正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

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.show more

Alex Veremeyenko

98,265 subscribers

16,147 次观看 • 21 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

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

200,950 次观看 • 1 年前

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 次观看 • 1 年前

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:

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:

Andrew Ng

131,069 次观看 • 2 个月前

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

Fireside Alpha

38,370 次观看 • 1 个月前

Micron is going to $4,000 and once you understand what inference actually is, the number stops sounding crazy (Save this). Dylan Patel just said that by 2030, OpenAI and Anthropic alone will need over 100 gigawatts of compute combined and by 2040, we may not even be measuring AI infrastructure in gigawatts anymore. We may be talking about terawatts. Every single one of those gigawatts needs memory to function. Without it, the compute is worthless. Most people heard that and thought about Nvidia but they should be thinking about Micron. Every AI model generating a response has two phases. The first is prefill, processing your prompt which is compute-heavy and the second is decode generating each word one token at a time and that phase is almost entirely memory-bound, not compute-bound. During decode, the GPU's processing units sit idle more than 95% of the time, waiting for data to arrive from memory. Google confirmed it in a research paper that decode-phase bottlenecks are dominated by memory bandwidth and capacity not raw compute. The GPU is not the bottleneck but the memory feeding the GPU is. This matters because inference is now where all the money lives. Training a model happens once, Inference happens billions of times a day every ChatGPT response, every Claude output, every agentic workflow running in the background and every one of those token streams is a billing event tied directly to memory performance. Adding more GPUs does not fix this because GPUs are already underutilized in inference because they are sitting idle waiting on memory. Adding more memory bandwidth and capacity is what directly reduces token cost, reduces latency, and allows the same cluster to serve dramatically more users simultaneously. Longer context windows compound the problem further, a model running a 1 million token context window requires dramatically more memory per session than a 10,000 token window, and every new model generation pushes context longer. The market treats memory as a downstream beneficiary of Nvidia orders. The correct framework is the opposite, Micron is the upstream constraint on how much value every Nvidia GPU can actually generate at inference scale. Micron guided Q4 to $50 billion in revenue, has HBM4 ramping at twice the pace of the prior generation, and CEO Sanjay Mehrotra has said supply will not catch demand before the end of 2027. At 8x forward earnings on $112 projected FY2027 EPS, Micron is the most undervalued infrastructure company in the entire AI stack. Inference is memory. Memory is Micron and the inference ramp has barely started. Milk Road Pro members are already up massively on this position and we're just getting started. If you want the full breakdown of what we're buying and why, come join us for just a dollar using the link below!

Micron is going to $4,000 and once you understand what inference actually is, the number stops sounding crazy (Save this). Dylan Patel just said that by 2030, OpenAI and Anthropic alone will need over 100 gigawatts of compute combined and by 2040, we may not even be measuring AI infrastructure in gigawatts anymore. We may be talking about terawatts. Every single one of those gigawatts needs memory to function. Without it, the compute is worthless. Most people heard that and thought about Nvidia but they should be thinking about Micron. Every AI model generating a response has two phases. The first is prefill, processing your prompt which is compute-heavy and the second is decode generating each word one token at a time and that phase is almost entirely memory-bound, not compute-bound. During decode, the GPU's processing units sit idle more than 95% of the time, waiting for data to arrive from memory. Google confirmed it in a research paper that decode-phase bottlenecks are dominated by memory bandwidth and capacity not raw compute. The GPU is not the bottleneck but the memory feeding the GPU is. This matters because inference is now where all the money lives. Training a model happens once, Inference happens billions of times a day every ChatGPT response, every Claude output, every agentic workflow running in the background and every one of those token streams is a billing event tied directly to memory performance. Adding more GPUs does not fix this because GPUs are already underutilized in inference because they are sitting idle waiting on memory. Adding more memory bandwidth and capacity is what directly reduces token cost, reduces latency, and allows the same cluster to serve dramatically more users simultaneously. Longer context windows compound the problem further, a model running a 1 million token context window requires dramatically more memory per session than a 10,000 token window, and every new model generation pushes context longer. The market treats memory as a downstream beneficiary of Nvidia orders. The correct framework is the opposite, Micron is the upstream constraint on how much value every Nvidia GPU can actually generate at inference scale. Micron guided Q4 to $50 billion in revenue, has HBM4 ramping at twice the pace of the prior generation, and CEO Sanjay Mehrotra has said supply will not catch demand before the end of 2027. At 8x forward earnings on $112 projected FY2027 EPS, Micron is the most undervalued infrastructure company in the entire AI stack. Inference is memory. Memory is Micron and the inference ramp has barely started. Milk Road Pro members are already up massively on this position and we're just getting started. If you want the full breakdown of what we're buying and why, come join us for just a dollar using the link below!

Milk Road AI

128,678 次观看 • 1 个月前

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 次观看 • 4 个月前

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

270,496 次观看 • 1 个月前

SK Hynix's Hoshik Kim admits high bandwidth memory "is not the answer" to the memory-wall problem, and SK is working on different solutions to solve the bottleneck. "No, actually, I kind of agree with Pat. So HBM is not the answer." [SK Hynix is admitting that HBM is a bad memory?] "Nothing is a bad memory. It's a good memory, but it's not the final answer. So actually, HBM will not solve the memory-wall problem. Memory-wall problem is an inherent AI problem, which you cannot avoid." "So we try our best to provide higher bandwidth with this memory, with HBM technology even though it has some limitation, as Pat already mentioned, with thermal, IOs, cost, a lot of stuff. "But that's the only thing we got right now to help ease this memory-wall problem, memory-bottleneck problem. But at the same time, we are working in different directions to solve this memory-wall problem with different solutions."

SK Hynix's Hoshik Kim admits high bandwidth memory "is not the answer" to the memory-wall problem, and SK is working on different solutions to solve the bottleneck. "No, actually, I kind of agree with Pat. So HBM is not the answer." [SK Hynix is admitting that HBM is a bad memory?] "Nothing is a bad memory. It's a good memory, but it's not the final answer. So actually, HBM will not solve the memory-wall problem. Memory-wall problem is an inherent AI problem, which you cannot avoid." "So we try our best to provide higher bandwidth with this memory, with HBM technology even though it has some limitation, as Pat already mentioned, with thermal, IOs, cost, a lot of stuff. "But that's the only thing we got right now to help ease this memory-wall problem, memory-bottleneck problem. But at the same time, we are working in different directions to solve this memory-wall problem with different solutions."

Fireside Alpha

140,200 次观看 • 20 天前

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Santiago

84,747 次观看 • 2 年前

Chamath: Two terms you need to pay attention to in AI are Prefill and Decode “There's two terms that I think you're going to hear a ton about over these next few years.” “The first term is prefill, and the next is decode.” “What prefill and decode are, are two very distinct ways of how models think, and how a model goes through the process of answering a question that you ask it.” “And so when you send a prompt to AI, what happens is that the model processes it. This is called the reading phase or prefill.” “It reads your entire prompt all at once. And then it does a bunch of math, calculates all these relationships between all the words, and it stores them in temporary memory.” “The problem is that this is really compute bound. So it requires massive brute force. And Nvidia GPUs crush here.” “And their architecture is designed for massive parallel processing, which makes them really amazing at digesting these long prompts.” “So the problem just gets bigger and bigger, Nvidia just completely dominates.” “But the next phase though, this critical phase, the decode phase, is the writing phase, right?” “So the model starts to generate a response, you ask it a question and its response, one token at a time.” “And then to pick the next token to pick the next word, it has to look back at everything it has said already so that it doesn't hallucinate.” “The problem is that this is incredibly memory bandwidth constrained.” “And in our architecture, a long time ago, we made these design decisions from day one.” “And so what we did was we took a very different architectural approach, we took a very conservative process technology. We weren't pushing the boundaries of physics.” “And we used a lot of what's called SRAM. So memory on the chip so that we could do this decode thing as well or better than everybody else.” “And so now when you put these two things together, I just think it's going to create a huge acceleration in the ability for this entire infrastructure layer to get much cheaper and much more valuable, which I suspect then it'll have a lot more developer pull, you'll get a lot more applications being built, billions and billions of more people using it.”

Chamath: Two terms you need to pay attention to in AI are Prefill and Decode “There's two terms that I think you're going to hear a ton about over these next few years.” “The first term is prefill, and the next is decode.” “What prefill and decode are, are two very distinct ways of how models think, and how a model goes through the process of answering a question that you ask it.” “And so when you send a prompt to AI, what happens is that the model processes it. This is called the reading phase or prefill.” “It reads your entire prompt all at once. And then it does a bunch of math, calculates all these relationships between all the words, and it stores them in temporary memory.” “The problem is that this is really compute bound. So it requires massive brute force. And Nvidia GPUs crush here.” “And their architecture is designed for massive parallel processing, which makes them really amazing at digesting these long prompts.” “So the problem just gets bigger and bigger, Nvidia just completely dominates.” “But the next phase though, this critical phase, the decode phase, is the writing phase, right?” “So the model starts to generate a response, you ask it a question and its response, one token at a time.” “And then to pick the next token to pick the next word, it has to look back at everything it has said already so that it doesn't hallucinate.” “The problem is that this is incredibly memory bandwidth constrained.” “And in our architecture, a long time ago, we made these design decisions from day one.” “And so what we did was we took a very different architectural approach, we took a very conservative process technology. We weren't pushing the boundaries of physics.” “And we used a lot of what's called SRAM. So memory on the chip so that we could do this decode thing as well or better than everybody else.” “And so now when you put these two things together, I just think it's going to create a huge acceleration in the ability for this entire infrastructure layer to get much cheaper and much more valuable, which I suspect then it'll have a lot more developer pull, you'll get a lot more applications being built, billions and billions of more people using it.”

The All-In Podcast

567,546 次观看 • 7 个月前

I had to test it myself to believe this unreal inference speed. 3,000 tokens/s for 1 user on standard datacenter GPUs. They leveraged a hidden efficiency gap in how GPUs generate tokens. Kog just achieved 3,000 tokens/s on 8× AMD MI300X GPUs and 2,100 on 8× NVIDIA H200 (FP16, no speculative decoding). Their tech preview is on a 2B model, and they show how their techniques will scale to large frontier MoE models at similar speeds. That's a huge number because normal low-batch GPU decoding for 2B to 8B models is usually closer to 100 to 300 tokens/s per request, so Kog is claiming something like a 10X to 30X jump in the speed one user actually feels. Their trick: they are getting the speed by treating LLM decoding as a memory streaming problem, not mainly a math problem. For 1 user at batch size 1, the GPU is not doing big, efficient matrix-matrix work like in training or large-batch serving; it is repeatedly pulling the model’s active weights from high-bandwidth memory for each new token, so speed depends on how smoothly those weights keep flowing. Normal inference stacks keep breaking that flow. They run many separate GPU programs for different parts of the model, move intermediate results through memory, wait at synchronization points, talk back to the CPU for scheduling or sampling, and then repeat this token after token. Kog’s answer is to co-design 3 things that are usually tuned separately: the runtime, the low-level GPU code, and the model architecture. The biggest engineering move is the monokernel, where the whole decode pass runs as 1 persistent GPU-resident program, including sampling, so the system does not keep stopping for kernel launches, CPU scheduling, and intermediate memory round trips. They also rebuilt synchronization, because their own measurements say grid sync was eating around 35% of token-generation time; instead of making every compute unit wait at a broad barrier, each unit waits only for the exact data it needs. On AMD MI300X, they also map memory access around the chiplet layout, because memory latency changes depending on which die makes the request. Then their Laneformer model uses Delayed Tensor Parallelism, which lets cross-GPU communication happen in the background instead of blocking every layer.

I had to test it myself to believe this unreal inference speed. 3,000 tokens/s for 1 user on standard datacenter GPUs. They leveraged a hidden efficiency gap in how GPUs generate tokens. Kog just achieved 3,000 tokens/s on 8× AMD MI300X GPUs and 2,100 on 8× NVIDIA H200 (FP16, no speculative decoding). Their tech preview is on a 2B model, and they show how their techniques will scale to large frontier MoE models at similar speeds. That's a huge number because normal low-batch GPU decoding for 2B to 8B models is usually closer to 100 to 300 tokens/s per request, so Kog is claiming something like a 10X to 30X jump in the speed one user actually feels. Their trick: they are getting the speed by treating LLM decoding as a memory streaming problem, not mainly a math problem. For 1 user at batch size 1, the GPU is not doing big, efficient matrix-matrix work like in training or large-batch serving; it is repeatedly pulling the model’s active weights from high-bandwidth memory for each new token, so speed depends on how smoothly those weights keep flowing. Normal inference stacks keep breaking that flow. They run many separate GPU programs for different parts of the model, move intermediate results through memory, wait at synchronization points, talk back to the CPU for scheduling or sampling, and then repeat this token after token. Kog’s answer is to co-design 3 things that are usually tuned separately: the runtime, the low-level GPU code, and the model architecture. The biggest engineering move is the monokernel, where the whole decode pass runs as 1 persistent GPU-resident program, including sampling, so the system does not keep stopping for kernel launches, CPU scheduling, and intermediate memory round trips. They also rebuilt synchronization, because their own measurements say grid sync was eating around 35% of token-generation time; instead of making every compute unit wait at a broad barrier, each unit waits only for the exact data it needs. On AMD MI300X, they also map memory access around the chiplet layout, because memory latency changes depending on which die makes the request. Then their Laneformer model uses Delayed Tensor Parallelism, which lets cross-GPU communication happen in the background instead of blocking every layer.

Rohan Paul

13,244 次观看 • 2 个月前

DeepMind CEO Demis Hassabis says the AI bottleneck is already moving from model ideas to physical supply chains. "How much of the restriction at this point is physical? I am talking memory shortage, GPU shortage, and electricity." Demis: "That is constraining a lot of the deployment for sure. We see so much more demand for our models - Gemini and the other models that we have - than we can possibly serve." Then the less obvious part: "It does constrain a little bit the research, because you need a lot of chips to be able to experiment on new ideas at big enough scale that you can actually see if they are going to work." And on memory: "The whole supply chain is kind of strained... It still in the end actually comes down to a few suppliers of a few of the key components. Anywhere where there is some kind of constraint on the capacity, there is a sort of choke point." Important framing: The next AI race is not just who has the best model architecture. It is who can secure memory, chips, power, suppliers, and enough experimental capacity to find the next breakthrough before everyone else.

DeepMind CEO Demis Hassabis says the AI bottleneck is already moving from model ideas to physical supply chains. "How much of the restriction at this point is physical? I am talking memory shortage, GPU shortage, and electricity." Demis: "That is constraining a lot of the deployment for sure. We see so much more demand for our models - Gemini and the other models that we have - than we can possibly serve." Then the less obvious part: "It does constrain a little bit the research, because you need a lot of chips to be able to experiment on new ideas at big enough scale that you can actually see if they are going to work." And on memory: "The whole supply chain is kind of strained... It still in the end actually comes down to a few suppliers of a few of the key components. Anywhere where there is some kind of constraint on the capacity, there is a sort of choke point." Important framing: The next AI race is not just who has the best model architecture. It is who can secure memory, chips, power, suppliers, and enough experimental capacity to find the next breakthrough before everyone else.

Karl Mehta

25,716 次观看 • 1 个月前

We asked James Wang why Cerebras can run large models so quickly. His answer: Inference speed is mostly a memory problem. Instead of constantly pulling weights from external memory, Cerebras splits the model across multiple wafers and pipelines the layers together. “Inference is all about memory bandwidth.” “You don’t want to store weights external to the wafer, because the second it’s outside, it becomes much slower.” “We put the weights on the inside.” “The team built a new software stack that basically lets us split the models by layer and store them layer by layer across multiple wafers.” “That allows us to never have to read from external memory, and that’s what makes it so fast.” “For context, Claude is about 100 tokens per second.”

We asked James Wang why Cerebras can run large models so quickly. His answer: Inference speed is mostly a memory problem. Instead of constantly pulling weights from external memory, Cerebras splits the model across multiple wafers and pipelines the layers together. “Inference is all about memory bandwidth.” “You don’t want to store weights external to the wafer, because the second it’s outside, it becomes much slower.” “We put the weights on the inside.” “The team built a new software stack that basically lets us split the models by layer and store them layer by layer across multiple wafers.” “That allows us to never have to read from external memory, and that’s what makes it so fast.” “For context, Claude is about 100 tokens per second.”

MTS

27,935 次观看 • 2 个月前

We've been testing Memory (short-term and long-term) on Perplexity for a while. The results are great, and we are rolling it out widely. You can ask personalized questions, questions about past chats, and use any model or search mode with personal context (both apps and web).

We've been testing Memory (short-term and long-term) on Perplexity for a while. The results are great, and we are rolling it out widely. You can ask personalized questions, questions about past chats, and use any model or search mode with personal context (both apps and web).

Aravind Srinivas

96,497 次观看 • 8 个月前

Chase Hughes just revealed a wild way to edit traumatic memories using hypnosis on the Joe Rogan Experience. HUGHES: “I figured out a way to edit memory… I can take you back and change your perspective on trauma.” “You go back to that event when you got kicked in the nuts and everybody laughed at you in elementary school… and you can reprocess that memory in a very short amount of time as an adult with the perspective of an adult.” “Instead of modifying the memory like ‘oh that never happened’ — the memory stays. The perspective changes.” “And it’s exactly what psychedelics do. This massive perspective shift on memory… their whole life can be different after that day.” He says he’s done this hundreds of times and it creates real downstream effects.

Chase Hughes just revealed a wild way to edit traumatic memories using hypnosis on the Joe Rogan Experience. HUGHES: “I figured out a way to edit memory… I can take you back and change your perspective on trauma.” “You go back to that event when you got kicked in the nuts and everybody laughed at you in elementary school… and you can reprocess that memory in a very short amount of time as an adult with the perspective of an adult.” “Instead of modifying the memory like ‘oh that never happened’ — the memory stays. The perspective changes.” “And it’s exactly what psychedelics do. This massive perspective shift on memory… their whole life can be different after that day.” He says he’s done this hundreds of times and it creates real downstream effects.

Defiant Ghost

26,912 次观看 • 1 个月前

HERMES AGENT NOW RUNS ON AN 8GB LAPTOP GPU JUST AS EASILY AS IT RUNS ON A 128GB MINI PC Nous Research shipped the official Hermes Agent Desktop App this week. Someone pointed it at a local llama server running on an RTX 4060 with 16GB system RAM. The integration took two minutes The model behind it: Gemma 4 26B MoE, QAT quantized, running on 8GB of VRAM. A 60k token prompt held a stable 20 tokens a second, flat, no slowdown as context grew. The flags were nothing exotic, just -cmoe -c 248000 on llama.cpp What that 8GB setup does out of the box: reads and patches its own code, runs it in a terminal, debugs errors, manages GitHub repos, spawns sub-agents for parallel work. Browses the web with vision to debug a UI. Schedules cron jobs in plain language. Connects to Notion, Google Workspace, Linear, and Obsidian to manage tasks on its own That's the same agent layer running on a Minisforum MS-S1 MAX with 128GB of unified memory, 96GB of it to the GPU, holding a 120B model at 56 tokens a second instead of a 26B model at 20. Same software, same tool execution, same zero API key. The only thing that changes between an $800 laptop and a $2,000 mini PC is how big a model you can afford to run underneath it The barrier to running a real autonomous agent locally didn't just drop. It dropped all the way down to hardware most people already own

HERMES AGENT NOW RUNS ON AN 8GB LAPTOP GPU JUST AS EASILY AS IT RUNS ON A 128GB MINI PC Nous Research shipped the official Hermes Agent Desktop App this week. Someone pointed it at a local llama server running on an RTX 4060 with 16GB system RAM. The integration took two minutes The model behind it: Gemma 4 26B MoE, QAT quantized, running on 8GB of VRAM. A 60k token prompt held a stable 20 tokens a second, flat, no slowdown as context grew. The flags were nothing exotic, just -cmoe -c 248000 on llama.cpp What that 8GB setup does out of the box: reads and patches its own code, runs it in a terminal, debugs errors, manages GitHub repos, spawns sub-agents for parallel work. Browses the web with vision to debug a UI. Schedules cron jobs in plain language. Connects to Notion, Google Workspace, Linear, and Obsidian to manage tasks on its own That's the same agent layer running on a Minisforum MS-S1 MAX with 128GB of unified memory, 96GB of it to the GPU, holding a 120B model at 56 tokens a second instead of a 26B model at 20. Same software, same tool execution, same zero API key. The only thing that changes between an $800 laptop and a $2,000 mini PC is how big a model you can afford to run underneath it The barrier to running a real autonomous agent locally didn't just drop. It dropped all the way down to hardware most people already own

NO1ennn

40,079 次观看 • 1 个月前

Everyone wrote Apple off as the AI loser, but one hardware spec might flip that story upside down (Save this). @jason called Apple a screaming buy on the back of a single chip detail. The rumored M7 Ultra, expected around 2028, is designed to support up to 1.5TB of unified memory, enough to run frontier class trillion parameter AI models locally, with no cloud required. The Street's bear case on Apple is straightforward. Apple has no frontier model of its own, Siri has stumbled for years and the company effectively rents OpenAI's models for its hardest queries. That narrative treats Apple as the one Magnificent Seven name that missed the AI wave entirely but the bull case flips that framing on its head. If frontier AI models keep shrinking and getting cheaper to run, Apple doesn't need the smartest model in the world, it just needs to own the device that model runs on. And unified memory is the mechanism that makes this possible. Unlike traditional systems where the CPU and GPU each need separate memory, Apple's architecture lets the CPU, GPU and Neural Engine draw from one shared pool. A fully specced M7 Ultra could theoretically run something on the scale of a 1.2 trillion parameter model locally and that capability plugs directly into the one advantage Apple has spent over a decade building: privacy. Apple has already shipped Private Cloud Compute, a system designed so even Apple can't access user data processed off device. Apple doubled down on this at WWDC 2026, framing on device privacy as non-negotiable while rivals default to the cloud. If the best AI models get small enough to run on Apple silicon, the moat stops being the model and becomes the hardware it has to sit on. Milk Road Pro remains bullish on Apple and it remains as one of our core positions, if you want the full thesis + our full AI trades, come join us using the link below for just a $1.

Everyone wrote Apple off as the AI loser, but one hardware spec might flip that story upside down (Save this). @jason called Apple a screaming buy on the back of a single chip detail. The rumored M7 Ultra, expected around 2028, is designed to support up to 1.5TB of unified memory, enough to run frontier class trillion parameter AI models locally, with no cloud required. The Street's bear case on Apple is straightforward. Apple has no frontier model of its own, Siri has stumbled for years and the company effectively rents OpenAI's models for its hardest queries. That narrative treats Apple as the one Magnificent Seven name that missed the AI wave entirely but the bull case flips that framing on its head. If frontier AI models keep shrinking and getting cheaper to run, Apple doesn't need the smartest model in the world, it just needs to own the device that model runs on. And unified memory is the mechanism that makes this possible. Unlike traditional systems where the CPU and GPU each need separate memory, Apple's architecture lets the CPU, GPU and Neural Engine draw from one shared pool. A fully specced M7 Ultra could theoretically run something on the scale of a 1.2 trillion parameter model locally and that capability plugs directly into the one advantage Apple has spent over a decade building: privacy. Apple has already shipped Private Cloud Compute, a system designed so even Apple can't access user data processed off device. Apple doubled down on this at WWDC 2026, framing on device privacy as non-negotiable while rivals default to the cloud. If the best AI models get small enough to run on Apple silicon, the moat stops being the model and becomes the hardware it has to sit on. Milk Road Pro remains bullish on Apple and it remains as one of our core positions, if you want the full thesis + our full AI trades, come join us using the link below for just a $1.

Milk Road AI

37,388 次观看 • 26 天前

Microsoft ran its graph system against vector RAG on 8k, 120k, and a full million token context window, and the million token giant lost Jonathan Larson from Microsoft Research put up the numbers → LazyGraphRAG won 92, 90, and 91 percent of data-local questions, the exact place plain RAG was supposed to be strong → and did it at a tenth of the cost of the million token run the lesson buried in the benchmark → a bigger context window is not memory, structured memory is throwing more tokens at the problem was just the expensive way to be wrong open source on GitHub, the lesson a $1500 course would charge you for

Microsoft ran its graph system against vector RAG on 8k, 120k, and a full million token context window, and the million token giant lost Jonathan Larson from Microsoft Research put up the numbers → LazyGraphRAG won 92, 90, and 91 percent of data-local questions, the exact place plain RAG was supposed to be strong → and did it at a tenth of the cost of the million token run the lesson buried in the benchmark → a bigger context window is not memory, structured memory is throwing more tokens at the problem was just the expensive way to be wrong open source on GitHub, the lesson a $1500 course would charge you for

NO1ennn

29,076 次观看 • 12 天前

Dylan Patel says the memory shortage is not a normal cycle "Memory capacity is only growing 20, 30% a year for the next three years. And yet, demand is doubling, is doubling...memory prices are going to keep soaring." "Our point there was memory isn't a shortage, and this is not a short-term shortage. It's a shortage that's going to last years." "Memory we've had the pricing go up like 4x and it's going to go up another 2x, 3x again, in addition to capacity growth. We talked about how KV cache was going to explode because of reasoning. And therefore memory was going to be the biggest winner." "We're still not at 85, 90% gross margins for memory, but we'll get there. And then at some point from there, it'll also halve back down to like 70s or maybe even lower."

Dylan Patel says the memory shortage is not a normal cycle "Memory capacity is only growing 20, 30% a year for the next three years. And yet, demand is doubling, is doubling...memory prices are going to keep soaring." "Our point there was memory isn't a shortage, and this is not a short-term shortage. It's a shortage that's going to last years." "Memory we've had the pricing go up like 4x and it's going to go up another 2x, 3x again, in addition to capacity growth. We talked about how KV cache was going to explode because of reasoning. And therefore memory was going to be the biggest winner." "We're still not at 85, 90% gross margins for memory, but we'll get there. And then at some point from there, it'll also halve back down to like 70s or maybe even lower."

Fireside Alpha

140,289 次观看 • 1 个月前

SK Hynix recently reached a $1 trillion market cap. SanDisk started the year at $35B and is now worth $230B. Dylan (Dylan Patel) on why memory prices will still double or triple from here: "People are like, the memory story is overplayed. No, you don't get it. DRAM will double or triple from here still, because that's how much capacity is required. Memory can only grow capacity low double digit percentages a year. 20-30% a year, even less for NAND, a little bit higher for DRAM. Even though the demand signal was very strong at the end of 2025, the memory companies immediately started reacting. None of that incremental capacity really gets here until 2028, even if they wanted to build as fast as possible. So the result is memory prices have gone through the roof. They're going to double and triple again. DRAM, especially. And they have to steal capacity from somewhere else. And the only way to steal capacity from somewhere else in a capitalist economy is demand destruction via higher pricing." Publish date: April 23, 2026

SK Hynix recently reached a $1 trillion market cap. SanDisk started the year at $35B and is now worth $230B. Dylan (Dylan Patel) on why memory prices will still double or triple from here: "People are like, the memory story is overplayed. No, you don't get it. DRAM will double or triple from here still, because that's how much capacity is required. Memory can only grow capacity low double digit percentages a year. 20-30% a year, even less for NAND, a little bit higher for DRAM. Even though the demand signal was very strong at the end of 2025, the memory companies immediately started reacting. None of that incremental capacity really gets here until 2028, even if they wanted to build as fast as possible. So the result is memory prices have gone through the roof. They're going to double and triple again. DRAM, especially. And they have to steal capacity from somewhere else. And the only way to steal capacity from somewhere else in a capitalist economy is demand destruction via higher pricing." Publish date: April 23, 2026

Invest Like the Best

105,024 次观看 • 1 个月前