alt.hn

3/25/2026 at 5:00:54 AM

TurboQuant: Redefining AI efficiency with extreme compression

 https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/

by ray__

3/25/2026 at 7:47:28 AM

This is a great development for KV cache compression. I did notice a missing citation in the related works regarding the core mathematical mechanism, though. The foundational technique of applying a geometric rotation prior to extreme quantization, specifically for managing the high-dimensional geometry and enabling proper bias correction, was introduced in our NeurIPS 2021 paper, "DRIVE" (https://proceedings.neurips.cc/paper/2021/hash/0397758f8990c...). We used this exact rotational approach and a similar bias correction mechanism to achieve optimal distributed mean estimation. I also presented this work and subsequent papers in a private invited talk at Google shortly after publication. Given the strong theoretical overlap with the mechanisms in TurboQuant and PolarQuant, I hope to see this prior art acknowledged in the upcoming camera-ready versions.

by amitport

3/25/2026 at 7:02:58 AM

This is the worst lay-people explanation of an AI component I have seen in a long time. It doesn't even seem AI generated.

by benob

3/25/2026 at 8:02:07 AM

It is AI generated. Or was written by someone a bit far from the technical advances IMHO. The Johnson-Lindenstrauss Lemma is a very specific and powerful concept, when in the article the QLJ explanation is vacuous. A knowledgeable human would not have left the reader wanting for how that relates to the Lemma.

by BenoitP

3/25/2026 at 7:04:42 AM

I think it is though-

“ TurboQuant, QJL, and PolarQuant are more than just practical engineering solutions; they’re fundamental algorithmic contributions backed by strong theoretical proofs. These methods don't just work well in real-world applications; they are provably efficient and operate near theoretical lower bounds.”

by spencerflem

3/25/2026 at 10:26:09 AM

Genius new idea: replace the em-dashes with semicolons so it looks less like AI.

by NoahZuniga

3/25/2026 at 10:30:50 AM

I read "this clever step" and immediately came to the comments to see if anyone picked up on it.

It reads like a pop science article while at the same time being way too technical to be a pop science article.

Turing test ain't dead yet.

by zarzavat

3/25/2026 at 7:09:43 AM

Maybe they quantized a bit too much the model parameters...

by benob

3/25/2026 at 8:08:56 AM

I also instinctively reacted to that fragment, but at this point I think this is overreacting to a single expression. It's not just a normal thing to say in English, it's something people have been saying for a long time before LLMs existed.

by integralid

3/25/2026 at 10:19:57 AM

Another instinctual reaction here. This specific formulation pops out of AI all the time, there might as well have been an emdash in the title

by g-mork

3/25/2026 at 8:31:23 AM

There are tells all over the page:

> Redefining AI efficiency with extreme compression

"Redefine" is a favorite word of AI. Honestly no need to read further.

> the key-value cache, a high-speed "digital cheat sheet" that stores frequently used information under simple labels

No competent engineer would describe a cache as a "cheat sheet". Cheat sheets are static, but caches dynamically update during execution. Students don't rewrite their cheat sheets during the test, do they? LLMs love their inaccurate metaphors.

> QJL: The zero-overhead, 1-bit trick

> It reduces each resulting vector number to a single sign bit (+1 or -1). This algorithm essentially creates a high-speed shorthand that requires zero memory overhead.

Why does it keep emphasizing zero overhead? Why is storing a single bit a "trick?" Either there's currently an epidemic of algorithms that use more than one bit to store a bit, or the AI is shoving in extra plausible-sounding words to pad things out. You decide which is more likely.

It's 1:30am and I can't sleep, and I still regret wasting my time on this slop.

by nvme0n1p1

3/25/2026 at 9:25:58 AM

Looks like Google canned all their tech writers just to pivot the budget into H100s for training these very same writers

by veunes

3/25/2026 at 8:43:35 AM

There is also the possibility that the article when through the hands of the company's communication department which has writers that probably write at LLM level.

by pqs

3/25/2026 at 10:18:14 AM

[dead]

by awesomelvin

3/25/2026 at 9:46:40 AM

The gap between how this is described in the paper vs the blog post is pretty wide. Would be nice to see more accessible writing from research teams — not everyone reading is a ML engineer

by zeeshana07x

3/25/2026 at 10:10:03 AM

Agreed. The practical implications are often more interesting than the math anyway — smaller models running locally means you can afford to run multiple models in parallel for cross-validation, which changes how you approach tasks like code analysis or bug detection.

by dev_tools_lab

3/25/2026 at 9:58:53 AM

These are very different media types with very different goals.

by om8

3/25/2026 at 6:42:33 AM

I did not understand what polarQuant is.

Is is something like pattern based compression where the algorithm finds repeating patterns and creates an index of those common symbols or numbers?

by bluequbit

3/25/2026 at 6:59:44 AM

https://mesuvash.github.io/blog/2026/turboquant-interactive/ has a little visualisation

by Maxious

3/25/2026 at 7:11:23 AM

I like the visualization, but I don’t understand the grid quantization. If every point is on the unit circle aren’t all the center grid cords unused?

by spencerflem

3/25/2026 at 7:39:08 AM

i think grid can be a surface of the unit sphere

by vincnetas

3/25/2026 at 6:56:57 AM

1. Efficient recursive transform of kv embeddings into polar coordinates 2. Quantize resulting angles without the need for explicit normalization. This saves memory via key insight: angles follow a distribution and have analytical form.

by mrugge

3/25/2026 at 7:15:46 AM

Reminds me vaguely of Burrows-Wheeler transformations in bzip2.

by quotemstr

3/25/2026 at 8:56:27 AM

The way I understand it, it's a way of compressing vectors by switching from their per-component representation to polar coordinates representation, where the nearby vectors are clumped together to a single line, allowing to describe them by different lengths

by viktorcode

3/25/2026 at 7:32:11 AM

Aren’t polar coordinates still n-1 + 1 for radius for n-dim vector? If so I understand that angles can be quantized better but when radius r is big the error is large for highly quantized angles right? What am I missing?

by moktonar

3/25/2026 at 7:34:18 AM

r is a single value per vector. You don't have to quantize it, you can keep it and quantize the billion+ other coordinates of the vector.

by amitport

3/25/2026 at 8:12:05 AM

What they're saying is that the error for a vector increases with r, which is true.

Trivially, with r=0, the error is 0, regardless of how heavily the direction is quantized. Larger r means larger absolute error in the reconstructed vector.

by mungoman2

3/25/2026 at 8:26:14 AM

Yes, the important part is that the normalized error does not increase with the dimension of the vector (which does happen when using biased quantizers)

It is expected that bigger vectors have proportionally bigger error, nothing can be done by the quantizer about that.

by amitport

3/25/2026 at 7:46:08 AM

I'm somewhat at a loss here other than understanding the fundamentals. Can someone tell me how the compression impact performance?

by maurelius2

3/25/2026 at 8:16:47 AM

If in short, for many inference tasks the bottleneck is memory bandwidth. Suppose you have a machine with a memory bandwidth of 256 GB/s, and let's say you want to do inference for 4B model (model with 4 billion parameters). If you will load the model in BF16 format (16 bits), each forward pass (i.e. each token generated) will require roughly ~8 GB of memory bandwidth. So, 256/8 = 32 t/s, and that's the generation speed you will be strictly capped at even if your processing power is measured in exaFLOPS. But let's say now that you have decided to instead quantize the model and then run the quantized version. Suppose you have made a Q4_K_M version (4 bits + some weights will take more). Now each of your forward passes will take roughly 2-3 GB (rough approximations, reality is different) of memory bandwith (actually, it will be around 2 GB), and even in the worst case 256/3 = 85.3, while 256/2 = 128 t/s. Quants can reduce quality of the model and lower it's performance, but in most modern quantization methods those losses are usually negligible (although, of course, they're still present). So, as you can see, it can be concluded that quantization "widens" (it's not removing it fully) memory bottleneck while still preserving (not always though) acceptable quality.

(Sorry for my terrible English, it's not my native language)

by dryarzeg

3/25/2026 at 8:27:17 AM

So let’s start with a really simple decoder transformer with a single layer and single attention head, and train it to predict the next token in a sequence of text. To predict the next token you need a few things: a query for the very last token in the sequence, and a key and value for every prior token. You take your query and compute a dot product with every prior key (two large vectors in, scaler attention score out). That scaler attention score first goes through softmax, and then becomes the weight you use to compute a weighted average of your values, new value goes through the mlp, mlp output is projected into the logits from which you sample your next token (that’s the general idea at least skipped a few steps).

The last query in the sequence will be new for every new token you predict, but the set of prior keys and values stay the same, ie keys and values are reusable. The key value cache gets bigger and bigger for each new token you add to the sequence, and that’s where compression comes in. You have to store the keys and values in vram, and you’d like to keep the size down by not storing the raw uncompressed tensors. To make this work well your compression needs two things: it needs to be fast so that you can compress and decompress on the fly, and it needs to play well with softmax attention. Prior attempts at compression usually suck at one or the other, either the speed to decompress is too slow and your token/s takes a hit, or you lose important precision and the model output quality suffers. The claim in the paper is that they’ve made progress on both.

by valine

3/25/2026 at 8:35:38 AM

So limiting max context length also reduces VRAM needs a bit? If cache is 20% of total, 1/10th of context as a limit would mean 18% total memory reduction.

by edg5000

3/25/2026 at 8:44:38 AM

Yup exactly, in principle it helps with both inference speed by reducing memory bandwidth usage and also reduces the memory footprint of your kvcache.

by valine

3/25/2026 at 8:50:56 AM

Sounds like Multi-Head Latent Attention (MLA) from DeepSeek

by lucrbvi

3/25/2026 at 9:52:01 AM

Nah, those are completely different beasts. DeepSeek's MLA solves the KV cache issue via low-rank projection - they literally squeeze the matrix through a latent vector at train time. TurboQuant is just Post-Training Quantization where they mathematically compress existing weights and activations using polar coordinates

by veunes

3/25/2026 at 9:21:20 AM

[dead]

by veunes

3/25/2026 at 9:37:46 AM

[dead]

by aledevv

3/25/2026 at 10:05:50 AM

[dead]

by dev_tools_lab

3/25/2026 at 8:32:41 AM

[dead]

by rsmtjohn

3/25/2026 at 8:00:46 AM

[dead]

by mohsen1

3/25/2026 at 7:08:21 AM

[dead]

by hikaru_ai

3/25/2026 at 9:05:50 AM

Pied Piper vibes. As far as I can tell, this algorithm is hardly compatible with modern GPU architectures. My guess is that’s why the paper reports accuracy-vs-space, but conveniently avoids reporting inference wall-clock time. The baseline numbers also look seriously underreported. “several orders of magnitude” speedups for vector search? Really? anyone has actually reproduced these results?

by mskkm

3/25/2026 at 9:49:31 AM

Classic academic move. If the authors show accuracy-vs-space charts but hide end-to-end latency, it usually means their code is slower in practice than vanilla fp16 without any compression. Polar coordinates are absolute poison for parallel GPU compute

by veunes