111–120 of 240 posts
Earlier quoted context omitted.
This isn't quite right: it'll run with the full model loaded to RAM, swapping in the experts as it needs. It has turned out in the past that experts can be stable across more than one token so you're not swapping as much as you'd think. I don't know if that's been confirmed to still be true on recent MoEs, but I wouldn't be surprised.
What you are describing would be uselessly slow and nobody does that.
Earlier quoted context omitted.
This isn't quite right: it'll run with the full model loaded to RAM, swapping in the experts as it needs. It has turned out in the past that experts can be stable across more than one token so you're not swapping as much as you'd think. I don't know if that's been confirmed to still be true on recent MoEs, but I wouldn't be surprised.
What you are describing would be uselessly slow and nobody does that.
Coolest part of Qwen3-Next, in my opinion, (after the linear attention parts) is that they do MTP without adding another un-embedding matrix. Deepseek R1 also has a MTP layer (layer 61) https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/main/mod... But Deepseek R1 adds embed_tokens and shared_head.head tensors, which are [129280, 7168] or about 2GB in size at FP8. Qwen3-Next doesn't have that: https://huggingface.co…
Could someone kindly point to a convenient all-on-one ELI5 of all these words? :')
LLMs take your input, upscale it into a very high dimensional space, and then downscale it back to 1D at the end. This 1D list is interpreted as a list of probabilities -- one for each word in your vocabulary. i.e f(x) = downscale(upscale(x)). Each of downscale() and upscale() are parameterized (billions of params). I see you have a gamedev background, so as an example: bezier curves are parameterized functions where bezier handles are the parameters. During training, these parameters are continuously adjusted so that the output of the overall function gets closer to the expected result. Neural networks are just really flexible functions for which you can choose parameters to get any expected result, provided you have enough of them (similar to bezier curves in this regard).
---
When training, you make an LLM learn that
I use arch = downscale(upscale(I use))
If you want to predict the next word after that, you do next in sequence the following:
I use arch btw = downscale(upscale(I use arch))
Now, multi-token prediction is having two downscale functions, one for each of the next two words, and learning it that way, basically, you have a second downscale2() that learns how to predict the next-to-next word.
i.e in parallel:
I use arch = downscale1(upscale(I use))
I use ____ btw = downscale2(upscale(I use))
However, this way you'll need twice the number of parameters downscale needs. And if you want to predict more tokens ahead you'll need even more parameters.
What Qwen has done, is instead of downscale1 and downscale2 being completely separately parameterized functions, they set downscale1(.) = lightweight1(downscale_common(.)) and downscale2(.) = lightweight2(downscale_common(.)). This is essentially betting that a lot of the logic is common and the difference between predicting the next and next-to-next token can be captured in one lightweight function each. Lightweight here, means less parameters. The bet paid off.
So overall, you save params.
Concretely,
Before: downscale1.params + downscale2.params
After: downscale_common.params + lightweight1.params + lightweight2.params
Edit: its actually downscale_common(lightweight()) and not the other way around as I have written above. Doesn't change the crux of the answer, but just including this for clarity.
Coolest part of Qwen3-Next, in my opinion, (after the linear attention parts) is that they do MTP without adding another un-embedding matrix. Deepseek R1 also has a MTP layer (layer 61) https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/main/mod... But Deepseek R1 adds embed_tokens and shared_head.head tensors, which are [129280, 7168] or about 2GB in size at FP8. Qwen3-Next doesn't have that: https://huggingface.co…
Could someone kindly point to a convenient all-on-one ELI5 of all these words? :')
Unlike many disciplines, AI is an arena that doesn't have a lot of intuitive simplified models that are accurate -- most of the simplified models available do not accurately describe what's going on enough to reason about and understand them. So, you just have to start reading!
The same week Oracle is forecasting huge data center demand and the stock is rallying. If these 10x gains in efficiency hold true then this could lead to a lot less demand for Nvidia, Oracle, Coreweave etc
Absolutely not; the trends have proven that people will just pay for the best quality they can get, and keep paying roughly the same money. Every time a new model is released, people abandon the old, lower quality model (even when it’s priced less), and instead prefer to pay the same for a better model. The same will happen with this.
The same week Oracle is forecasting huge data center demand and the stock is rallying. If these 10x gains in efficiency hold true then this could lead to a lot less demand for Nvidia, Oracle, Coreweave etc
Absolutely not; the trends have proven that people will just pay for the best quality they can get, and keep paying roughly the same money. Every time a new model is released, people abandon the old, lower quality model (even when it’s priced less), and instead prefer to pay the same for a better model. The same will happen with this.
Earlier quoted context omitted.
https://en.wikipedia.org/wiki/Jevons_paradox
Sure but where is the demand going to come from? LLMs are already in every google search, in Whatsapp/Messenger, throughout Google workspace, Notion, Slack, etc. ChatGPT already has a billion users. Plus penetration is already very high in the areas where they are objectively useful: programming, customer care etc. I just don't see where the 100-1000x demand comes from to offset this. Would be happy to hear other vie…
Earlier quoted context omitted.
https://en.wikipedia.org/wiki/Jevons_paradox
Sure but where is the demand going to come from? LLMs are already in every google search, in Whatsapp/Messenger, throughout Google workspace, Notion, Slack, etc. ChatGPT already has a billion users. Plus penetration is already very high in the areas where they are objectively useful: programming, customer care etc. I just don't see where the 100-1000x demand comes from to offset this. Would be happy to hear other vie…
Case in point; I'd like something that realtime assesses all the sensors and API endpoints of stuff in my home and as needed bubbles up summaries, diaries, and emergency alerts. Right now that's probably a single H200, and well out of my "value range". The number of people in the world that do this now at scale is almost certainly less than 50k.
If that inference cost went to 1%, then a) I'd be willing to pay it, and b) there'd be enough of a market that a company could make money integrating a bunch of tech into a simple deployable stack, and therefore c) a lot more people would want it, likely enough to drive more than 50k H200s worth of inference demand.
Alibaba keeps releasing gold content I just tried Qwen3-Next-80B-A3B on Qwen chat, and it's fast! The quality seem to match Qwen3-235B-A22B. Quite impressive how they achieved this. Can't wait for the benchmarks at Artificial analysis According to Qwen Chat, Qwen3-Next has the following limits: Maximum context length: 262,144 tokens Max summary generation length: 32,768 tokens This is 2x higher on context length and…
My take on long context for many frontier models is not about support but the accuracy drops drastically as you increase the context. Even if a model claims to support 10M context, reality is it doesn’t perform well when you saturate. Curious to hear others perspective on this
Earlier quoted context omitted.
Strong disagree. Llama 4's release in 2025 is (deservedly) panned, but Llama 3.1 405b does not deserve that slander. https://artificialanalysis.ai/#frontier-language-model-intel... Do not compare 2024 models to the current cutting edge. At the time, Llama 3.1 405b was the very first open source (open weights) model to come close to the closed source cutting edge. It was very very close in performance to GPT-4o and Cl…
He is definitely talking about Llama4.
> dense
> 405B model
Llama4 does not match any of these details. Maybe the commenter thinks their comment is about Llama4 (I don't see a reason to believe so) but readers familiar with these details know they are referring to Llama3.1.