Nice to see a provider being transparent about KV cache quantisation. I've been suspecting that some providers do this silently whilst heavily promoting their unquantised weights, even though KV quantisation can degrade quality more than weight quantisation.
However, I wish their testing were more detailed. Firstly, some model families are more sensitive to KV quantisation than others (only Kimi K2.6 was tested). Secondly, the evaluation suite they use to claim that FP8 KV quantisation is indistinguishable is noticeably lacking coding benchmarks; in long-running tasks, minor tool call errors compound over time.
> None of this would matter if it changed the model's answers
If they want to assert that the answers don’t change, then perhaps they should calculate the statistical distance between the token probability outputs or something to that effect. I doubt the results would indicate that the answers don’t change by any reasonable interpretation.
vLLM's study also concluded that "FP8 can deliver meaningful latency and capacity gains with small or negligible accuracy loss". Their benchmarks include LiveCodeBench 6.
What is the typical job title and/or skillset for this?
It can be any one of many jobs depending on how high close to the metal one's focus is, but the highest headcount role is usually SRE/Infra/Ops with GPU knowledge sprinkled on top. That is to say Linux sysadmin, networking, fleet management, scaling, incident troubleshooting, etc.
So they quantize models, only tell about it in the blog post (instead of a warning on the model page), and even in the blog post pretend there's no difference by benchmarking on small context tasks many of which are saturated. Coding agents will probably be severely negatively affected by KV quantization.
I'd say serving quantized models without saying so on the "store" page is fraud.
I was interested in reading this until my slop detector went off at the paragraph starting with “It's worth being precise about where the benefit comes from, because it isn't raw speed.”
I have had to stop commenting this because it would end up on 50% of the posts here. I really wish we could flag prose as ai-generated on here and just filter it out.
Meta recently added a filter as a requirement for posts on Facebook. if it was ai generated, you are required to check off a box for that on your posts.
This is only to help them filter out AI generations for their own training data. There is no way you can "block" all "ai generated content" from your view.
Next up, LinkedIn starts using this feedback to train a classifier. They then announce an officially approved "not slop" classification only for LinkedIn Gold member posts. The classified posts have a wider reach due to everyone filtering out AI slop. Non-members automatically get bucketed in with the slop bc they don't pay to have the verified classifier run on them.
seems like at that point it'd be a bigger benefit for LinkedIn to provide the one and only model trained to subvert the human capability to detect slop using that new negative corpus.
> “It's worth being precise about where the benefit comes from, because it isn't raw speed.“
What's funny it's that is as if AI "learned" to speak english but not really. People simply don't speak using those strange constructs: those sentences sound a bit like if a "Karen" was trying to make a point.
What's scary, to me, as a dev using AI, is that those LLMs do the same thing with code: it looks like proper code, but it really ain't so once you dig a bit.
It's verbose and doesn't add anything: it's just infinite verbiage / sloppy-pasta.
Crazy thing though it's that it's 2026 and apparently devs can't be bothered to copy/paste their sloppy-pasta LLMish into a de-sloppifier before publishing blog posts.
I'm getting somewhat conflicted. There is a Twitter account I follow that has great takes and lots of deeply personal posts... BUT he obviously uses an LLM in his writing pipeline somewhere. Too many AIisms scattered about to ignore.
I'm not thrilled with it, but he is obviously using it to improve his writing overall- to communicate some great ideas that are personal and germane. I've decided that being too inflexible serves no one. If it is true slop, I'll not revisit the writer in the future- if they are using AI to polish writing that at its core is a unique voice, I'll accept it and learn to live with it...
Cloudflare blogs are not meant to be human-read, AFAIK. They're raw material meant to be fed into an agent to be filtered down. I rarely read the contents because they are usually word-expanded to a greater degree than an article from The Atlantic.
I don't disagree, but at some point in the last year they ended up severely word-expanded. So in a revealed sense, they are no longer meant for human consumption except for those who don't significantly value their own time. There is very little information in the post that an agent can't pull for you:
* they use quantized models
* they quantize KV cache
* they have a cache tagging mechanism to prevent cache misuse (neat)
The agent can extract numbers without filler prose as well.
We let all traffic get MITM'd, now we're letting our AI conversation get tracked. Cloudflare reeks like a US Honeypot.
However, I wish their testing were more detailed. Firstly, some model families are more sensitive to KV quantisation than others (only Kimi K2.6 was tested). Secondly, the evaluation suite they use to claim that FP8 KV quantisation is indistinguishable is noticeably lacking coding benchmarks; in long-running tasks, minor tool call errors compound over time.
> None of this would matter if it changed the model's answers
If they want to assert that the answers don’t change, then perhaps they should calculate the statistical distance between the token probability outputs or something to that effect. I doubt the results would indicate that the answers don’t change by any reasonable interpretation.
Maybe the results are still good enough.
https://vllm-project.github.io/2026/04/22/fp8-kvcache.html
Why… I wanted to see if it’s worth it to use cloudflare’s endpoint but I can’t even see the pricing
Input tokens (per 1M)$3.00 Cached input tokens (per 1M)$0.30 Output tokens (per 1M)$15.00
What is the typical job title and/or skillset for this?
Applied ML Research Engineer or something maybe, not that I have ever seen that title. Maybe just catchall ML Engineer...
It can be any one of many jobs depending on how high close to the metal one's focus is, but the highest headcount role is usually SRE/Infra/Ops with GPU knowledge sprinkled on top. That is to say Linux sysadmin, networking, fleet management, scaling, incident troubleshooting, etc.
I'd say serving quantized models without saying so on the "store" page is fraud.
I love AI, but I really hate reading it.
HN is for conversation between humans[1] (about AI generated blogspam, apparently)
1. https://news.ycombinator.com/newsguidelines.html
I've been asking for that for some time.
I can only read so many (either pro OR against) ".. IT'S AI! ! !" comments before skipping the thread. I can't be the only one.
How well it would work on this site, I'm not sure.
then we're all screwed I guess. lol
What's funny it's that is as if AI "learned" to speak english but not really. People simply don't speak using those strange constructs: those sentences sound a bit like if a "Karen" was trying to make a point.
What's scary, to me, as a dev using AI, is that those LLMs do the same thing with code: it looks like proper code, but it really ain't so once you dig a bit.
It's verbose and doesn't add anything: it's just infinite verbiage / sloppy-pasta.
Crazy thing though it's that it's 2026 and apparently devs can't be bothered to copy/paste their sloppy-pasta LLMish into a de-sloppifier before publishing blog posts.
I'm not thrilled with it, but he is obviously using it to improve his writing overall- to communicate some great ideas that are personal and germane. I've decided that being too inflexible serves no one. If it is true slop, I'll not revisit the writer in the future- if they are using AI to polish writing that at its core is a unique voice, I'll accept it and learn to live with it...
* they use quantized models
* they quantize KV cache
* they have a cache tagging mechanism to prevent cache misuse (neat)
The agent can extract numbers without filler prose as well.