GLM-5.2 – How to Run Locally

by TechTechTech • 1 month ago

617 305 1 month ago

xrd2• 1 month ago

So close! My machine with 192GB RAM + RTX 3090 24GB can almost run this. It says it needs 24GB of VRAM and 256GB of RAM for MoE offloading.

https://unsloth.ai/docs/models/glm-5.2#usage-guide

In a prior thread, someone said it would take $500k in hardware:

https://news.ycombinator.com/item?id=48629970

mgambati• 1 month ago

With 2 wouldn’t have good results. Ideal range for coding is at least Q8.

kibibu• 1 month ago

According to this very article, 4-bit dynamic is essentially lossless

cheema33• 1 month ago

I have the RAM, but not the VRAM. What kind of speed/tps could you expect from a 3090 with 24GBs of RAM? I am somewhat tempted to pick a GPU with 24GBs of RAM.

phamilton2• 1 month ago

Generation is basically just memory bandwidth math.

Each token has to read all the active weights. I think that's around 40B parameters active. At a 4-bit quant that's 20GB. With 100GB/s (replace with whatever your bandwidth is) and you get 5 tokens per second.

ekidd• 1 month ago

A GPU with 24GBs of RAM is mostly useful for running a very carefully squeezed Qwen3.6 27B (4-bit Unsloth quants, 8-bit K/V cache, possibly MTP, 128k context). This is a fun little model that's smart enough to do debugging, refactoring, and implementing "clean" specs that don't force it to make complicated design choices. I've seen it rip through a 9-year-old Terraform AWS config, and (without using the network) correctly identify nearly everything that would need to be upgraded or migrated for modern AWS. But if I give it some poorly conceived spec with lurking design headaches, then it goes on an endless thinking binge and ultimately fails.

Speed-wise, I don't have numbers, but it feels subjectively faster than Opus in Claude Code. YMMV.

Once you go above "a used 3090 at a decentish price", then I strongly recommend renting cloud GPUs or at least testing models using paid APIs. This allows testing your use case before spending piles of money.

elliotbnvl• 1 month ago

$500k is a vast overestimation. For massive concurrency at FP8 or even BF16 maybe.

NVFP4 at reasonable speeds (~120 tok/s) and concurrency is possible at a $80/90k figure with today's prices, maybe even less. That buys you 6 RTX 6000 PRO Blackwells, a decent CPU and motherboard, power supply. 576gb of VRAM.

You could do it for under $50k if you're OK with 40 tok/s decode, ~1200 tok/s prefill.

__m• 1 month ago

How fast will the hardware become outdated? Are there big improvements expected in the next 3 years?

hbbio• 1 month ago

Yes, a single GB300 workstation also does it, probably even more than 120tok/s.

Official price 85k...

throwawayffffas2• 1 month ago

You can get a 1TB of HBM2 vram for like 10k, https://www.ebay.com/itm/177571378959

The problem is the backplane I have not managed to find a single baseboard, and getting a random baseboard to work with random modules is probably a crap shoot.

uberex4• 1 month ago

Funny I casually asked Gemini and it said 500k for unquantized with decent throughput.

stymaar2• 1 month ago

This is why you shouldn't believe uncritically an answer from an LLM (neither should you do for any answer from a human either though).

j45• 1 month ago

LLMs aren't discrete calcluators or estimators of things unless framed and guided to do so.

colinsane3• 1 month ago

i asked gemini and it replied with "Error: 400 Your prompt was blocked by safety filters. Please revise and try again."

nijave5• 1 month ago

That's fair for new hardware. You probably want to prompt "homelab" or "used hardware" to compare what's in this thread.

ijidak• 1 month ago

Crossing my fingers that this boom jumpstarts 90's like improvements in computing hardware.

I feel like part of the reason for the relative stagnation in hardware over the last twenty years was simply the lack of use cases to justify hardware refreshes by businesses.

Most of the money and energy went to mobile for the last fifteen years.

Affordable local inference might be the gravy train the server, desktop, and laptop manufacturers need to get back in gear.

linzhangrun• 1 month ago

Physical limitation of the manufacturing process may be more significant factor, starting from the TSMC 10nm ten years ago

gruez• 1 month ago

I feel like part of the reason for the relative stagnation in hardware over the last twenty years was simply the lack of use cases to justify hardware refreshes by businesses.

No, we're running into limits of moore's law, and it's showing in prices for new nodes, where they're getting denser but not cheaper.

0xbadcafebee• 1 month ago

Definitely the stagnation was due to a lack of use cases, but this isn't a bad thing. We don't need most of the hardware advancement we got.

Business hardware got beefier because businesses demanded more data (or more specifically: the industry told businesses they needed more data), with no idea of what to actually do with it once they got it. To get all that data, bandwidth needed to be increased, with more iops to read/write it, more storage to keep it, and more memory and cpu to process it. But 99% of the data is junk. Companies have "data lakes" so big they need to come up with excuses to use the data, or risk somebody pointing out that they're spending a fortune hoarding bits.

Consumer hardware hasn't had a new use case since like 2012. Faster wifi for broadband & local file transfers, and higher-resolution video, are the only reasons one needed newer hardware. We actually got a resolution so high it makes no perceivable difference. And yeah we got faster CPUs and memory, but as soon as we did it got all eaten up by the most inefficient, wasteful software conceivable. Same use cases as 13 years ago, just more expensive, harder to use, and buggier. We should've gotten a new CPU architecture that was faster and more energy efficient. Finally it was delivered, but with a moat around the golden Apple.

Here we are two and a half decades into the Internet era, and my damn bluetooth earbuds and webcam microphone don't work half the time that I open a video conferencing app. Hardware can stay exactly like it is for the next few decades and I'd be happy. I just want software that works, and doesn't get continuously slower, forcing me to buy bigger hardware; or more draconian, locking me out of being able to use it how I want.

BobbyTables2• 1 month ago

Yeah, even Windows managed to not drive terribly dramatic upgrades in general computing (besides Windows’ absurd RAM usage and now requiring a TPM).

In the old days, Microsoft Entertainment Pack games were somewhat visibly taxing on some lower end systems.

bbor• 1 month ago

I’m kinda lost here… do y’all really have machines in your houses with hundreds of gigs of RAM?? Am I just behind the times?

The page advertises the 8-bit quant as taking ~800GB, which seems like it would require at least 3 consumer motherboards fully stacked w/ 4x64GB cards each.

Maybe “locally” has slowly come to imply “…on your homelab”?

cpburns20092• 1 month ago

RAM wasn't expensive even a year ago. I maxed out a used Dell Precision T5610 with 128 GB DDR3 for $250 in 2021.

numpad0• 1 month ago

DRAM prices at mid-2025 rates were ~$2.5/GB for DDR5, and ~$1.5/GB for DDR4. "Hundreds of gigs" of RAM used to be under $500. 128GB of cheapest RAM used to be like $200. It seemed to go over heads for a lot of people that you could get hypothetical future machines on CS/CE textbooks were attainable for that little, for some reason - there seemed to be some fixation on the idea that 16GB is all you need.

Gracana• 1 month ago

You don't have to have a server, workstation motherboards support lots of memory channels.

I was lucky to buy a lot of RAM before prices skyrocketed. I knew I wanted to play with this stuff, so I spent what felt like a lot of money at the time to buy 8x96GB DDR5-6400 RDIMMs. Now the same RAM costs at least 6x more.

nijave5• 1 month ago

I got a 2U rackmount with 192Gi DDR4 for $1.1k USD in 2023. Around 1.5 yrs ago, server RAM could be had pretty cheap--especially slower LRDIMMs (I wanna say 512Gi DDR4 was <$500 USD). I checked a couple old ServeTheHome threads and seeing maybe around $50/32GB RDIMM although thought it was cheaper than that for a little while

oceanplexian3• 1 month ago

As soon as Llama came out I had a realization what was coming and went all-in on hardware with the assumption open source would catch up with GPT4. Surprise, it did, we now have small models that absolutely crush GPT4 in performance.

It wasn’t that absurdly expensive for a hobby, I bought 64GB DDR4 ECC sticks between $70-$100 on eBay before everything took off. Now everyone is in here debating if open source is 1 month or 3 months behind SOTA. The future is obviously local.

nozzlegear2• 29 days ago

I've got an M1 Mac Studio Ultra with 64gb RAM, which I bought when it first released, i.e. not for LLMs at all. I'd love to have one of the newer generations with twice or quadruple the amount of RAM, but while it has no chance of running GLM 5.2, I've been able to run local, quantized LLMs like Qwen 3.6 using oMLX + Opencode pretty comfortably and productively.

zuzululu3• 1 month ago

wonder if AMD's new ai chip can run this with ease? I'm seriously consider buying it. GLM 5.2 is just shy of GPT 5.4 so I would welcome offloading any grunt work locally

I am very excited for local LLMs I think we may have GPT 5.5-xhigh level of performance for under 2000 EUR

This should put more pressure on the frontier models to avoid sitting on any fancy stuff and lower token prices as a whole.

Nothing beats a local LLM disconnected from the cloud.

Iolaum• 1 month ago

At full quantization GLM 5.2 may be close to GPT 5.4. But at Q2 or whatever one needs in order to run it on a pro-sumer device it will be worse.

Also I m not sure where you are getting the under 2k value. I bought a Framework desktop 128GB last year and my setup was around 2.7k. The same setup now sells for around 4.7k.

kccqzy• 1 month ago

The AMD 395 supports up to 128GB unified RAM. So still not enough even at 1-bit quant unfortunately.

monksy2• 1 month ago

96gb vram is the max it supports.

benjiro292• 1 month ago

"GLM 5.2 is just shy of GPT 5.4"... If your running the full model. As in have 750 (FP8) to 1.5TB(FP16) of memory available.

Do not mix the benchmark results of GLM 5.2 FP16/FP8 with FP4 or FP2.

* FP4 will mean a accuracy loss of about 3%. Not noticeable but more chance for mistakes.

* FP2 ... what is what most people are able to run at home, for a "reasonable" price. Your looking at over 17% loss in accuracy.

At that point, your running at less then claude-sonnet-4.6, as the issues compound with accuracy losses. And reasonable priced is still in the ~ $5000 range (192GB + GPU 32GB active/kv cache system).

For that price your using a Codex / Claude Pro subscription for the next 4+ years with better models (by default), let alone with a FP2 GLM 5.2 version. And your looking at < 10 fps. A MacStudio with 512GB will net you 18 a 20fps+ with FP4, but ... i mean, those used to be $10.000.

Unfortunately the local hardware cost is a major issue for running large models like that.

Edit: Its funny whenever the issue of cost and what you need to give up vs the subscription services, there are always people who downvote in bad faith.

zuzululu3• 1 month ago

you are right that means GLM is still quite far off from truly competitive

i think your answer was perfect not sure why you are being downvoted

kgeist• 1 month ago

The cost of local hardware is amortized if a whole team uses it instead of just 1 dev (GPUs are extremely underutilized if you launch just 1 generation stream). I'm not sure why everyone always assumes solo devs with Macs. We've just ordered a large datacenter-grade node for use by the whole dev team, and the calculations show that it's going to cost the same amount of money if we kept using AWS Bedrock (infosec reasons) for a couple years but... it gives us 100% privacy, we're immune to all the AI regulation dramas in the US/EU, all the random outages, and the developers won't have to think about token limits/weekly caps etc. ever again. And all that with a model which is Opus-grade

(it's not our first AI server, we already have experience deploying LLMs for our clients, so the numbers look solid)

nh43215rgb• 1 month ago

Even with upcoming AI Max+ PRO 495 we are capped with 192GB, so no...

nl2• 1 month ago

I am very excited for local LLMs I think we may have GPT 5.5-xhigh level of performance for under 2000 EUR

We are maybe 10 years off that.

RAM prices are going to continue to increase for the next 2 years at least.

Even putting that aside it's currently around 40-70,000 EUR to run this with a FP8 quantization (which you need to get close to maximum performance).

To actually get GPT 5.5-xhigh performance in the real world you need more headroom to support things like subagents (which will fill up your KV cache).

I like local models but realism is important. The sweet spot for the next 3 years will continue to be ~35B MoE models. They might match GPT 5.5-xhigh for chat-style problems but not for coding.

hsuduebc2• 1 month ago

I wonder, if in the near future any acquisitions of some RAM producers with intent to just keep RAM prices up, will happen from the AI companies. It could seriously hurt their business, if companies will be able to host their AI in some time.

UncleOxidant5• 1 month ago

Are you talking about Medusa Halo? It's going to support up to 256GB unified memory (up from 128GB for Strix Halo and 192GB for Gorgon Halo). That might just be barely enough to run a 2-bit quant GLM-5.2. It will expand memory bus to 384-bits, vs. 256-bits for Strix Halo which will help with bandwidth (projected to be around 500 GB/sec). But don't expect Madusa Halo-based machines to appear until sometime in 2028.

The other way this could go is that Z.ai could decide to release a smaller model targeted towards coding. They've done that before (GLM-4.7-Flash had 30B params). It would be great if they decided to release something in the 80B-100B param range. Something that size would easily run in a current Strix Halo system.

zuzululu3• 1 month ago

yeah you are correct 2 bit quant won't be enough

guess we'll be paying $200/month for a while

monksy2• 1 month ago

Strix Halo only supports 96gb of video memory then it goes to 32gb to the host system.

pheggs5• 1 month ago

I feel like the gap is closing to be able to run good enough models locally even for coding and I would assume it could make some companies a bit nervous. Am I wrong about that?

fny• 1 month ago

The RAM requirements are still pretty painful.

yieldcrv2• 1 month ago

equilibrium in one or two more years on the consumer/prosumer side

think Apple M6 or M7 with a currently unforeseen denser memory style, 256gb RAM

a couple inference or cache improvements on the algorithmic side, using less ram for context windows and doubling token speed again

denser open source models, packing more experts for smaller active layers

it'll still be expensive but like $8,000 - $13,000 instead of $450,000 worth of B200s

CamouflagedKiwi• 1 month ago

The hardware requirements to run this locally are still very high. Seems far enough off mainstream for those companies not to be too worried yet.

cogman104• 1 month ago

I don't think so. I could easily see a company deciding to host and run these models for their own development. If you have a dev team of about 10 people, a one time $50k investment in an LLM server has to be pretty tempting. Unlimited tokens, decent performance, upgrade options, and potential product integrations.

For companies wanting LLMs in their products in general, I have to think going the local llm route is even more tempting. Somewhat dumb models are more than good enough for a lot of the things people are integrating LLMs into their products.

eventualcomp• 1 month ago

Where is $50k coming from again?

twelvechairs2• 1 month ago

Surely for most the desire is just an LLM provider that doesnt store or sell their queries (including by national actors). As long as that is allowed to happen surely its the answer for the vast majority.

UncleOxidant5• 1 month ago

If we didn't have a RAM/GPU shortage right now they would be more nervous than they are. But as it is very few people are going to be able to afford a rig that can run this model effectively. That's probably not going to change for several more years yet. I think if the Z.ai folks decide to come out with a flash version of GLM-5.2 specialized for coding that came in about about 80B params, then the US frontier labs would probably be more worried. Overall, the Chinese AI companies have been showing the way to do the same amount with less (sometimes much less) and as that trend continues it's going to make the frontier labs worried - but even the Chinese AI companies are going to want to protect their moat by not releasing capable models that are significantly smaller than their current flagship models. AliBaba Qwen seems to be there now - it's gotten mighty quiet from them lately - their latest 395B model is just too large for most folks to run at home and they don't seem to be making any noises about releasing smaller ones this time around.

gpm• 1 month ago

The ram/gpu shortage won't last forever though. Moreover we can be pretty confident that long-term the prices will obey wrights law and come down in cost significantly (from the pre-shortage prices) as we learn to produce them more efficiently.

LLM companies are valued as if they're going to have some enduring monopoly that they can extract money from... GLM-5.2 and similar models make that valuation very very questionable.

elorant• 1 month ago

Very few people, but quite a lot of companies especially after per token pricing took effect and companies see their invoices skyrocketing. You pay an upfront cost once and you’re done.

verdverm2• 1 month ago

I suspect the time horizon is shorter because of software advances. We are getting more capability out of smaller models

Alibaba released Qwen 3.6 "tiny" models not that long ago, they punch way above their weight(s)

dannyw5• 1 month ago

When a large open weight model is released, a lab, startup, or a rich hoist can easily do logit-level distillation and create a XXb param model or whatever, and in theory you should get a really good distill.

bawana2• 1 month ago

is it possible that ai companies ordered a bunch of ram just so that models cannot be run locally? they are betting new fabs wont be built before quantum takes hold.

notatoad• 1 month ago

locally on what hardware? something like the new dgx spark, ryzen halo, or mac studio will cost you ~ $4k plus whatever you pay for power. at the rate AI is currently progressing, i think you'd be optimistic to consider that as having a 2 year depreciation.

for $4k, you can get 20 months of claude max 200. i'd take claude over the hardware.

anthropic will have something to worry about when you can run a local model on your macbook that can code. but i think we're quite a ways off from that.

chatmasta2• 1 month ago

Just a hunch, but I think the most cost effective “local” deployment method right now is renting GPU clusters by the hour and running all the inference software on them yourself. This will be cheaper than capital expenditure on hardware that will depreciate and become last-gen, and cheaper than OpenRouter pay per token.

tomr75• 1 month ago

people who can't afford Claude max 200 are using qwen 3.6 27b for local coding assistance already

fc417fc802• 1 month ago

at the rate AI is currently progressing, i think you'd be optimistic to consider that as having a 2 year depreciation.

How so? Model capability at a fixed hardware level has been consistently (and rapidly) increasing. You might or might not be able to run state of the art 2 (or 4 or whatever) years from now but you can reasonably expect the hardware to last upwards of a decade with model performance consistently improving over that time frame.

You can get a tolerable (at least by some metrics) experience using 10 year old hardware today.

c7b2• 1 month ago

You can get a 128GB Strix Halo for under $3k. Used to be under $2k. Even if you believe it'll be completely obsolete for AI in two years, it'll still be good for many other things. Games for at least several more years, a great home server and/or desktop almost indefinitely. Plus, we might actually reach good enough levels for some AI use cases, if we're not already there.

And never underestimate the potential for enshittification. Your local rig will only deliver better performance over time as more and more tweaks come out. With cloud services expect the opposite to happen as subsidies run out. It's entirely possible that they will intersect on a bang per buck basis within two years.

SXX2• 1 month ago

You forget that after 2 years you still gonna have said Mac Studio that can be sold off for 30-50% of the price.

Of course its gonna lose value faster if something magical happen with hardware manufacturing, but you'll likely get 25% back at least.

On other side you cant really predict how valuable claude max gonna be in a year because Anthropic can further enshittify it.

oceanplexian3• 1 month ago

Yeah, 20 months of Claude Max until they rugpull you. I’m spending 7-10k/month in raw token costs on Claude Max. Having an alternative is a nice insurance policy.

stymaar2• 1 month ago

Honestly, Qwen3.6 is already what you need for the large majority of tasks.

(I only ask Opus every 5 to 10 requests, when my local Qwen fails or when I encounter a situation that is too world-knowledge specific to be worth asking, but that way you can live easily with Claude's cheapest plan without ever facing usage limit).

simplyluke• 1 month ago

You don't even need to run them locally for them to be a threat. Plenty of companies are looking at paying third party companies to host these models and they come in at fractions of the price of the frontier labs.

pheggs5• 30 days ago

thats true. also, I watched the glm prices and it didnt take long before the prices dropped even lower for some providers. its like another layer of competition between hosters

scosman• 1 month ago

It's not economic to run them locally. It's amazing for privacy, and fun hobby. But you're either looking at super slow CPU builds with $10k in RAM, $90k worth of GPUs, or a really quantized model that doesn't compare in quality.

I might build one for fun, but it's not going to change the economics alone. Still exciting it's possible.

oceanplexian3• 1 month ago

It depends what you’re using it for. Real time interactive Claude code session? No, it’s kind of impractical.

But if you already have agent loops dialed in (For example I have one that uses a browser testing framework), it wouldn’t really affect me at all if it crunched away at 7 tokens per second all night long.

leansensei2• 1 month ago

Not really, you can do great things without them. I've been summarizing hundreds of documents. I've added MCP servers to my internal business tools (Elixir apps) and can chat with the Nous Hermes agent over Telegram about pending orders, inventory level, historical product prices, etc., without having to click/dick around with a web UI.

Sure, it cannot replace SOTA models for agentic coding, except for small, well-scoped refactorings. But even a model like ministral-3:8b or qwen3.5:9b is a boon for so many smaller use cases!

fsuts3• 1 month ago

Why do you think they are rushing to IPO!!

andai• 1 month ago

How is this model half the size of DeepSeek V4 Pro? Is it because DeepSeek did more aggressive cost cutting on the attention mechanism?

skiing_crawling• 1 month ago

"it can fit" on 256GB of RAM, but it will be heavily quantized and still run very slowly. The headline number is not token generation, its prompt processing. So if you get 10 tok/s and an API gives you 20-30 tok/s, it doesn't seem that bad on its face, but a mac studio or any other machine that's not loading all of it into GPU will do PP 20-50X slower than a purely GPU based setup, which is what actually makes this unusable without $50k in GPUs.

On top of that, you will still be heavily quantized.

gerdesj2• 1 month ago

A nvidia spark thingie has 128GB unified RAM. They also have a dual port version of one of these things: https://www.nvidia.com/content/dam/en-zz/Solutions/networkin.... ie 2 x 100GB/s ports, they may even be 2 x 200GB/s. Once I've got my paws on one, I'll know more.

You can cluster these beasts too. Two and three (with two IP subnets) is fairly obvious. Four or more might need a switch depending on how much network latency affects things.

Apple seem to have forgotten about M series with gobs of RAM. I can't get the Apple shop to show more than 96GB of unified RAM and that costs a kidney.

Computer0• 1 month ago

128 gb of much slower ram than Apple.

mapontosevenths4• 1 month ago

I have one, and I love it. That said my buddies Mac smokes it for inference workloads in terms of tokens per second AND its more usable for other things.

If you are training and doing research it's great, if you want to cluster them it cant be beat, but if you just want local inference on a single box buy a mac or even a strix halo device.

jauntywundrkind• 1 month ago

200 Gb / s (not GB/s)!

(Still potentially very useful! But not magically ultra fast.)

justincormack• 1 month ago

It is 2x200Gb/s physically but the PCIe bandwidth is basically only 200Gb/s so it may as well be one, and actually its a weird 2xPCIe4 not 1xPCIe8 so it appears in software as dual 100Gb/s. Its a bit odd.

nullc2• 1 month ago

Just running cpu only w/ Q6 on 9684X I get about 1tok/s ... also still get about 1tok/s/stream when running 16 in parallel.

nullc2• 30 days ago

6.5t/s if I add one RTX R6000 gpu (where is uses about 25GB).

Wowfunhappy• 1 month ago

The full model requires 1.51TB of disk space

...a bit of an odd question: how well do LLMs losslessly compress, as in for cold storage?

I definitely don't have the hardware to run this model at any kind of reasonable speed (and I don't want to use a super aggressive quantization that would kill performance). Even so, I think it would be cool to retain an offline copy, in case... I don't really know, a solar flare destroys the internet some day, or maybe a zombie apocalypse. It would just be cool to have.

But 1.5 TB is a bit too much! If it could be compressed down into something semi kind of reasonable, that would be fun!

SirMadam• 1 month ago

SOTA LLM specific compression achieves around ~54%! https://arxiv.org/abs/2505.06252v3

redox992• 1 month ago

Probably not at all, considering weights are randomly initialized.

gcr• 1 month ago

There are two forms of compression relevant to LLMs:

1. Reduce the number of parameters

2. Reduce the resolution of each parameter (quantization)

For 1, changing the architecture is typically only possible by the labs producing the models, which is why each OSS model release tends to feature a small number of carefully chosen model sizes (for example, Gemma4 comes in e2B, e4B, 12B, 26Ba4B, and 31B sizes).

Generally, models with higher parameter counts have more world knowledge. For coding models, this shows up as a stronger command of uncommon libraries/languages. Very small models (<20B) also lack “smarts.”

Reducing the resolution of each parameter is easier which is why lots of practitioners have their own quantizations, but this makes it harder for a model to “think” fluently. Interacting with heavily quantized models feels like interacting with someone who didn’t get any sleep the night before.

Models that have higher-fidelity quantization take more RAM and have higher “smarts,” but don’t necessarily have more world knowledge. Models with aggressive quantization tend to be more likely to make rookie mistakes, emit malformed tool calls, get stuck in loops, or even exhibit signs of “neuroticism” / “distress” in their thinking tokens.

Parameter counts = world knowledge, quantization = “smarts.”

This is a soft rule of thumb, the difference isn’t very strong.

throwdbaaway• 1 month ago

On ZFS with zstd compression, I am getting 1.34x compressratio for the BF16 weights (across multiple models).

Here's the du output for GLM-5.2:

    $ du -s -BG /cube/models/zai-org/GLM-5.2/
    1099G   /cube/models/zai-org/GLM-5.2/

walrus0110• 1 month ago

...a bit of an odd question: how well do LLMs losslessly compress, as in for cold storage?

TBH this is like the near last ranking consideration in cost for being able to download and run this. Even though HDD and SSD prices have gone nuts as a result of the recent demand/shortage, it's not like 1.5TB of space costs a lot.

Even if you fed it into xzip with the most cpu intensive compression options and it didn't compress at all (eg: like trying to xzip an AV1 video, or whatever), it's still the cost of a single fast food hamburger meal in $/TB. The real concern is the RAM to run it.

But anyways, anecdotally, many 16-bit full precision GGUF files will compress to about 65% of original size with default xz options. I have a log here showing that's what IBM Granite 4.1 30b compressed to, which I'm keeping around but in lukewarm storage.

CGamesPlay• 1 month ago

Can somebody help me understand the Quantization Analysis? It says "dynamic 4-bit UD-Q4_K_XL and dynamic 5-bit UD-Q5_K_XL are generally lossless" while showing a top-1% token agreement on the chart of 97.5%. Not what I would consider "generally lossless". Is this implying that some post-processing is going to account for the 2.5% loss? Beam search?

dannyw5• 1 month ago

Generally 97.5% token agreement is very positive. Like the article explains, the difference isn’t the model thinking the capital of France isn’t Paris, but rather maybe saying “The capital of France is Paris” instead of “Paris is the capital of France”.

hxii• 1 month ago

Any time I see one of these posts about models of this size a quote comes to mind – "Your Scientists Were So Preoccupied With Whether Or Not They Could, They Didn’t Stop To Think If They Should".

Only a select few have the hardware required to run this to begin with, and even then the forecasted performance makes me wonder if it’s worth it at all.

segmondy11• 1 month ago

Completely worth it. At 6tk a second. If I can get 2 hrs of token generation. That's 2hrs * 3600secs * 6tk = 43200 tokens, at about 10tk to a line of code, that's about 4320 lines. Let's even trim it more and slice it by half. That's 2160 lines of code a day. Most professional programmers can't deliver that much consistently in a day.

The key to a model this large is (1) Use it to plan, generate lots of plan and farm out to a smaller model. Then for very specific and complicated portions precisely prompt for what you need.

uberex4• 1 month ago

Thats not a complete reasoning. Even frontiers need to revisit and fix things. Add 10 loops to that and it is 20 hours. Still great compared to a 2023 human, but why am I not just paying pocket money for Claude Pro instead?

ramgine• 1 month ago

I have up to 1tb of ddr4 in my server but it only has a 12gb vram 3060. Would getting a 24gb vram make this a viable system or am I throwing money away?

segmondy11• 1 month ago

You can run it today with that 12gb vram 3060, but I would suggest getting 2 3090s. Use cmoe option. This will keep the attention/route tensors on the GPU and offload the rest to system memory. Try it now and see the performance.

rnewme• 1 month ago

Should work yes.

dofm• 1 month ago

Can't run this myself.

But I do like Unsloth Studio, quite a lot. It's nicely designed.

snootypoot• 1 month ago

if sam altman didnt exist i could afford to run this

numlock862• 1 month ago

if sam altman didn't exists this model would most likely not exist as well

jonathanhefner• 1 month ago

Runing GLM-5.2 on local hardware

Do the runes make it smarter or just run faster (or both)?

nicman232• 1 month ago

depends on the color

segmondy11• 1 month ago

I run Q4_K_XL. All it takes to run to get about 6tk/sec is 512gb of ram and 2 3090 GPUs with llama.cpp -cmoe. I also have crappy DDR4, 2400mhz, 3200mhz will bring that speed up to about 9tk/sec. I also have ok 32core epyc CPU, a better 64core would bring it up to about 11tk/sec. I did a budget build before the crazy hardware cost and I regret it everyday. Nevertheless, it's fantastic being able to run this model at home. It's great for planning, one shot prompting once you have a plan or all the context you need. This entire hardware cost $2400 when it was built. If you're willing to be resourceful, you can find ways to run these models at home. I often get the silly question of why, and suggestions about how much I can save using cloud API, but the Fable drama has opened up eyes on why it's good for us to be independent. Thanks team unsloth, Q4_K_XL is solid, if you are going to grab a quant, make sure to get the K_XL variant if it can fit.

redox992• 1 month ago

That's crazy good for $2400.

edg50003• 1 month ago

Very cool. So it's not just about GPU VRAM which I incorrectly thought. I though you'd need 512 GB GPU VRAM. I don't think it cost only 2400; 512GB ram would be more expensive though back in the day. But not mortgage-grade 200.000 which I estimated myself (which assumed running in 100% VRAM; overkill for a single user probably).

segmondy11• 1 month ago

you can use system ram with a system like llama.cpp which offloads to system ram. token generation is a function of system bandwidth, the faster the bandwidth the better. so I'm on 8 channel 2400mhz. if I had a 12 ddr channel, I would get 1.5x the speed at 2400mhz. of course ddr5 is much faster, so a 12 ddr at 4800mhz will provide 3x the speed for token generation or roughly 18tk/sec. prompt processing is all about compute, so the more cpu cores you have the faster it can do PP.

nijave5• 1 month ago

Well, it's about GPU VRAM if you want something competitive with cloud-hosted offerings at the performance levels showing in benchmarks. This is a heavy quant with quality degradation and significantly lower performance.

Cloud offerings are 80-200tk/sec versus single digit tk/sec.

That said, I'm also surprised it runs at all locally. I do think it'd be painfully slow for anything interactive so you're relying on another model for a comprehensive design or you're hoping a one-shot with somewhat degraded quality turns out correctly.

nextaccountic2• 1 month ago

How can you combine CPU cores and multiple GPU? Are you running some layers in cpu, others in gpu #1, and others in gpu #2? What about the bandwidth and latency between them?

Or maybe the model itself only runs at gpus, and the cpu memory only store the weights for experts not corrently activated? If so, then what's the 32 or 64 cpu cores for?

I'm a big fan of fully utilizing one's hardware and it's kinda sad that it's not the norm to run things on either gpu, cpu or both, dynamically choosing at runtime, for everyday software

segmondy11• 1 month ago

checkout llama.cpp, the entire point of the project is for us mere mortals and GPU poor.

nodja• 1 month ago

Pipeline parallelism. Instead of splitting layers by row/column. You split at the layer edges. So instead of having this huge bottleneck of bandwidth you only need to transfer about 4KB per token when changing devices on a model like Qwen 3 30BA3.

xrd2• 1 month ago

This is a good place to start reading about dual gpus.

https://github.com/noonghunna/club-3090/blob/master/docs/DUA...

fsuts3• 1 month ago

6 tokens per second?

Can you put up with that? As seems very slow. I aim for 40t/s on a laptop and choose models that deliver that speed over larger slower ones

segmondy11• 1 month ago

I have been putting up with it forever. We are spoiled by MixtureOfExperts. Folks were delighted to run llama3-70B at such speed. We were happy with 15-20tk/sec with 8b models, and if you could run llama3-405B at 1tk/sec you were a god. To each their own. I can live with 6 high quality tokens. If I could get a Fable equivalent model, I'll gladly take 2tk/sec if that's what it took to run it locally.

walrus0110• 1 month ago

I get a lot done with something that's also approximately 6 tokens/second, if you're willing to give it a well defined set of prompts and projects to work on, leave it for an hour or two, then come back and check what it's done. And often to remember to give it something of more consequence to do for at least 3-4 hours of wall clock runtime before heading to bed.

zozbot2342• 1 month ago

AIUI the llama.cpp implementation for this model is still quite half-baked due to missing the support for DSA sparse attention mechanism. This leads to running the model with a different mechanism that it has not been trained for, which has been shown to lead to lower quality and performance.

Anyway, I think GLM 5.2 in many ways is not as interesting as DeepSeek V4 series, which uses an even more advanced attention mechanism and can save a lot of memory capacity for KV cache, especially at larger contexts. Which in turn opens up wide batching especially on consumer platforms. GLM doesn't have that, in some ways it feels broadly similar to Kimi 2.6 wrt. the underlying performance architecture. Both are a bit too heavy to run reasonably at full quality on ordinary hardware.

trollbridge2• 1 month ago

Particularly DeepSeek 4.1, which they appear to be A/B testing on the API and which also seems available on the free chat interface.

It also has an input image modality, which is a game changer. The cheap Sinofrontier models have generally been lacking in this regard.

Basically, Chinese competition is fierce - DeepSeek set the pricing tier, and the question for each lab now is how to justify charging a little more.

MiMo-2.5-Pro has gone with UltraSoeed, pumping out 1000t/s for a 3X price hike.

GLM has gone with 5.2, hitting Opus levels of reasoning at a fraction of the cost.

DeepSeek will probably keep their pricing model and just keep getting better and better.

Qwen-3.7 is the dark horse. Some rumours are Alibaba is simply making these models because they need them internally.

The real question is why this level of innovation and competition isn’t happening in America or Europe. In particular I see no reason Europe doesn’t have a lab competing on these terms.

dxuh• 1 month ago

"All it takes to run" might be fair if you paid $2400, but right now the total price is way closer to $10k (almost 5k for the RAM and 2k each for the GPUs). Today that is a lot of expensive hardware.

segmondy11• 1 month ago

512gb 2400mhz ddr4 ram = $1600 not $5000. https://www.ebay.com/itm/188284985172 You can get creative and source 2-3 2080ti 22gb from China for about $250 a piece. You can either be resourceful and find a way or find a whole bunch of excuses.

Astropolak• 26 days ago

Agree'd here. I am seeing a pattern in this thread - Everyone here seems to justify paying substantially more money to self-host hardware thinking they are saving in the medium-long term. Frankly from the perspective of the company I'm with - lost opportunity cost is more important than trying to future proof on a hunch that AI vendors are going to pull the board out under you. There's alot of competition in this space and we're seeing these frontier models continue to get better for the foreseeable future. As such, you're dumping alot of money upfront on hardware which is not guaranteed to be relevant in a few years time. The benefits to me seem pretty speculative, minus the point of security.

discordance2• 1 month ago

Running that full load is at least 600 W, so in a day ~14 kWh. At $0.2 a kWH, that would be $2.80/day or $1k a year of op-ex in electricity.

Unless you really want privacy or the fuzzy feeling of owning your own, it’s cheaper, more convenient and has much faster tok/s if you pay a hyper scaler.

That said, I do like the direction we are heading and look forward to seeing what host your own hardware we get in 2 years.

SXX2• 1 month ago

I guess you missed recent news. Problem is that cloud LLM might just sliently sabotage your work by downgrading output model with no notice.

Or cloud LLM might just refuse to sell to you because it dont like your passport.

walrus0110• 1 month ago

Not everyone lives in a place where electricity is $0.20 a kWh. For instance BC Hydro residential rates are $0.11 (CAD) for the first tier and $0.14 for the second tier of consumption in a month. At current exchange rate $0.14 CAD is $0.099 USD a kWh. Hydro Quebec is even cheaper.

At a theoretical 6 tok/s, 86400 seconds in a day, approx 500,000 tokens of GLM5.2 output for 2 bucks a day seems like a pretty good bargain to me. Of course not counting the one time cost of the hardware to run it. But I see people dropping $4000-5000 on all kinds of much less useful stuff.

Additionally in a place where people use electric baseboard heating or electric in floor radiant heating, or really any other heating element based system in winter that's less efficient than a heat pump, additional electrical from a computing load is basically "free" since you would be spending that same money otherwise to heat your house. If a computer with 512GB of RAM is dumping the waste heat into your room, it accomplishes a portion of the same thing as a baseboard.

Not to mention there is a whole other less measurable benefit of having a locally hosted model that can't be turned off or arbitrarily restricted by a service provider, and where all of your queries and context cache aren't subject to surveillance by any third party.

dzjkb• 1 month ago

how do you rent 2 3090s for $2.80/day?

matheusmoreira4• 1 month ago

We do want privacy, and we also want to own the hardware so the US can't just turn it off whenever it feels like it.

I think the main reason not to run locally is to get the full models instead of quantized versions.

poulpy123• 1 month ago

which hyper scaler would you suggest ?

throwawayffffas2• 1 month ago

So in my experience with 2 7900XTs with models that sit fully in VRAM it's more like 400W the gpus spend a lot of time waiting for each other.

DrScientist• 1 month ago

Depends on whether you've also gone for self-hosted electricity generation or not.

segmondy11• 1 month ago

No one locally runs full load all day. The only way to see that is if you're training. We are talking about inference. I limit my GPU to 300watts. You can limit them down to 200w. Since everything is not in GPU and the bottleneck is between CPU/system ram. The GPUs don't even get to spike, I see 160w-180w for each GPU during inference. So redo your calculation again. Figure about 6 hrs of daily inference, and we are down to roughly $125 a year. Thanks again for your speculation.

downut2• 1 month ago

I have rooftop solar and I have been building credit with my electric utility even though the daily high temperature is well over 100F outside and a comfortable 75F inside. That includes running three AMD 12 thread 128GB systems with obsolete GPUs 24x7x365. I'm not a gamer, so 6 years ago I went low-end low-power GPUs. Boy am I dumb. Currently running the qwen3.6:27b, 35b, and gemma4:31b models just fine.

As soon as VRAM prices drop to sanity I'm going to load up and I could care less about the power draw.

Some parts of the future are absolutely great.

bawana2• 1 month ago

Even on a macStudio w 512 gig memory?

eptcyka• 30 days ago

Isn't that still cheaper than the 100 or 200$ plan that Anthropic wants from you?

radku• 1 month ago

I have pretty much almost this exact setup with 2x3090s and with slightly faster DDR4 512GB and 64 core Epyc! [0] I've been enjoying it a lot. Can't wait to give this model a try.

Apart of running local models I use this rig as my main remote development platform. All Claude Code sessions are running there in tmux now. And my fingers can't be happier not having to deal with constantly hot laptop. Not to mention that Claude Code is such a battery hog.

[0] https://medium.com/@rathko/i-built-an-epyc-64-core-512gb-ram...

effisfor• 1 month ago

I applaud all you tinkerers for pushing on the state of the home-brewed art here. Like crypto, AI is drowned out by hucksters, very few people talk about developing resilience. Or the researchers who will push on open source models in efforts to cram them onto an electric toothbrush or tamagotchi. Bravo to you all.

pizza234• 1 month ago

LOL, sure this works if one has a time machine or a LOT of money to burn.

32 CPU Epyc (Epyc is required for faster memory access) + 32 GB VRAM + 512 GB RAM is stupid expensive nowadays, and in best case, it will just downgrade to "very" expensive at some point in the future.

This makes sense only if 1. one is paranoid about privacy or 2. they have money to smoke or 3. they need to workaround cloud model restrictions, AND they have to do it routinely (because if not, a oneshot cloud bare metal setup is way cheaper, faster, and allows more powerful models, due to VRAM offering).

I did spend stupid money as well and yet, the system is 2x slower than cloud providers for comparable performance on vision tasks (I still have to test coding). Oh, and it's hot as hell.

ikari_pl• 1 month ago

I can work out max 90GB to the agents. Advise. :)

SwellJoe• 30 days ago

6 tokens per second is not fit for interactive use. I find Gemma 4 (QAT 4-bit, MTP) to be tolerable at about 30 tokens per second on my old GPUs. Anything slower than 15 is annoying. I tried DS4 on my Strix halo (1-bit quantization of DeepSeek V4 Flash, the biggest model that can realistically run on 128GB, right now), and it tops out at something like 10 or 11 with a long time to first response, and that's quite painful to use. I'd definitely rather spend money to use the big models on cloud infrastructure.

And, the several thousand dollars it costs to run these things unusably slowly buys a lot of tokens on the cheap Chinese models.

Frannky• 1 month ago

There is a push from multiple directions at the same time:

- new AI desktops with GB10s. They are relatively cheap and you can cluster them and load 1TB of VRAM

- Nvidia, amd, intel, Cerebras etc pushing new hardware

- oss models getting crazy good, like glm 5.2

- flash models getting very good like deepseek V4 flash

- quantizations

- harnesses being able to use different models (big for difficult stuff, small for grunt work)

So hopefully soon for the ones who want to break free from APIs, we will be able to host at home a cluster of AI desktops at a reasonable price with Opus-level capabilities, can't wait!!

khafra• 1 month ago

I feel like "relatively" is doing a lot of work, there: at about $4k per GB10, that's $36k for a 1TB cluster. Cheap compared to equivalent H200's, but out of reach for home labs that aren't funded with OpenAI or Anthropic RSUs.

snarfy• 1 month ago

When the AI bubble pops those hardware prices will pop too.

matheusmoreira4• 1 month ago

Hope you're right! Can't wait!

Tepix7• 1 month ago

My hope is on Intel Crescent Island with 480GB. I don't need 8x H200 performance (and cost), but I would like to run GLM 5.2 Q8.

MaKey• 1 month ago

I'd love to too, but I guess Crescent Island with 480 GB will cost something like $10-12k or even more.

drudolph914• 1 month ago

GLM 5.2 is the first time I'm actually excited about AI! I'm not the most bullish on AI code for several few reasons, but the biggest reason is the ownership model. We all know we're near the tail end of the "subsidized pricing" window for AI, and I've been hoping for so long to get an open weight model that is _close enough_ to the SOTA before this window closes - and we actually got it! I'm excited to be able to in the near future run GLM locally, and use these things like a tool instead of living in this for-rent model for the rest of my life. I'm excited to actually enjoy programming again

numlock862• 1 month ago

Is this really worth it, though? Throughout the years my experience with quantized models has been that they feel like a lobotomized version of the original. Doesn't matter if it's an LLM, dedicated diffusion model or some other dedicated task. Sure, they get the job done. But a lot worse. The only ones that can somewhat hold up are the ones provided by the vendor directly. Gemma4 comes to mind. However I suspect they have some secret sauce other than just "let's quantize this" since they have the original model and its data at hand.

There should be more native 4bit, 1.25bit and likewise models. Those actually work great while making them smaller in comparison. But I guess there is some reason for them being pretty niche.

nicman232• 1 month ago

it is not a flat quant but a dynamic

iaw3• 1 month ago

On a model of this size quantization has much less impact on quality of output. I'm running a 3bit version and find it comparable to sonnet, almost opus.

edg50003• 1 month ago

One advantage about local LLM: You could serialize the context yourself, without being constrained by APIs. And let's not forget, the Big 2 encrypt their thinking. If you use custom clients, which is a very grey area alreay, being able to produce the context string raw is a big bonus. Takes away a lot of annoying constraints and needless mystique/obfuscation.

But I don't know how usable GLM 5.2 is vs the Big 2.

c7b2• 1 month ago

Can someone explain the math to me? Why is 1-bit only ten percent less memory than 2-bit?

incognito124• 1 month ago

Keyword dynamic, the parameters are quantized on a case by case basis

idonotknowwhy• 1 month ago

2 reasons.

First, it's not really "1 bit", actually much closer to 2-bit. IQ1_M is actually 1.75bit and IQ2_XXS is 2.06bit This is from the ./llama-quantize --help with most of the quant types and their size in bpw: https://pastebin.com/bCUqGfeE

And to elaborate on the "dynamic" aspect inconito said in the other comment, if you click on one of the .gguf files in huggingface:

https://huggingface.co/unsloth/GLM-5.2-GGUF/blob/main/UD-IQ1...

There are a lot of Q5_K, Q6_K, etc tensors. Only the routed experts (ffn_gate_exps.weight, ffn_up_exps.weight, ffn_down_exps.weight) are heavily quantized, and it looks like the down_proj is actually iq3_xxs for this model.

suyash• 1 month ago

We really need a quantized version for regular laptop

Havoc• 1 month ago

I bet OpenAI and Anthropic hate the timing of glm 5.2.

Kinda shows they have a headstart rather than a magic moat

achrono• 1 month ago

Nope, GLM 5.2 is only the latest and greatest in a long line of open-weights models. There are even fully open source models that are comparable to o1-mini (OLMo), or almost-fully-open ones that are comparable to o3 (Nemotron).

I'm super grateful to the open labs (who, importantly, do not have the word 'Open' in their name), all the more so to the likes of Ai2.

There is no magic moat indeed. It is math, engineering and of course copious amounts of data (and the political maneuvering required to secure it, e.g. how most everyone has trained on Anna's Archive by this point).

walrus0110• 1 month ago

I really don't think anyone is going to have a good time trying to run it on anything with 256GB of RAM no matter what the post says. 512 is the much more realistic minimum. I'm fortunate enough to have two 512GB RAM dual xeon workstations in my home office that I bought cheap before the price rise to mess around with things...

zkmon2• 1 month ago

I have high respect for unsloth's work, helping millions to get started with local AI, but this post appears kind of download bait.

Offloading too many layers to CPU is not going to work at all. I have tried this many times and had to rm -rf on those heavy hf cache folders. Also I doubt 1-bit or 2-bit quants of GLM 5.2, running mostly outside of VRAM can beat Q8_0 of Qwen3.6-27B fully loaded in VRAM - on usefulness.

iaw3• 1 month ago

I run 3bit GLM5.2 and full precision Qwen3.6-27B. GLM is much much closer to frontier models in it's breadth and ability to plan. If you just need to implement Python code from an existing plan Qwen is your choice but it has problem succeeding with more complex tasks that GLM5.2 does not.

As I type this my local GLM5.2 is troubleshooting bugs that Qwen would not be able to handle.

zkmon2• 30 days ago

Not sure how much of your GLM is offloaded to CPU. I was contending the suggestion of using system RAM + VRAM.

maxignol• 1 month ago

Lucky me, I never go out without my 256gb unified ram mac x)

jzer0cool• 1 month ago

1 bit requirement (1-bit 223 GB wowza). What you all recommend with 24-48 vram, or is this approach much out dated now.

cjbprime3• 1 month ago

I've got access to a 192GB RAM Mac Studio, which is below the stated minimum RAM. Can swapping off fast disk be used to make it work out, especially since it's MoE?

walrus0110• 1 month ago

Seems like a good way to shorten the lifespan of an NVME SSD significantly by using up its TB written lifespan, if you let it extensively swap. Also the performance will be absolutely abysmal like 0.1 tok/second.

smallerize• 1 month ago

The LLM tools are smart enough to keep the weights on the disk and read-write stuff in RAM.

cjbprime3• 30 days ago

Both sentences are likely wrong. It's a written lifespan (technically an "erased" one), not a read lifespan. The weights are only needed from disk read-only. And Mac NVMe interfaces are surprisingly fast.

Edit: Oh, I think you maybe thought I meant swapping working RAM off disk? I didn't. I meant swapping weights off disk into RAM on-demand.

bilekas• 1 month ago

this can directly fit on a 256GB unified memory Mac

And yet Apple won't sell them to you anymore. And I'm not too confident it will be even possible to hand then 10k to get one again.

chakintosh• 1 month ago

Breaking even in 2069

jessinra98• 1 month ago

Anyone here tried both Qwen and GLM families on the same setup and found a clear winner for one task vs the other?

stackedinserter• 1 month ago

TLDR: realistically, you can't.

antirez• 1 month ago

DwarfStar work in progress numbers: I see 14 tokens/sec generation, that slopes to 10 t/s with longer 10k or more context size. Consider that the indexed attention requires evaluating 2048 selected rows, 2x DeepSeek and with less compression, so the performances with larger contexts here to south faster. Prefill can be 180 t/s on small contexts to 150 t/s and less with larger contexts. I used DeepSeek v4 PRO in this conditions, it is usable but it is far from the 35 t/s 400 t/s prefill you get with DeepSeek v4 Flash 2 bit on a MacBook m5 max. But likely my implementation is yet not optimized enough, so a bit more performance can be obtained. I'm using 4 bit quants. The model is also definitely less sparse than DeepSeek v4, so it activates a bigger percentage of parameters. If it works decently at 2-bit, that would be a win even for machines where 4-bit fits, since this would mean 2x memory (equivalent) bandwidth basically for the routed experts.

Local inference needs really hard a 1.2 / 1.5 T/s memory bandwidth system with 512GB and 2/3 times the GPU compute of Mac Studio M3 Ultra, at an affordable 10/15k price point. A variant with 1TB memory would also be welcomed at 20k price point.

reasonabl_human• 1 month ago

Thank you for your work on DwarfStar! It is truly helping democratize access to frontier tech.

rbrener• 28 days ago

It is quite cool indeed..!

zozbot2342• 1 month ago

10k context is not a whole lot, this model theoretically supports up to 1M. But the KV cache storage takes up a whole lot more memory capacity at full context than DeepSeek V4 Pro, let alone Flash. (About ~96GB according to readily available KV cache calculators, might be more in practice. For comparison DeepSeek Flash is ~10GB and Pro is at least in that ballpark.) So I'm not sure that this model is a good deal for memory-constrained machines unless you're specifically interested in very short contexts only. This could still be worth it if it came with a game-changing increase in smarts but that seems a bit unlikely so far.

It will be interesting to see how this model does under a SSD streaming scenario, the lower sparsity should ideally be favorable.

Local inference needs really hard a 1.2 / 1.5 T/s memory bandwidth system with 512GB and 2/3 times the GPU compute of Mac Studio M3 Ultra, at an affordable 10/15k price point. A variant with 1TB memory would also be welcomed at 20k price point.

Are these realistic specs at present? Not that clear to me, 1.5 T/s seems really high.

draginol• 1 month ago

The most interesting part of this to me is not the benchmark table, but the packaging.

A model like GLM-5.2 being available as GGUF, usable through llama.cpp/Ollama/vLLM/SGLang/LM Studio, and wrapped for local agent workflows changes the category. It stops being an impressive open model exists and starts becoming this is something a small team can actually put into its development stack.

For instance, company buys an RX6000 setup for say $15k total. They could use this for handling data heavy sifting that would otherwise be a lot of Claude tokens.

It doesn't need to be as good as frontier-best. Just good enough.

I could see a business of people packaging this and handing it to companies who want Help Desk bots without any extra setup.

giancarlostoro• 1 month ago

For instance, company buys an RX6000 setup for say $15k total. They could use this for handling data heavy sifting that would otherwise be a lot of Claude tokens.

Considering they might be spending thousands per month on API costs already, dropping 15K to save on one process might not be bad. On the other hand, also an opportunity to sell GLM 5.2 inference at near cost to other companies for less than whatever Claude costs. In theory it costs anywhere from $0.51 to less than $2 an hour to run it and use it 24/7 that's still wildly cheaper than calling Opus which doesn't bill per hour, but per million tokens, drastically higher. Hell, you could probably bill at $5 per GPU hour and still be cheaper. Whether you're looking to self-host or sell hosting for it, it looks way cheaper regardless. I think most decent open models will continue to fit in at least 32GB of VRAM so a 6000 Pro GPU is more than enough. alternatively, even on a 5090 you can get a reasonable amount of inference for way less than paying for Opus, Qwen would be your friend there though.

storus• 1 month ago

So a minimum of 3x RTX Pro 6000 to run 1-bit at ~76% accuracy or MacStudio 512GB RAM to run 4-bit at ~97% accuracy.

iaw3• 1 month ago

No. Unsloth has CPU offloading. It'll be slow but it'll work even with SSD offloading.

GLM-5.2 – How to Run Locally

Related Stories