning text-to-image models, StyleDrop implemented on Muse convincingly outperforms other methods, including DreamBooth and textual inversion on Imagen or Stable Diffusion. paper page:

AK

56,408 views • 3 years ago

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

Andrew Ng

125,146 views • 1 year ago

Excited to share our latest progress on DYNA-1 pre-training! 🤖 The base model now can perform diverse, dexterous tasks (laundry folding, package sorting, …) without any post-training, even in unseen environments. This powerful base also allows extremely efficient fine-tuning to ~100% success on challenging new tasks with as little as 1 hour of data! 🤯 Watch it master two of them: cup stacking & celery chopping on repeat, no failures. 👇

Excited to share our latest progress on DYNA-1 pre-training! 🤖 The base model now can perform diverse, dexterous tasks (laundry folding, package sorting, …) without any post-training, even in unseen environments. This powerful base also allows extremely efficient fine-tuning to ~100% success on challenging new tasks with as little as 1 hour of data! 🤯 Watch it master two of them: cup stacking & celery chopping on repeat, no failures. 👇

Dyna Robotics

68,198 views • 11 months ago

$QVAC SDK will support in 0.9.0 (gonna be release in ~10 days) LoRA fine-tuning directly on-device, letting developers customize LLMs with their own data without sending anything to the cloud. You just load a base model, point it at your training dataset, and get a lightweight LoRA adapter back — all running locally. The fine-tuned model can then be used for inference immediately, with no extra setup. Why it matters: LoRA (Low-Rank Adaptation) fine-tuning lets you specialize a general-purpose language model for your specific use case — like matching a brand's tone, mastering domain terminology, or following a particular output format — using a fraction of the compute a full fine-tune would require. QVAC handles the entire workflow locally: dataset preparation, training with configurable hyperparameters, checkpoint saving, and seamless inference with the resulting adapter. Your data never leaves the device. The developer experience: Fine-tuning with QVAC is as simple as calling "sdk.finetune()" with your dataset and a few hyperparameters. Training runs entirely on your local hardware, produces a compact LoRA adapter file, and supports pause/resume so you can stop a job and pick it back up without losing progress. The result plugs straight into QVAC's inference pipeline — no model conversion, no deployment step, just immediate local completions with your fine-tuned model.$

QVAC SDK will support in 0.9.0 (gonna be release in ~10 days) LoRA fine-tuning directly on-device, letting developers customize LLMs with their own data without sending anything to the cloud. You just load a base model, point it at your training dataset, and get a lightweight LoRA adapter back — all running locally. The fine-tuned model can then be used for inference immediately, with no extra setup. Why it matters: LoRA (Low-Rank Adaptation) fine-tuning lets you specialize a general-purpose language model for your specific use case — like matching a brand's tone, mastering domain terminology, or following a particular output format — using a fraction of the compute a full fine-tune would require. QVAC handles the entire workflow locally: dataset preparation, training with configurable hyperparameters, checkpoint saving, and seamless inference with the resulting adapter. Your data never leaves the device. The developer experience: Fine-tuning with QVAC is as simple as calling "sdk.finetune()" with your dataset and a few hyperparameters. Training runs entirely on your local hardware, produces a compact LoRA adapter file, and supports pause/resume so you can stop a job and pick it back up without losing progress. The result plugs straight into QVAC's inference pipeline — no model conversion, no deployment step, just immediate local completions with your fine-tuned model.

Paolo Ardoino 🤖

42,413 views • 5 months ago

We have news! We created a new robotics model called Loop Model 1. On the zip-tie insertion task, it achieves 20x more throughput per unit of data than "Pi06 + RLT" from Physical Intelligence, a top model for such tasks. It’s the missing piece that makes MicroFactory work, because now deployment becomes so simple and fast that our users can do it themselves.

We have news! We created a new robotics model called Loop Model 1. On the zip-tie insertion task, it achieves 20x more throughput per unit of data than "Pi06 + RLT" from Physical Intelligence, a top model for such tasks. It’s the missing piece that makes MicroFactory work, because now deployment becomes so simple and fast that our users can do it themselves.

Igor Kulakov (MicroFactory)

77,133 views • 4 months ago

TokenFlow: Consistent Diffusion Features for Consistent Video Editing paper page: The generative AI revolution has recently expanded to videos. Nevertheless, current state-of-the-art video models are still lagging behind image models in terms of visual quality and user control over the generated content. In this work, we present a framework that harnesses the power of a text-to-image diffusion model for the task of text-driven video editing. Specifically, given a source video and a target text-prompt, our method generates a high-quality video that adheres to the target text, while preserving the spatial layout and motion of the input video. Our method is based on a key observation that consistency in the edited video can be obtained by enforcing consistency in the diffusion feature space. We achieve this by explicitly propagating diffusion features based on inter-frame correspondences, readily available in the model. Thus, our framework does not require any training or fine-tuning, and can work in conjunction with any off-the-shelf text-to-image editing method. We demonstrate state-of-the-art editing results on a variety of real-world videos.

TokenFlow: Consistent Diffusion Features for Consistent Video Editing paper page: The generative AI revolution has recently expanded to videos. Nevertheless, current state-of-the-art video models are still lagging behind image models in terms of visual quality and user control over the generated content. In this work, we present a framework that harnesses the power of a text-to-image diffusion model for the task of text-driven video editing. Specifically, given a source video and a target text-prompt, our method generates a high-quality video that adheres to the target text, while preserving the spatial layout and motion of the input video. Our method is based on a key observation that consistency in the edited video can be obtained by enforcing consistency in the diffusion feature space. We achieve this by explicitly propagating diffusion features based on inter-frame correspondences, readily available in the model. Thus, our framework does not require any training or fine-tuning, and can work in conjunction with any off-the-shelf text-to-image editing method. We demonstrate state-of-the-art editing results on a variety of real-world videos.

AK

143,469 views • 3 years ago

Show it a single human demonstration of a long-horizon task and the robot performs it right away. No fine-tuning, no retraining. Skild has released S1, a robot foundation model where you specify the task with a video demonstration instead of a language instruction. Why video, not language Language works for "hand me the mug." It falls apart for anything delicate or long-horizon. You don't learn to fold a fitted sheet from a sentence. So S1 takes a video demo into its context window and translates it to its own body and scene. What it does - Tasks up to 10 minutes, dozens of sequential steps - Unseen tasks: potting plants, pour-over coffee, kit assembly, pancakes. Skild says even the underlying primitives, like flipping a pancake, weren't in pre-training - Recovers from its own mistakes without being trained to - Improves on flawed demos. Demonstrator drops an egg, S1 doesn't make that mistake. It reads the demo as a goal, not a trajectory to copy.

Show it a single human demonstration of a long-horizon task and the robot performs it right away. No fine-tuning, no retraining. Skild has released S1, a robot foundation model where you specify the task with a video demonstration instead of a language instruction. Why video, not language Language works for "hand me the mug." It falls apart for anything delicate or long-horizon. You don't learn to fold a fitted sheet from a sentence. So S1 takes a video demo into its context window and translates it to its own body and scene. What it does - Tasks up to 10 minutes, dozens of sequential steps - Unseen tasks: potting plants, pour-over coffee, kit assembly, pancakes. Skild says even the underlying primitives, like flipping a pancake, weren't in pre-training - Recovers from its own mistakes without being trained to - Improves on flawed demos. Demonstrator drops an egg, S1 doesn't make that mistake. It reads the demo as a goal, not a trajectory to copy.

The Humanoid Hub

19,027 views • 1 month ago

🤔 Why do we still rely on the final layer of an LLM, when different layers encode different information? 🤔 In our new work, “Improving LLM Final Representations with Inter-Layer Geometry” (ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling) we show that actually, LLMs do not have one “best” layer. We introduce the Cayley-Encoder: an efficient and effective geometric encoder that learns one strong representation from all layer representations of the LLM, without biasing the representation toward any specific layer. While adding at most 0.1% learned parameters to the LLM, the Cayley-Encoder achieves large empirical gains over LoRA fine-tuning, final-layer representations, expensive attention-based aggregation, and methods that optimize specific layers for the task.

🤔 Why do we still rely on the final layer of an LLM, when different layers encode different information? 🤔 In our new work, “Improving LLM Final Representations with Inter-Layer Geometry” (ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling) we show that actually, LLMs do not have one “best” layer. We introduce the Cayley-Encoder: an efficient and effective geometric encoder that learns one strong representation from all layer representations of the LLM, without biasing the representation toward any specific layer. While adding at most 0.1% learned parameters to the LLM, the Cayley-Encoder achieves large empirical gains over LoRA fine-tuning, final-layer representations, expensive attention-based aggregation, and methods that optimize specific layers for the task.

Maya Bechler-Speicher

16,703 views • 4 months ago

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

Andrew Ng

86,697 views • 1 year ago

Small Language Models (SML) are the future of AI. "Small" (SML) instead of "Large" (LLM). These small models are highly specialized models with superhuman abilities on specific tasks. Here are two techniques to build these models: • Spectrum • Model Merging I give you a short introduction in the attached video, but here is a quick summary: Spectrum helps us identify the most relevant layers to solve one specific task. We can ignore everything else and focus on fine-tuning these layers. Using Spectrum, we can fine-tune models in a heartbeat. Model Merging combines multiple models into a unique, much better model than any of the individual input models. You can also combine models specialized in different tasks and get a model with multiple abilities. This is the state of the art of productizing models. It's what Arcee.ai's platform does behind the scenes. Arcee collaborated with me on this post and is sponsoring it. There are three main steps to produce a model for your particular use case: 1. You create a dataset by uploading your data. 2. You train a model. At this step, Arcee uses Spectrum and Model Merging to produce a highly specialized model for your task. 3. You can deploy that model to any environment you want. Three important notes: • Training process is 2x faster and 2x cheaper than regular fine-tuning. • Resultant models are smaller and have higher accuracy. • They create these specialized models from open-source models. Check this site so you can fully appreciate how this works: If you want to fine-tune an open-source model, consider Arcee's platform. This is the state of the art.

Small Language Models (SML) are the future of AI. "Small" (SML) instead of "Large" (LLM). These small models are highly specialized models with superhuman abilities on specific tasks. Here are two techniques to build these models: • Spectrum • Model Merging I give you a short introduction in the attached video, but here is a quick summary: Spectrum helps us identify the most relevant layers to solve one specific task. We can ignore everything else and focus on fine-tuning these layers. Using Spectrum, we can fine-tune models in a heartbeat. Model Merging combines multiple models into a unique, much better model than any of the individual input models. You can also combine models specialized in different tasks and get a model with multiple abilities. This is the state of the art of productizing models. It's what Arcee.ai's platform does behind the scenes. Arcee collaborated with me on this post and is sponsoring it. There are three main steps to produce a model for your particular use case: 1. You create a dataset by uploading your data. 2. You train a model. At this step, Arcee uses Spectrum and Model Merging to produce a highly specialized model for your task. 3. You can deploy that model to any environment you want. Three important notes: • Training process is 2x faster and 2x cheaper than regular fine-tuning. • Resultant models are smaller and have higher accuracy. • They create these specialized models from open-source models. Check this site so you can fully appreciate how this works: If you want to fine-tune an open-source model, consider Arcee's platform. This is the state of the art.

Santiago

164,162 views • 2 years ago

Learn a development pattern to systematically improve the accuracy and reliability of LLM applications in our new short course, Improving Accuracy of LLM Applications, built in partnership with Lamini and Meta, and taught by Lamini’s CEO Sharon Zhou, and Meta’s Senior Director of Partner Engineering, Amit Sangani. (Disclosure: I am an investor in Lamini.) The path to tuning an LLM application can be complex. In this course, you'll learn a systematic sequence of steps for improving accuracy by reducing hallucinations: - Create an evaluation dataset to measure model accuracy - Add prompt engineering and self-reflection - Fine-tune your model including "memory-tuning" which is a new method of embedding facts in an LLM Using the Llama 3-8B parameter model, you will: - Build a text-to-SQL agent with a custom schema and simulate situations where it hallucinates - Understand the difference between instruction fine-tuning, which gives pre-trained LLMs instructions to follow, and memory fine-tuning - See how Performance-Efficient Fine-tuning (PEFT) techniques like Low-Rank Adaptation (LoRA) reduce training time by 100x and Mixture of Memory Experts (MoME) reduces it even further I appreciate Meta releasing the Llama's family of open models -- this course gives an example of the unique type of work that developers can do with such models. Please sign up here:

Learn a development pattern to systematically improve the accuracy and reliability of LLM applications in our new short course, Improving Accuracy of LLM Applications, built in partnership with Lamini and Meta, and taught by Lamini’s CEO Sharon Zhou, and Meta’s Senior Director of Partner Engineering, Amit Sangani. (Disclosure: I am an investor in Lamini.) The path to tuning an LLM application can be complex. In this course, you'll learn a systematic sequence of steps for improving accuracy by reducing hallucinations: - Create an evaluation dataset to measure model accuracy - Add prompt engineering and self-reflection - Fine-tune your model including "memory-tuning" which is a new method of embedding facts in an LLM Using the Llama 3-8B parameter model, you will: - Build a text-to-SQL agent with a custom schema and simulate situations where it hallucinates - Understand the difference between instruction fine-tuning, which gives pre-trained LLMs instructions to follow, and memory fine-tuning - See how Performance-Efficient Fine-tuning (PEFT) techniques like Low-Rank Adaptation (LoRA) reduce training time by 100x and Mixture of Memory Experts (MoME) reduces it even further I appreciate Meta releasing the Llama's family of open models -- this course gives an example of the unique type of work that developers can do with such models. Please sign up here:

Andrew Ng

66,549 views • 2 years ago

Progress in open models is keeping Big AI labs up at night, and I'm here for it! We have a brand new open-weight multimodal model optimized for long-horizon tasks. This model is really good at something: it can work on tasks that keep evolving over time. • 280B total parameters, but only 16B active • 512K context window • Understands text, images, and audio • Strong reasoning, coding, and tool use But the best of all: the model learns and adapts to new information! Imagine you start running an agent today to solve a problem, and while it's working, you get new information that changes the initial conditions, or you change your mind. The agents you run today don't have issues with short tasks and goals that don't change, but reality is messy, and that makes it hard for long-horizon agents to succeed. The new dots3-note Preview model introduces TEMPO. TEMPO is a new reinforcement learning technique that lets the model periodically pause and critique its own progress. Basically, from time to time, the agent asks itself: "Am I getting closer to the goal, or am I wasting my time?" The same model switches between actor and critic. The actor works on the problem. The critic looks at the current state, reasons about how much progress it has made, and determines what should happen next. TEMPO gives the model feedback along the way. This is huge for any agent that can work on long-horizon tasks without wasting its time.

Santiago

80,792 views • 1 month ago

Introducing FLUX 3 Action. An open weights 7B World Action Model that achieves first place on the RoboLab benchmark. It outperforms the previous best open model by 6.1 percentage points while using 56% fewer parameters and running up to 3.95x faster.⁠⁠ FLUX 3 Action removes the usual trade-off between world action model performance and VLA speed: it still predicts video and actions together, but plans more than twice as far ahead and runs faster per second of robot motion than the strongest open VLA. Teams can fine-tune FLUX 3 Action on their own demonstrations to create policies for a particular robot and task. Together with NVIDIA, we also integrated FLUX 3 Action natively into Hugging Face's LeRobot, with fine-tuning recipes included and edge deployment on NVIDIA Jetson. Beyond robotics, we’re also seeing promising results training task-specific policies for acting in simulated environments like gaming, controlling a vehicle, computer use, and wherever else a model needs to understand a visual environment and then choose what to do next. FLUX 3 Action builds on the same image, video, and audio pretraining as FLUX 3, but uses a smaller architecture designed for practical deployment. In midtraining, we trained the model to predict actions and future frames together. We’re releasing the weights, code, fine-tuning recipe, benchmarks, and reproducible examples so researchers and developers can build on the model with their own robots, environments, and tasks (see below).

Introducing FLUX 3 Action. An open weights 7B World Action Model that achieves first place on the RoboLab benchmark. It outperforms the previous best open model by 6.1 percentage points while using 56% fewer parameters and running up to 3.95x faster.⁠⁠ FLUX 3 Action removes the usual trade-off between world action model performance and VLA speed: it still predicts video and actions together, but plans more than twice as far ahead and runs faster per second of robot motion than the strongest open VLA. Teams can fine-tune FLUX 3 Action on their own demonstrations to create policies for a particular robot and task. Together with NVIDIA, we also integrated FLUX 3 Action natively into Hugging Face's LeRobot, with fine-tuning recipes included and edge deployment on NVIDIA Jetson. Beyond robotics, we’re also seeing promising results training task-specific policies for acting in simulated environments like gaming, controlling a vehicle, computer use, and wherever else a model needs to understand a visual environment and then choose what to do next. FLUX 3 Action builds on the same image, video, and audio pretraining as FLUX 3, but uses a smaller architecture designed for practical deployment. In midtraining, we trained the model to predict actions and future frames together. We’re releasing the weights, code, fine-tuning recipe, benchmarks, and reproducible examples so researchers and developers can build on the model with their own robots, environments, and tasks (see below).

Black Forest Labs

177,424 views • 11 days ago

🚀 Excited to release AIBuildAI — an AI agent that automatically builds AI models 🏆 #1 on OpenAI MLE-Bench 💻 GitHub: Building AI models still requires a lot of manual work — designing models, writing code, training them, tuning hyperparameters, and iterating on results. We developed AIBuildAI to automate much of this process. The system runs an agent loop that automatically: • analyzes the task • designs models • writes code to implement them • trains the models • tunes hyperparameters • evaluates model performance • iteratively improves the models 🏆 On OpenAI’s MLE-Bench benchmark, AIBuildAI ranked #1:

🚀 Excited to release AIBuildAI — an AI agent that automatically builds AI models 🏆 #1 on OpenAI MLE-Bench 💻 GitHub: Building AI models still requires a lot of manual work — designing models, writing code, training them, tuning hyperparameters, and iterating on results. We developed AIBuildAI to automate much of this process. The system runs an agent loop that automatically: • analyzes the task • designs models • writes code to implement them • trains the models • tunes hyperparameters • evaluates model performance • iteratively improves the models 🏆 On OpenAI’s MLE-Bench benchmark, AIBuildAI ranked #1:

Pengtao Xie

152,252 views • 6 months ago

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

Sudo su

32,589 views • 3 months ago

Physics-based Motion Retargeting from Sparse Inputs paper page: Avatars are important to create interactive and immersive experiences in virtual worlds. One challenge in animating these characters to mimic a user's motion is that commercial AR/VR products consist only of a headset and controllers, providing very limited sensor data of the user's pose. Another challenge is that an avatar might have a different skeleton structure than a human and the mapping between them is unclear. In this work we address both of these challenges. We introduce a method to retarget motions in real-time from sparse human sensor data to characters of various morphologies. Our method uses reinforcement learning to train a policy to control characters in a physics simulator. We only require human motion capture data for training, without relying on artist-generated animations for each avatar. This allows us to use large motion capture datasets to train general policies that can track unseen users from real and sparse data in real-time. We demonstrate the feasibility of our approach on three characters with different skeleton structure: a dinosaur, a mouse-like creature and a human. We show that the avatar poses often match the user surprisingly well, despite having no sensor information of the lower body available. We discuss and ablate the important components in our framework, specifically the kinematic retargeting step, the imitation, contact and action reward as well as our asymmetric actor-critic observations. We further explore the robustness of our method in a variety of settings including unbalancing, dancing and sports motions.

AK

106,527 views • 3 years ago