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 course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with Red Hat and taught by Cedric Clyburn. Efficient LLM serving requires efficient memory management. A 70B-parameter model takes ~140 GB just... to load the weights. On top of that, every active request needs its own chunk of GPU memory, the KV cache, to store the token context it has built up so far. In this course, you'll learn to reduce a model's memory footprint with quantization and serve it using vLLM, which handles many concurrent requests efficiently through smart memory management. Skills you'll gain: - Quantize a model and measure the accuracy tradeoff - Serve a model with vLLM and watch it handle concurrent requests efficiently - Benchmark your deployment and make informed tradeoffs between speed, cost, and accuracy Join and learn to serve LLMs efficiently:show more

Andrew Ng

1,837,975 subscribers

133,542 views • 2 months ago •via X (Twitter)

Education Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Learn how to build an optimized LLM inference system from the ground up in our new short course, Efficiently Serving LLMs, built in collaboration with Predibase by Rubrik and taught by Travis Addair. Whether you're serving your own LLM or using a model hosting service, this course will give you a deep understanding of the optimizations required to efficiently serve many users at once. - Learn how LLMs generate text one token at a time, and how techniques like KV caching, continuous batching, and quantization speed things up and optimize memory usage for serving multiple users. - Benchmark the performance of these LLM optimizations to explore the trade-offs between quickly responding to an individual user’s request vs. serving many users at once. - Use techniques like low-rank adaptation (LoRA) to efficiently serve hundreds of unique, custom fine-tuned models on a single device, without sacrificing throughput. - Use Predibase's LoRAX framework to see optimization techniques in action on a real LLM server. Sign up here:

Learn how to build an optimized LLM inference system from the ground up in our new short course, Efficiently Serving LLMs, built in collaboration with Predibase by Rubrik and taught by Travis Addair. Whether you're serving your own LLM or using a model hosting service, this course will give you a deep understanding of the optimizations required to efficiently serve many users at once. - Learn how LLMs generate text one token at a time, and how techniques like KV caching, continuous batching, and quantization speed things up and optimize memory usage for serving multiple users. - Benchmark the performance of these LLM optimizations to explore the trade-offs between quickly responding to an individual user’s request vs. serving many users at once. - Use techniques like low-rank adaptation (LoRA) to efficiently serve hundreds of unique, custom fine-tuned models on a single device, without sacrificing throughput. - Use Predibase's LoRAX framework to see optimization techniques in action on a real LLM server. Sign up here:

Andrew Ng

104,727 views • 2 years ago

New course: Efficient Inference with SGLang: Text and Image Generation, built in partnership with LMSys LMSYS Org and RadixArk RadixArk, and taught by Richard Chen Richard Chen, a Member of Technical Staff at RadixArk. Running LLMs in production is expensive, and much of that cost comes from redundant computation. This short course teaches you to eliminate that waste using SGLang, an open-source inference framework that caches computation already done and reuses it across future requests. When ten users share the same system prompt, SGLang processes it once, not ten times. The speedups compound quickly, especially when there's a lot of shared context across requests. Skills you'll gain: - Implement a KV cache from scratch to eliminate redundant computation within a single request - Scale caching across users and requests with RadixAttention, so shared context is only processed once - Accelerate image generation with diffusion models using SGLang's caching and multi-GPU parallelism Join and learn to make LLM inference faster and more cost-efficient at scale!

New course: Efficient Inference with SGLang: Text and Image Generation, built in partnership with LMSys LMSYS Org and RadixArk RadixArk, and taught by Richard Chen Richard Chen, a Member of Technical Staff at RadixArk. Running LLMs in production is expensive, and much of that cost comes from redundant computation. This short course teaches you to eliminate that waste using SGLang, an open-source inference framework that caches computation already done and reuses it across future requests. When ten users share the same system prompt, SGLang processes it once, not ten times. The speedups compound quickly, especially when there's a lot of shared context across requests. Skills you'll gain: - Implement a KV cache from scratch to eliminate redundant computation within a single request - Scale caching across users and requests with RadixAttention, so shared context is only processed once - Accelerate image generation with diffusion models using SGLang's caching and multi-GPU parallelism Join and learn to make LLM inference faster and more cost-efficient at scale!

Andrew Ng

101,794 views • 4 months 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

66,435 views • 2 years ago

New course: Agent Memory: Building Memory-Aware Agents, built in partnership with Oracle and taught by Richmond Alake and Nacho Martínez. Many agents work well within a single session but their memory resets once the session ends. Consider a research agent working on dozens of papers across multiple days: without memory, it has no way to store and retrieve what it learned across sessions. This short course teaches you to build a memory system that enables agents to persist memory and thereby learn across sessions. You'll design a Memory Manager that handles different memory types, implement semantic tool retrieval that scales without bloating the context, and build write-back pipelines that let your agent autonomously update and refine what it knows over time. Skills you'll gain: - Build persistent memory stores for different agent memory types - Implement a Memory Manager that orchestrates how your agent reads, writes, and retrieves memory - Treat tools as procedural memory and retrieve only relevant ones at inference time using semantic search Join and learn to build agents that remember and improve over time!

New course: Agent Memory: Building Memory-Aware Agents, built in partnership with Oracle and taught by Richmond Alake and Nacho Martínez. Many agents work well within a single session but their memory resets once the session ends. Consider a research agent working on dozens of papers across multiple days: without memory, it has no way to store and retrieve what it learned across sessions. This short course teaches you to build a memory system that enables agents to persist memory and thereby learn across sessions. You'll design a Memory Manager that handles different memory types, implement semantic tool retrieval that scales without bloating the context, and build write-back pipelines that let your agent autonomously update and refine what it knows over time. Skills you'll gain: - Build persistent memory stores for different agent memory types - Implement a Memory Manager that orchestrates how your agent reads, writes, and retrieves memory - Treat tools as procedural memory and retrieve only relevant ones at inference time using semantic search Join and learn to build agents that remember and improve over time!

Andrew Ng

160,148 views • 5 months ago

New course: Build and Train an LLM with JAX, built in partnership with Google and taught by Chris Achard. JAX is the open-source library behind Google's Gemini, Veo, and other advanced models. This short course teaches you to build and train a 20-million parameter language model from scratch using JAX and its ecosystem of tools. You'll implement a complete MiniGPT-style architecture from scratch, train it, and chat with your finished model through a graphical interface. Skills you'll gain: - Learn JAX's core primitives: automatic differentiation, JIT compilation, and vectorized execution - Build a MiniGPT-style LLM using Flax/NNX, implementing embedding and transformer blocks - Load a pretrained MiniGPT model and run inference through a chat interface Come learn this important software layer for building LLMs!

New course: Build and Train an LLM with JAX, built in partnership with Google and taught by Chris Achard. JAX is the open-source library behind Google's Gemini, Veo, and other advanced models. This short course teaches you to build and train a 20-million parameter language model from scratch using JAX and its ecosystem of tools. You'll implement a complete MiniGPT-style architecture from scratch, train it, and chat with your finished model through a graphical interface. Skills you'll gain: - Learn JAX's core primitives: automatic differentiation, JIT compilation, and vectorized execution - Build a MiniGPT-style LLM using Flax/NNX, implementing embedding and transformer blocks - Load a pretrained MiniGPT model and run inference through a chat interface Come learn this important software layer for building LLMs!

Andrew Ng

193,392 views • 5 months ago

New course: Transformers in Practice. You'll get a practical view of how transformer-based LLMs work, so you can reason about their behavior, diagnose problems like slow inference, and make smarter decisions about deployment. This course is built in partnership with AMD and taught by Sharon Zhou. You'll see how transformers generate text one token at a time, how the model decides which earlier words matter most when predicting the next one, and how techniques like quantization speed up inference on GPUs. This is not a video-only course; interactive visualizations throughout let you play with these concepts and build intuition that sticks. Skills you'll gain: - Understand why LLMs hallucinate, and RAG and chain-of-thought shape what they generate - Look inside the model to see how attention and layers combine to predict the next token - Diagnose inference bottlenecks and learn the techniques that speed up transformers on GPUs Join and understand what's really happening inside your LLMs:

New course: Transformers in Practice. You'll get a practical view of how transformer-based LLMs work, so you can reason about their behavior, diagnose problems like slow inference, and make smarter decisions about deployment. This course is built in partnership with AMD and taught by Sharon Zhou. You'll see how transformers generate text one token at a time, how the model decides which earlier words matter most when predicting the next one, and how techniques like quantization speed up inference on GPUs. This is not a video-only course; interactive visualizations throughout let you play with these concepts and build intuition that sticks. Skills you'll gain: - Understand why LLMs hallucinate, and RAG and chain-of-thought shape what they generate - Look inside the model to see how attention and layers combine to predict the next token - Diagnose inference bottlenecks and learn the techniques that speed up transformers on GPUs Join and understand what's really happening inside your LLMs:

Andrew Ng

123,383 views • 3 months ago

LLMs can take gigabytes of memory to store, which limits what can be run on consumer hardware. But quantization can dramatically compress models, making a wider selection of models available to developers. You can often reduce model size by 4x or more while maintaining reasonable performance. In our new short course Quantization Fundamentals taught by Hugging Face's @younesbelkada and Marc Sun, you'll: - Learn how to quantize nearly any open source model - Use int8 and bfloat16 (Brain float 16) data types to load and run LLMs using PyTorch and the Hugging Face Transformers library - Dive into the technical details of linear quantization to map 32-bit floats to 8-bit integers As models get bigger and bigger, quantization becomes more important for making models practical and accessible. Please check out the course here:

LLMs can take gigabytes of memory to store, which limits what can be run on consumer hardware. But quantization can dramatically compress models, making a wider selection of models available to developers. You can often reduce model size by 4x or more while maintaining reasonable performance. In our new short course Quantization Fundamentals taught by Hugging Face's @younesbelkada and Marc Sun, you'll: - Learn how to quantize nearly any open source model - Use int8 and bfloat16 (Brain float 16) data types to load and run LLMs using PyTorch and the Hugging Face Transformers library - Dive into the technical details of linear quantization to map 32-bit floats to 8-bit integers As models get bigger and bigger, quantization becomes more important for making models practical and accessible. Please check out the course here:

Andrew Ng

288,316 views • 2 years ago

New short course: LLMs as Operating Systems: Agent Memory, created with Letta, and taught by its founders Charles Packer and Sarah Wooders. An LLM's input context window has limited space. Using a longer input context also costs more and results in slower processing. So, managing what's stored in this context window is important. In the innovative paper MemGPT: Towards LLMs as Operating Systems, its authors (which include the instructors) proposed using an LLM agent to manage this context window. Their system uses a large persistent memory that stores everything that could be included in the input context, and an agent decides what is actually included. Take the example of building a chatbot that needs to remember what's been said earlier in a conversation (perhaps over many days of interaction with a user). As the conversation's length grows, the memory management agent will move information from the input context to a persistent searchable database; summarize information to keep relevant facts in the input context; and restore relevant conversation elements from further back in time. This allows a chatbot to keep what's currently most relevant in its input context memory to generate the next response. When I read the original MemGPT paper, I thought it was an innovative technique for handling memory for LLMs. The open-source Letta framework, which we'll use in this course, makes MemGPT easy to implement. It adds memory to your LLM agents and gives them transparent long-term memory. In detail, you’ll learn: - How to build an agent that can edit its own limited input context memory, using tools and multi-step reasoning - What is a memory hierarchy (an idea from computer operating systems, which use a cache to speed up memory access), and how these ideas apply to managing the LLM input context (where the input context window is a "cache" storing the most relevant information; and an agent decides what to move in and out of this to/from a larger persistent storage system) - How to implement multi-agent collaboration by letting different agents share blocks of memory This course will give you a sophisticated understanding of memory management for LLMs, which is important for chatbots having long conversations, and for complex agentic workflows. Please sign up here!

New short course: LLMs as Operating Systems: Agent Memory, created with Letta, and taught by its founders Charles Packer and Sarah Wooders. An LLM's input context window has limited space. Using a longer input context also costs more and results in slower processing. So, managing what's stored in this context window is important. In the innovative paper MemGPT: Towards LLMs as Operating Systems, its authors (which include the instructors) proposed using an LLM agent to manage this context window. Their system uses a large persistent memory that stores everything that could be included in the input context, and an agent decides what is actually included. Take the example of building a chatbot that needs to remember what's been said earlier in a conversation (perhaps over many days of interaction with a user). As the conversation's length grows, the memory management agent will move information from the input context to a persistent searchable database; summarize information to keep relevant facts in the input context; and restore relevant conversation elements from further back in time. This allows a chatbot to keep what's currently most relevant in its input context memory to generate the next response. When I read the original MemGPT paper, I thought it was an innovative technique for handling memory for LLMs. The open-source Letta framework, which we'll use in this course, makes MemGPT easy to implement. It adds memory to your LLM agents and gives them transparent long-term memory. In detail, you’ll learn: - How to build an agent that can edit its own limited input context memory, using tools and multi-step reasoning - What is a memory hierarchy (an idea from computer operating systems, which use a cache to speed up memory access), and how these ideas apply to managing the LLM input context (where the input context window is a "cache" storing the most relevant information; and an agent decides what to move in and out of this to/from a larger persistent storage system) - How to implement multi-agent collaboration by letting different agents share blocks of memory This course will give you a sophisticated understanding of memory management for LLMs, which is important for chatbots having long conversations, and for complex agentic workflows. Please sign up here!

Andrew Ng

201,127 views • 1 year ago

New short course: Long-Term Agentic Memory with LangGraph. Learn to build an agent with long-term memory in this course developed in collaboration with taught by its Co-Founder and CEO, Harrison Chase! Personal assistance and productivity tasks have become important use cases for agents. An important feature of an AI assistant, such as a coding or calendar assistant, is its ability to keep improving over time from its experience. Agent memory is the key capability that enables this. To add memory to an agent, you must first figure out what to store and what to retrieve when it is time to use the information. Additionally, you’ll have to decide when to update the stored information. For example, you might update in each iteration loop of the agent or perform updates in the background, with a helper agent. In this course, you will learn a mental framework to build agents with long-term memory. You'll create a useful email assistant that can respond, ignore, and notify using writing, scheduling, and memory-management tools. You’ll develop your agent's memory by adding facts to its memory store, provide examples to learn the user's preferences, and optimize system prompts to evolve instructions based on previous responses. In detail, you’ll: - Learn how the three types of memory--semantic, episodic, and procedural–and the two update mechanisms–via hot path and in the background–apply to your agents. - Build an email agent with writing, scheduling, and availability tools, along with a router that triages incoming email and handles it accordingly by ignoring, responding, or notifying the user. - Add tools to your email agent that allow it to operate on semantic memory by learning facts about the user, storing them in a long-term memory store, and searching over them in future interactions. - Incorporate episodic memory, in the form of few-shot examples, in the triage step of your agents to help them learn and update user preferences. - Add procedural memory as system prompts, optimized with feedback to improve the instructions the agent follows. Learn how to approach memory in agents, and start building agents with long-term memory with LangGraph! Please sign up here:

New short course: Long-Term Agentic Memory with LangGraph. Learn to build an agent with long-term memory in this course developed in collaboration with taught by its Co-Founder and CEO, Harrison Chase! Personal assistance and productivity tasks have become important use cases for agents. An important feature of an AI assistant, such as a coding or calendar assistant, is its ability to keep improving over time from its experience. Agent memory is the key capability that enables this. To add memory to an agent, you must first figure out what to store and what to retrieve when it is time to use the information. Additionally, you’ll have to decide when to update the stored information. For example, you might update in each iteration loop of the agent or perform updates in the background, with a helper agent. In this course, you will learn a mental framework to build agents with long-term memory. You'll create a useful email assistant that can respond, ignore, and notify using writing, scheduling, and memory-management tools. You’ll develop your agent's memory by adding facts to its memory store, provide examples to learn the user's preferences, and optimize system prompts to evolve instructions based on previous responses. In detail, you’ll: - Learn how the three types of memory--semantic, episodic, and procedural–and the two update mechanisms–via hot path and in the background–apply to your agents. - Build an email agent with writing, scheduling, and availability tools, along with a router that triages incoming email and handles it accordingly by ignoring, responding, or notifying the user. - Add tools to your email agent that allow it to operate on semantic memory by learning facts about the user, storing them in a long-term memory store, and searching over them in future interactions. - Incorporate episodic memory, in the form of few-shot examples, in the triage step of your agents to help them learn and update user preferences. - Add procedural memory as system prompts, optimized with feedback to improve the instructions the agent follows. Learn how to approach memory in agents, and start building agents with long-term memory with LangGraph! Please sign up here:

Andrew Ng

131,850 views • 1 year ago

researchers gave a tiny local model human-style memory and its context limit basically stopped existing a team from MBZUAI, Princeton and Weizmann took a 1B model and rebuilt how it reads. instead of attending to everything at once, the model reads in 1,024 token chunks and passes the important stuff forward through an associative memory, the same way you carry the plot of a book between chapters without rereading them. the design mirrors human memory on purpose. full attention inside a chunk works as short-term memory. the module that carries information between chunks works as long-term memory. they even trained it like a person, starting with short easy texts and raising the difficulty gradually, because memory thrown into the deep end learns nothing. the numbers back it up. the normal model burns 40GB of GPU memory on a long document and collapses hard past its limit, dropping from 0.86 to 0.32 accuracy. the memory version holds 0.71 at double that length while using a flat 12GB no matter how long the input gets. it also needs about 30% fewer FLOPs. the part i keep thinking about is that nobody scaled anything here. they didn't build a bigger model, didn't stretch the window, didn't add compute. they looked at how a brain handles a long day and copied the architecture. a model small enough to run on a consumer gpu now survives documents its own architecture used to choke on. we keep treating intelligence as a compute problem. sometimes it's a memory problem.

researchers gave a tiny local model human-style memory and its context limit basically stopped existing a team from MBZUAI, Princeton and Weizmann took a 1B model and rebuilt how it reads. instead of attending to everything at once, the model reads in 1,024 token chunks and passes the important stuff forward through an associative memory, the same way you carry the plot of a book between chapters without rereading them. the design mirrors human memory on purpose. full attention inside a chunk works as short-term memory. the module that carries information between chunks works as long-term memory. they even trained it like a person, starting with short easy texts and raising the difficulty gradually, because memory thrown into the deep end learns nothing. the numbers back it up. the normal model burns 40GB of GPU memory on a long document and collapses hard past its limit, dropping from 0.86 to 0.32 accuracy. the memory version holds 0.71 at double that length while using a flat 12GB no matter how long the input gets. it also needs about 30% fewer FLOPs. the part i keep thinking about is that nobody scaled anything here. they didn't build a bigger model, didn't stretch the window, didn't add compute. they looked at how a brain handles a long day and copied the architecture. a model small enough to run on a consumer gpu now survives documents its own architecture used to choke on. we keep treating intelligence as a compute problem. sometimes it's a memory problem.

Alex Veremeyenko

16,147 views • 1 month ago

New course announcement: Semantic Caching for AI Agents, taught by Tyler Hutcherson and Iliya Zhechev from Redis. Semantic caching can significantly reduce your AI application's inference costs and latency. If someone asks "How do I get a refund?" and another later asks "I want my money back," semantic caching recognizes these mean the same thing so it can use a cached response instead of making another model call. This short course takes you from building your first semantic cache from scratch to implementing production-ready systems using Redis' open-source tools. Skills you'll gain: - Build semantic caches from scratch, then implement them using Redis' SDK with production features - Measure cache performance using hit rate, precision, recall, and latency - Enhance accuracy with threshold tuning, cross-encoders, LLM validation, and fuzzy matching Join and learn to reduce your agentic AI's costs and improve speed!

New course announcement: Semantic Caching for AI Agents, taught by Tyler Hutcherson and Iliya Zhechev from Redis. Semantic caching can significantly reduce your AI application's inference costs and latency. If someone asks "How do I get a refund?" and another later asks "I want my money back," semantic caching recognizes these mean the same thing so it can use a cached response instead of making another model call. This short course takes you from building your first semantic cache from scratch to implementing production-ready systems using Redis' open-source tools. Skills you'll gain: - Build semantic caches from scratch, then implement them using Redis' SDK with production features - Measure cache performance using hit rate, precision, recall, and latency - Enhance accuracy with threshold tuning, cross-encoders, LLM validation, and fuzzy matching Join and learn to reduce your agentic AI's costs and improve speed!

Andrew Ng

62,497 views • 9 months ago

A tricky LLM interview question: You're serving a reasoning model on vLLM, and it keeps running out of GPU memory on long traces. So you add KV cache compression and evict 90% of the cached tokens. VRAM usage stays as is and GPU still runs out of memory. Why? (answer below) Evicting 90% of the KV cache can free almost none of the memory it was using. This sounds counterintuitive, but it follows directly from how production servers store the cache today. The KV cache grows with every token a model generates. Each token appends its key and value vectors across every layer, and nothing is freed while generation continues. This is the dominant memory cost for reasoning models. If a 32K-token CoT caches ~32K tokens of KV vectors, a Qwen3-32B with 4-bit weights will run out-of-memory around 24K tokens on a 24GB GPU. One obvious solution is to keep the important tokens and drop the rest, since attention is sparse enough to allow it. But this does not solve the memory problem yet. The reason is paged attention, which is the memory manager behind vLLM and most production servers. Under the hood, it splits GPU memory into fixed physical blocks, each one holds the KV for about 16 tokens. This block returns to the allocator only when every slot inside it is empty. Since the eviction logic selects tokens by importance, and such tokens are scattered across blocks... ...so despite eviction, almost every block is left with at least some survivor tokens. For instance, if the logic evicts 14k of 16k tokens across 1,000 blocks, most likely every block will still have a token. This means the allocator frees almost nothing. Placing the new tokens into those freed slots is not ideal because it breaks the cache's layout. Say token 16,001 arrives, and it's placed in the slot the 40th token used to hold. The cache now reads position 38, then 16,001, then 41, so the cache is no longer in token order. Attention can still compute the right answer from that, but only if every slot now carries a separate note recording which position it actually holds. This introduces another bookkeeping cost that an in-order layout inherently avoids. So the cache is logically 90% smaller and still physically the same size. Many compression results miss this because they measure on pre-allocated contiguous tensors rather than a paged server. There's another problem. Eviction methods pick which tokens to keep by looking at the attention scores themselves (as expected). But fast attention kernels used in production, like FlashAttention, never save those scores. They compute attention in small pieces and throw the full score grid away as they go, which is also why they're fast. So the exact signal eviction methods need isn't available in memory. The workaround is to fall back to eager attention and build the full matrix, which gives up the speed FlashAttention was there to provide. NVIDIA published a method called TriAttention to solve both these problems. It never needs attention scores. Instead, it scores tokens from the geometry of the model's key and query vectors before RoPE is applied, where those vectors sit in stable clusters. For the memory problem, it runs a compaction pass every 128 decoded tokens. The surviving tokens slide forward to close the holes eviction creates, so whole blocks empty out and return to the allocator while the cache stays in token order. On long reasoning traces, the approach matches full-attention accuracy while decoding 2.5x faster and using 10.7x less KV memory. KV cache compression is a big infrastructure problem. The number that decides whether it works is the count of freed blocks, not the count of evicted tokens. You can find the NVIDIA write-up here: I wrote a first-principles breakdown of how the KV cache works. It walks through why the model stores keys and values at all, why the cache grows with every token, and a comparison of LLM generation speed with and without KV caching. Read it below.

A tricky LLM interview question: You're serving a reasoning model on vLLM, and it keeps running out of GPU memory on long traces. So you add KV cache compression and evict 90% of the cached tokens. VRAM usage stays as is and GPU still runs out of memory. Why? (answer below) Evicting 90% of the KV cache can free almost none of the memory it was using. This sounds counterintuitive, but it follows directly from how production servers store the cache today. The KV cache grows with every token a model generates. Each token appends its key and value vectors across every layer, and nothing is freed while generation continues. This is the dominant memory cost for reasoning models. If a 32K-token CoT caches ~32K tokens of KV vectors, a Qwen3-32B with 4-bit weights will run out-of-memory around 24K tokens on a 24GB GPU. One obvious solution is to keep the important tokens and drop the rest, since attention is sparse enough to allow it. But this does not solve the memory problem yet. The reason is paged attention, which is the memory manager behind vLLM and most production servers. Under the hood, it splits GPU memory into fixed physical blocks, each one holds the KV for about 16 tokens. This block returns to the allocator only when every slot inside it is empty. Since the eviction logic selects tokens by importance, and such tokens are scattered across blocks... ...so despite eviction, almost every block is left with at least some survivor tokens. For instance, if the logic evicts 14k of 16k tokens across 1,000 blocks, most likely every block will still have a token. This means the allocator frees almost nothing. Placing the new tokens into those freed slots is not ideal because it breaks the cache's layout. Say token 16,001 arrives, and it's placed in the slot the 40th token used to hold. The cache now reads position 38, then 16,001, then 41, so the cache is no longer in token order. Attention can still compute the right answer from that, but only if every slot now carries a separate note recording which position it actually holds. This introduces another bookkeeping cost that an in-order layout inherently avoids. So the cache is logically 90% smaller and still physically the same size. Many compression results miss this because they measure on pre-allocated contiguous tensors rather than a paged server. There's another problem. Eviction methods pick which tokens to keep by looking at the attention scores themselves (as expected). But fast attention kernels used in production, like FlashAttention, never save those scores. They compute attention in small pieces and throw the full score grid away as they go, which is also why they're fast. So the exact signal eviction methods need isn't available in memory. The workaround is to fall back to eager attention and build the full matrix, which gives up the speed FlashAttention was there to provide. NVIDIA published a method called TriAttention to solve both these problems. It never needs attention scores. Instead, it scores tokens from the geometry of the model's key and query vectors before RoPE is applied, where those vectors sit in stable clusters. For the memory problem, it runs a compaction pass every 128 decoded tokens. The surviving tokens slide forward to close the holes eviction creates, so whole blocks empty out and return to the allocator while the cache stays in token order. On long reasoning traces, the approach matches full-attention accuracy while decoding 2.5x faster and using 10.7x less KV memory. KV cache compression is a big infrastructure problem. The number that decides whether it works is the count of freed blocks, not the count of evicted tokens. You can find the NVIDIA write-up here: I wrote a first-principles breakdown of how the KV cache works. It walks through why the model stores keys and values at all, why the cache grows with every token, and a comparison of LLM generation speed with and without KV caching. Read it below.

Avi Chawla

271,839 views • 2 months ago

OpenAI just announced API access to o1 (advanced reasoning model) yesterday. I'm delighted to announce today a new short course, Reasoning with o1, built with OpenAI, and taught by Colin Jarvis, Head of AI Solutions at OpenAI, to show you how to use this effectively! Unlike previous language models which generate output directly, o1 “thinks before it responds,” and generates many reasoning tokens before returning a more thoughtful and accurate response. It is great at complex reasoning -- including planning for agentic workflows, coding, and domain-specific reasoning in STEM fields like law. But how you should use it is quite different from other LLMs. I think o1 will be a game changer for many AI applications; and in this course, you'll learn how to use it effectively. In detail, you’ll: - Learn to recognize what tasks o1 is suited for, and when to use a smaller model, or combine o1 with a smaller model - Understand the new principles of prompting reasoning models: Be simple and direct; no explicit chain-of-thought required; use structure; show rather than tell - Implement multi-step orchestration in which o1 plans, and hands tasks over to gpt-4o-mini to execute specific steps; this illustrates a design pattern to optimize intelligence (accuracy) and cost - Use o1 for a coding task to build a new application, edit existing code, and test performance by running a coding competition between o1-mini and GPT 4o - Use o1 for image understanding and learn how it performs better with a "hierarchy of reasoning," in which it incurs the latency and cost upfront, preprocessing the image and indexing it with rich details so it can be used for Q&A later - Learn a technique called meta-prompting, in which you use o1 to improve your prompts. Using a customer support evaluation set, you'll iteratively use o1 to modify a prompt to improve performance You'll also learn about how OpenAI used reinforcement learning to produce a model that uses "test-time compute" to improve performance. I think you'll find this course enjoyable and valuable. Please sign up for it here:

OpenAI just announced API access to o1 (advanced reasoning model) yesterday. I'm delighted to announce today a new short course, Reasoning with o1, built with OpenAI, and taught by Colin Jarvis, Head of AI Solutions at OpenAI, to show you how to use this effectively! Unlike previous language models which generate output directly, o1 “thinks before it responds,” and generates many reasoning tokens before returning a more thoughtful and accurate response. It is great at complex reasoning -- including planning for agentic workflows, coding, and domain-specific reasoning in STEM fields like law. But how you should use it is quite different from other LLMs. I think o1 will be a game changer for many AI applications; and in this course, you'll learn how to use it effectively. In detail, you’ll: - Learn to recognize what tasks o1 is suited for, and when to use a smaller model, or combine o1 with a smaller model - Understand the new principles of prompting reasoning models: Be simple and direct; no explicit chain-of-thought required; use structure; show rather than tell - Implement multi-step orchestration in which o1 plans, and hands tasks over to gpt-4o-mini to execute specific steps; this illustrates a design pattern to optimize intelligence (accuracy) and cost - Use o1 for a coding task to build a new application, edit existing code, and test performance by running a coding competition between o1-mini and GPT 4o - Use o1 for image understanding and learn how it performs better with a "hierarchy of reasoning," in which it incurs the latency and cost upfront, preprocessing the image and indexing it with rich details so it can be used for Q&A later - Learn a technique called meta-prompting, in which you use o1 to improve your prompts. Using a customer support evaluation set, you'll iteratively use o1 to modify a prompt to improve performance You'll also learn about how OpenAI used reinforcement learning to produce a model that uses "test-time compute" to improve performance. I think you'll find this course enjoyable and valuable. Please sign up for it here:

Andrew Ng

357,661 views • 1 year ago

New short course on Pretraining LLMs! Developed with Upstage and taught by their CEO Sung Kim and CSO Lucy Park. While prompting or fine-tuning existing models works well for many general language tasks, pretraining is valuable for specialized domains or languages with limited representation in current models. This course walks you through the LLM pretraining pipeline: 1. Data preparation: Learn to source, clean, and prepare training data using HuggingFace. 2. Model architecture: Configure transformer networks, including modifying existing models. 3. Training: Set up and run training using open-source libraries. 4. Evaluation: Benchmark performance using popular evaluation strategies. As an example use case, you'll also compare the output of a base model with its fine-tuned and further pretrained variants, to see the impact of pretraining on a model's ability to write Python. The course also explores an innovative technique called depth up-scaling, which Upstage used to train their Solar model family, reducing pretraining compute costs by up to 70%. This technique works by first duplicating layers of a smaller pretrained model to form a larger model, and then further pretraining the result. Sign up here!

New short course on Pretraining LLMs! Developed with Upstage and taught by their CEO Sung Kim and CSO Lucy Park. While prompting or fine-tuning existing models works well for many general language tasks, pretraining is valuable for specialized domains or languages with limited representation in current models. This course walks you through the LLM pretraining pipeline: 1. Data preparation: Learn to source, clean, and prepare training data using HuggingFace. 2. Model architecture: Configure transformer networks, including modifying existing models. 3. Training: Set up and run training using open-source libraries. 4. Evaluation: Benchmark performance using popular evaluation strategies. As an example use case, you'll also compare the output of a base model with its fine-tuned and further pretrained variants, to see the impact of pretraining on a model's ability to write Python. The course also explores an innovative technique called depth up-scaling, which Upstage used to train their Solar model family, reducing pretraining compute costs by up to 70%. This technique works by first duplicating layers of a smaller pretrained model to form a larger model, and then further pretraining the result. Sign up here!

Andrew Ng

85,714 views • 2 years ago

New short course on Fine-tuning LLMs! Many developers are moving beyond only prompting, to also fine-tuning LLMs - that is, taking a pre-trained model and training it further on your own data, which can deliver superior results inexpensively. In this course, Sharon Zhou, CEO of Lamini (disclosure: I’m a minor shareholder) shows you how to recognize when fine-tuning can be help, and how to train an open-source LLM on your own data. I hope you enjoy the course!

New short course on Fine-tuning LLMs! Many developers are moving beyond only prompting, to also fine-tuning LLMs - that is, taking a pre-trained model and training it further on your own data, which can deliver superior results inexpensively. In this course, Sharon Zhou, CEO of Lamini (disclosure: I’m a minor shareholder) shows you how to recognize when fine-tuning can be help, and how to train an open-source LLM on your own data. I hope you enjoy the course!

Andrew Ng

502,866 views • 3 years ago

New short course on Reinforcement Learning from Human Feedback! RLHF is one of the key techniques that led to the rise of modern LLMs. It is used to align LLMs with human preferences, to make them more honest, helpful and harmless, by (i) learning a reward function that mimics human preferences, as expressed in human-provided labels, then, (ii) tuning an LLM to generate outputs that receive a high reward. In this course, taught by Nikita Namjoshi, Developer Advocate for GenAI at Google Cloud, you'll learn the details of how RLHF works, including how to apply it to tune an LLM for your own applications. You'll also use an open source library to tune a base LLM to align with human preferences expressed in a training set, and evaluate the tuned model by comparing its responses before and after RLHF-tuning. Please sign up here!

New short course on Reinforcement Learning from Human Feedback! RLHF is one of the key techniques that led to the rise of modern LLMs. It is used to align LLMs with human preferences, to make them more honest, helpful and harmless, by (i) learning a reward function that mimics human preferences, as expressed in human-provided labels, then, (ii) tuning an LLM to generate outputs that receive a high reward. In this course, taught by Nikita Namjoshi, Developer Advocate for GenAI at Google Cloud, you'll learn the details of how RLHF works, including how to apply it to tune an LLM for your own applications. You'll also use an open source library to tune a base LLM to align with human preferences expressed in a training set, and evaluate the tuned model by comparing its responses before and after RLHF-tuning. Please sign up here!

Andrew Ng

205,569 views • 2 years ago

Learn to train an LLM with distributed data while ensuring privacy using federated learning in a new two-part short course, Intro to Federated Learning and Federated Fine-tuning of LLMs with Private Data, created with Flower and taught by Daniel J. Beutel and Nicholas Lane. Federated learning allows a single model to be trained across multiple devices, such as phones, or multiple organizations, such as hospitals, without the need to share data to a central server. This two-part course gives you an introduction to federated learning, and then teaches you how to fine-tune your large language model with distributed data using Flower Lab’s open source federated learning framework. You’ll learn: - How to use federated learning to train a variety of models, ranging from speech and vision models to LLMs, across distributed data while offering data privacy options to users and organizations. - Privacy Enhancing Technologies like differential privacy (DP), which obscures individual data by adding calibrated noise to query results. - Two variants of differential privacy - Central and Local - and how to choose depending on your use case. - How to measure and decrease bandwidth usage to make federated learning more practical and efficient with techniques like using pre-trained models and Parameter-Efficient Fine-Tuning - How federated LLM fine-tuning reduces the risk of leaking training data. Sign up here!

Learn to train an LLM with distributed data while ensuring privacy using federated learning in a new two-part short course, Intro to Federated Learning and Federated Fine-tuning of LLMs with Private Data, created with Flower and taught by Daniel J. Beutel and Nicholas Lane. Federated learning allows a single model to be trained across multiple devices, such as phones, or multiple organizations, such as hospitals, without the need to share data to a central server. This two-part course gives you an introduction to federated learning, and then teaches you how to fine-tune your large language model with distributed data using Flower Lab’s open source federated learning framework. You’ll learn: - How to use federated learning to train a variety of models, ranging from speech and vision models to LLMs, across distributed data while offering data privacy options to users and organizations. - Privacy Enhancing Technologies like differential privacy (DP), which obscures individual data by adding calibrated noise to query results. - Two variants of differential privacy - Central and Local - and how to choose depending on your use case. - How to measure and decrease bandwidth usage to make federated learning more practical and efficient with techniques like using pre-trained models and Parameter-Efficient Fine-Tuning - How federated LLM fine-tuning reduces the risk of leaking training data. Sign up here!

Andrew Ng

64,609 views • 2 years ago

$Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.$

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

Alexander Whedon

13,477,513 views • 3 months ago

New course! Generative AI with Large Language Models, created with Amazon Web Services and hosted on Coursera. This course goes deep into the technical foundations of LLMs and how to use them. You can sign up here: You’ll work through the full life-cycle of a generative AI project, and learn specific techniques like RLHF; zero-shot, one-shot, and few-shot learning with LLMs; advanced prompting frameworks like ReAct; even fine-tuning LLMs, and gain hands-on practice with all of these techniques. Instructors Antje Barth Chris Fregly Shelbee Eigenbrode and Mike G Chambers all do incredible Generative AI work at AWS, and have supported many companies to build creative LLM applications. They bring tremendous practical LLM expertise to this course. I'm confident you’ll finish this course with a deeper understanding of how LLMs work, and how to use them. I hope you enjoy the course!

New course! Generative AI with Large Language Models, created with Amazon Web Services and hosted on Coursera. This course goes deep into the technical foundations of LLMs and how to use them. You can sign up here: You’ll work through the full life-cycle of a generative AI project, and learn specific techniques like RLHF; zero-shot, one-shot, and few-shot learning with LLMs; advanced prompting frameworks like ReAct; even fine-tuning LLMs, and gain hands-on practice with all of these techniques. Instructors Antje Barth Chris Fregly Shelbee Eigenbrode and Mike G Chambers all do incredible Generative AI work at AWS, and have supported many companies to build creative LLM applications. They bring tremendous practical LLM expertise to this course. I'm confident you’ll finish this course with a deeper understanding of how LLMs work, and how to use them. I hope you enjoy the course!

Andrew Ng

467,976 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