deadmanoz

- Independent Researcher
- Perth WA BitDevs & Meetups
- perth.bitdevs.com.au
- pq-bitcoin.org

Public Key

npub12msugd9s2xz5p6flnm9ygsx2zcgnnr29tw0ezq53z3p06xag70vqv6pvdp

NIP-05 Address

Profile Code

nprofile1qqs9dcwyxjc9rp2qayleajjygr9pvyfe34z4h8u3q2g3gsharw508kqprpmhxue69uhhqun9d45h2mfwwpexjmtpdshxuet5qy28wumn8ghj7un9d3shjtnyv9kh2uewd9hsh7q25n

Publishing to

Author Public Key

npub12msugd9s2xz5p6flnm9ygsx2zcgnnr29tw0ezq53z3p06xag70vqv6pvdp

Show more details

Last Notes

2026-07-22 11:34:56 CEST

I've published my Bitcoin merge-mining research repo: data and tooling for about 26 chains that have merge mined with Bitcoin. Recovered stale blocks a focus.
Or, put another way, I ran nodes for about 20 non-Bitcoin 💩 chains, so others didn't have to!
https://github.com/deadmanoz/merge-mining-research

2026-07-20 08:21:26 CEST

- reply

“Honest take”
#nevent1q…p2kz

2026-07-19 14:03:20 CEST

Self hosting will increasingly become necessary, for privacy AND defensive purposes
Recount from huggingface regarding a recent security incident, model provider guardrails hampered investigation and response effort, fortunately they could self host
https://huggingface.co/blog/security-incident-july-2026
https://blossom.primal.net/5060f2f64b9505c201a95184b5faec94fe5830906f6950256498d4543a221919.jpg
https://blossom.primal.net/1d88ec1ead91f256a6222b30d5dd5dd84c4e21bcb934e990c5d5ffca0f88cf65.jpg

2026-07-19 13:56:08 CEST

- reply

A Cambrian explosion of chains though, one for each late converter

2026-07-19 04:48:44 CEST

- reply

Get around your local BitDevs!!
#nevent1q…edh2

2026-07-19 04:45:02 CEST

AI Psychosis in software dev is having to read phrases like “load-bearing”, “seams”, “sidecar” constantly

2026-07-18 08:00:21 CEST

- reply

Too bad it’s also really slow..
Chinese labs need to catch-up on inference infrastructure for scale!!
#nevent1q…zv86

2026-07-17 12:21:24 CEST

It might be true..
I’ve been using Kimi occasionally for a few months, mainly only when I had no usage remaining on Claude/Codex, because it was much, much weaker
But K3 does actually seem to be a very capable model from my brief experimentation
Now if only I had the hardware to self host it 🥲

2026-07-16 13:18:11 CEST

- reply

I got 21 at the cafe in Kings Park today

2026-07-16 13:17:36 CEST

It has been a while, good return - daughter “slayed it”
WORD5 #650 3/6
⬛⬛⬛🟧⬛
🟪🟧🟧🟧⬛
🟪🟪🟪🟪🟪
https://otherstuff.ai/word5/

2026-07-16 13:13:30 CEST

- reply

I can’t wait to travel international and have internet on the plane

2026-07-15 04:49:36 CEST

- reply

Haha wow

2026-07-15 04:46:19 CEST

- reply

I’m going to be attending and speaking!!
#nevent1q…rcrw

2026-07-02 13:43:26 CEST

- reply

Always has been

2026-07-02 11:19:35 CEST

An interesting preliminary observation from the Merge Mining Monitor, 20 stale blocks in 20 hours produced by Foundry on 11/12 September 2025.
An event that cost them dearly, at least ~46.9 BTC lost in fees.
https://mmm.deadmanoz.xyz/?tree_time=2025-09-18T16%3A21%3A51Z&selected=00000000000000000000ba485fa4bbba0416890572ff6b5853cdc24e92d7434f
https://bnoc.xyz/t/introducing-the-merge-mining-monitor/146/3
https://blossom.primal.net/f39895ecce427039f35ab8b3b4f481ac0cd448f1f181726e569b734919b5d56e.png

2026-07-02 08:05:57 CEST

Introducing the Merge Mining Monitor, another tool in service of the Bitcoin monitoring mission.
Merge mining lets child chains ride on Bitcoin's PoW, preserving the parent headers miners hashed.
Bitcoin keeps no durable record of valid blocks it later discards (stales/orphans). Merge-mined chains do.
The monitor turns that merge-mining evidence into a browsable record of who mined each Bitcoin block, which child chains rode along, and which discarded blocks survived in child-chain history.
The current dataset recovers 2,200+ stale blocks back to 2011, when Namecoin first brought merge mining to Bitcoin.
https://mmm.deadmanoz.xyz/

2026-07-02 08:05:09 CEST

2026-06-15 11:07:54 CEST

This looks really fucking rad from Block, h/t @nprofile…l4na & @nprofile…yxhn & @nprofile…660p
https://github.com/block/buzz

2026-06-10 10:11:18 CEST

Fable 5.. probably at least 80% of my sessions are “safety measure” downgraded within the first few prompts
Guess I have no reason for “extra usage” come June 22nd

2026-05-21 16:18:04 CEST

- reply

And nice work to the original author too!!

2026-05-14 07:31:56 CEST

I needed to understand merge mining for a particular research direction I am pursuing (relating to stale blocks), and created the following visualisation-rich piece on the topic.
Hoping this makes it easier for some to grok how it all works, especially AuxPoW!
https://deadmanoz.xyz/posts/2026/merge-mining

2026-05-12 13:12:11 CEST

- reply

Love the sound and audio, so tight!!!
Hard to play on mobile, gonna give it a whirl on computer

2026-05-11 01:20:07 CEST

- reply

Unfortunately this isn’t a constructive discussion, and doesn’t appear like it ever will be, cheers

2026-05-10 03:40:05 CEST

- reply

@nprofile…jrtm I have read the Capture articles, the main thing I notice is following aspect of human nature coming through
Humans will tend to preferentially associate with and select for individuals they get along with, like and respect

2026-05-09 15:57:40 CEST

Great overview video of the priorities, methodologies and achievements of Bitcoin Core from @nprofile…mn80 at OPNEXT’26 a few weeks ago
https://youtu.be/F3ONZsG-SY0

2026-05-02 16:28:43 CEST

- reply

That’s cool!!
#nevent1q…gq0p

2026-05-01 11:47:50 CEST

When an agent is deep in some obscure low-level code I don’t really have an interest in, and my eyes start to glaze over..
..I finally have a glimpse of what it is like to be a non-technical product manager

2026-04-27 06:49:49 CEST

Torn between liberal democracies needing to win the AI race and the best and most competitive open source efforts not coming from liberal democracies

2026-04-23 12:22:16 CEST

- reply

Look at those dream boats

2026-04-17 14:28:56 CEST

- reply

The Big Stuff
https://blossom.primal.net/2379a18a8452e108d0bfb6c0bd254444ac145fd98464111af51742aa00651996.png

2026-04-17 14:28:22 CEST

- reply

For Science™️…
https://fountain.fm/episode/1mieRW00BVL4hsQQHBYd

2026-04-02 08:11:00 CEST

Realised I had a lot of knowledge gaps to fill so I filled some of them in writing this piece
http://deadmanoz.xyz/posts/2026/bitcoin-node-roles-p2p

2026-03-25 06:05:12 CET

Pick almost any aspect of Bitcoin and prepare to be humbled and ready for a deep rabbit hole.
Currently digging into addr, addrv2, getaddr, et al and boy, there’s much more to these aspects than appears on the surface.
"Token bucket" rate limiters, 6 or 7 different peer types, 24-hour cache for getaddr response, etc etc 😅

2026-03-23 05:19:30 CET

Pretty impressed with qwen3.5 35B as a private local model.
Integrated vision, decent tool calling, reasoning/thinking that can be toggled.
It’s slow but has proven adequate in an OpenClaw harness for a few workflows I’ve got - private document interpretation/processing/organisation
I do wonder what the hardware requirements are for such local models to be capable of most general consumer use cases (this model at 35B is not there), probably100s of GB at this point in time?

2026-03-21 11:44:16 CET

Remember machine-to-machine payment discussions that were all the rage a few years ago??
Maybe we will finally see some traction in this area with agentic payments!!

2026-03-19 06:49:40 CET

- reply

Codex definitely has a higher pass mark than Opus, often finds issues that Opus misses
That being said, Opus is a better collaborator, it’s more explicit in what it does and is just more “fun” to work with

2026-03-19 01:51:14 CET

- reply

True, I had not even thought of that!!

2026-03-19 01:50:52 CET

- reply

The stories are also so very varied, what an imagination

2026-03-19 01:46:06 CET

- reply

Yeah, that Sanderson finished the series is good supporting evidence.
I’ll give it a whirl, thanks @npub1jy2…9p6d

2026-03-18 16:28:35 CET

- reply

Yeah I’ve got that a lot, I think the heavy magic focus (from what I understand) has prevented me from checking it out.. but yeah I really did enjoy Mistborn 1-3, which pretty much involves magic

2026-03-18 16:04:53 CET

When I was younger I read widely but was very much into Tom Clancy, Matthew Reilly et al.. stories that were real or realistic(-ish).
Not much sci-fi or high fantasy (although I smashed Harry Potter like many of the generation)
In the last few years I’ve really enjoyed branching out into these genres.
Iain M. Banks’ Culture Series… the opening scene of Consider Phlebas had me hooked!
And the trilogy of first Mistborn books was 🔥🔥🔥
Open to recommendations

2026-03-18 09:22:46 CET

Very happy to renew @npub1xsg…zazx for a 3rd year 🎉
Kagi News (free for all) has been a great addition to the product suite in the last year
And looking forward to Search API coming out of beta one day soon too…

2026-03-16 06:08:06 CET

I think forging higher living standards in the developed world now means higher expenditure of “intelligence tokens”
(recognising that this just reduces to energy usage, like the original correlation)

2026-03-15 06:12:50 CET

Self-hosted (private) agents that can navigate the web (well) to do your bidding would be 🔥🔥🔥

2026-03-12 13:32:55 CET

- reply

Yep, just ask McKinsey
https://codewall.ai/blog/how-we-hacked-mckinseys-ai-platform
#nevent1q…gk59

2026-03-12 04:41:08 CET

- reply

Great Aussie Tallow Soap!!
#nevent1q…frkk

2026-03-12 04:39:57 CET

I don’t really understand all the hype about Karpathy’s autoresearch.. I mean it’s on the path to the self-improving AI flywheel, but it’s like a small step from hyperparameter tuning which we’ve been doing for many years..
.. unrelated, I do wish there were more ML/AI researchers/practitioners posting here on Nostr!

2026-03-11 14:25:40 CET

- reply

I’d wear it!!

2026-03-09 00:01:11 CET

- reply

Ditto
WORD5 #520 5/6
⬛🟧⬛⬛⬛
⬛⬛⬛🟪🟪
⬛⬛⬛🟪🟪
⬛🟪⬛🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-03-07 02:56:55 CET

- reply

Yep qwen3.5 is a very good model

2026-03-05 01:41:20 CET

- reply

With a H

2026-03-04 04:27:17 CET

What’s the go here @npub14m9…f2c4??
https://blossom.primal.net/8d8d3711a63da55e6195fc03c45fde330f298bd7b48aee3429ebe65c64508091.png

2026-03-03 14:56:07 CET

📣📣📣
Perth BitDevs/Bitcoin Meetup for March 2026
When: Thursday 5th March, 6pm
Where: Rocky Ridge Brewing Co, Burswood
Topics:
* Bitcoin Core v31 feature freeze - cluster mempool (SFL), embedded ASMap, private tx broadcast
* Gloria Zhao steps down as Bitcoin Core maintainer
* BIP-110 update - cache poisoning bug, defections, OP_RETURN war, prediction markets <10%, first signalled block
* NestedMuSig2 security proof (Nadav Kohen, Chaincode Labs)
* Quantum resistance - BIP-360 P2MR, algorithmic agility, the quantum freeze
* PIPEs v2 - covenants & ZK proofs without consensus changes
* Binohash - transaction introspection via DER signature length
* P2P protocol BIPs - stale tip relay & BIP-434 peer feature negotiation
* FIBRE resurrected - UDP + FEC block relay back online (~30ms propagation)
* Sloth - secure group messaging over Nostr (MLS/Marmot protocol)
* Phishing attacks targeting Cake Wallet & Sparrow Wallet users
* ...and more!
Suggestions are open for additional topics to discuss!
https://github.com/PerthBitDevs/PerthBitDevs/issues/32
And be sure to check out the new fresh paint on the website
https://perth.bitdevs.com.au/

2026-03-02 13:02:19 CET

- reply

WORD5 #514 5/6
⬛⬛⬛⬛⬛
⬛🟧⬛⬛⬛
⬛⬛🟧🟧🟪
⬛🟪🟪🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai
American spelling annoys me 😢

2026-03-01 05:01:55 CET

- reply

Cool story, bro

2026-02-28 10:27:34 CET

- reply

BLM - Black Lives Matter? I’m not steeped in this enough to know about these intricacies.. I’m a foreigner, albeit still a citizen of the 5 eyes (Australia), so I’m seeing this from afar!!
I did like Palmers take, and resonated with a lot of what he discussed on Joe Rogan, but I think it needs to be recognised that Anthropic have probably always held the position they have now, and are just “sticking to their guns”?
As in, this was always their stance, just now that their tech has become super-capable, and increasingly so, the US government wants it used for purposes that Anthropic never agreed to?

2026-02-28 08:59:01 CET

- reply

Where’s that from?

2026-02-28 08:58:30 CET

- reply

I would like to think it’s not but who knows what’s actually happening in the background

2026-02-28 08:50:39 CET

- reply

Do you think Anthropic’s stance is just theatre?

2026-02-28 08:06:14 CET

Amen:
“Whatever the U.S. government does to threaten Anthropic, the AI company should know that their corporate customers, the public, and the engineers who make their products are expecting them not to cave. They, and all other technology companies, would do best to refuse to become yet another tool of surveillance.”
https://www.eff.org/deeplinks/2026/02/tech-companies-shouldnt-be-bullied-doing-surveillance

2026-02-28 00:41:27 CET

- reply

Was looking grim for a moment..
WORD5 #511 4/6
⬛⬛⬛⬛⬛
⬛⬛⬛⬛⬛
⬛⬛🟧⬛⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-27 00:35:44 CET

- reply

WORD5 #510 4/6
⬛⬛🟧⬛🟧
⬛⬛🟧🟧⬛
⬛🟧⬛🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-26 15:41:49 CET

Life imitates Bluey - Sleepytime
Probably familiar to many parents, there are at least 3 bed swaps every night in our household, sometimes more
Here’s the full episode (480p), the sound and music in this one is top-notch
https://blossom.primal.net/a4e8a347fb894ef0b1fd24ae87cef88c0ca1c2d7644893413981974cd9dc123a.mp4

2026-02-26 00:43:50 CET

- reply

Baller for sure!!
WORD5 #509 5/6
⬛⬛⬛🟧⬛
⬛⬛🟪⬛⬛
⬛⬛🟪🟪🟪
⬛⬛🟪🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-25 00:09:41 CET

- reply

Yes, there are limits, just like Bitcoin’s consensus rules.
The boundary of free speech, and Bitcoins censorship resistance, should be what is maximally permissible.
Any changes to this boundary should be heavily scrutinised and require significant agreement to actually occur. Maybe even require a higher bar than “rough consensus”?

2026-02-24 23:51:05 CET

IMO, there is a strong analogy between Bitcoin’s censorship resistance properties and free speech
Just as free speech permits abhorrent speech as the necessary cost of a free society...
...Bitcoin's censorship resistance necessitates non-monetary uses as the cost of sound money.

2026-02-23 13:17:36 CET

- reply

Hot damn.. first I’ve actually failed
WORD5 #507 X/6
⬛🟧⬛⬛⬛
⬛⬛🟧⬛⬛
⬛🟪⬛🟪⬛
⬛🟪⬛🟪⬛
⬛🟪⬛🟪🟪
⬛🟪🟧🟪🟪
https://word5.otherstuff.ai

2026-02-23 05:17:08 CET

- reply

Oh that is a great analogy for compaction!!!
Yeah played a bit of Factorio long ago, never really had the staying power of Civilisation for me though (which I been playing since Civ II). Mainly cause I played with mates and I was too slow..

2026-02-23 00:16:06 CET

- reply

WORD5 #506 4/6
⬛⬛🟧⬛⬛
⬛⬛🟪🟧⬛
🟪🟪🟪⬛🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-22 05:40:28 CET

For me, AI coding has a bit of a Civilisation (the game) kinda vibe..
..Just one more prompt (turn)

2026-02-20 13:44:26 CET

- reply

WORD5 #504 5/6
⬛⬛🟧⬛⬛
🟧⬛🟧⬛⬛
⬛🟪⬛🟪⬛
🟪🟪🟪🟪⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-19 15:39:23 CET

- reply

Thanks mate, I need to get this onto some new hardware…
Spinning rust is not ideal right? I recall you have some issues like 8 months ago with slow disks IIRC?

2026-02-19 15:31:28 CET

- reply

Yeah.. came to me finally as I was brushing teeth
WORD5 #503 5/6
🟪⬛⬛⬛⬛
🟪⬛🟪🟧⬛
🟪🟪🟪⬛⬛
🟪🟪🟪⬛⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-19 13:17:21 CET

- reply

What’s the DB size like these days?
I ran out of space on my M1 so had to delete a bunch of of shit, including BRK 😢

2026-02-18 13:43:08 CET

Ugh cascading computer hardware failures are the worst 🤬

2026-02-17 14:48:49 CET

- reply

Agents allow compartmentalisation of capabilities and access, which I think can open many use cases not possible with single models
And that’s to say nothing about the preservation, or rather distribution of context, between agents, potentially allowing much better performance on long task endeavours

2026-02-17 03:42:14 CET

I am a bit of a data hoarder but finally today took the plunge and deleted TBs of data from previous employment and projects.
TB because it was a lot of video and imagery used to train ML!
It’s a world I am not going back to, so about time I had a clean break.
Now I can start to hoard data for my new line of work, Bitcoin network monitoring 👹

2026-02-13 16:54:51 CET

In a matter of a few hours I’ve got a completely private via Signal and self-hosted LLM (qwen2.5-tools) agent to interact with and use. It’s “self-configuring” and pretty capable for what it is (32B parameter model).
It’s a tool I can comfortably use with my private documents, without worry of leaking to a provider.
I understand the hype..

2026-02-13 08:07:53 CET

Got a separate (e)SIM to experiment with sandboxed OpenClaw
Not quite “fuck around and find out” but let’s play
https://blossom.primal.net/eeebce6ddd889b63831865dc1faab4596c628b6b279465de415393a9634752cf.jpg

2026-02-12 23:30:02 CET

- reply

Quite a few variants possible here…
WORD5 #496 4/6
🟧⬛🟧⬛⬛
⬛🟪🟪🟪🟪
⬛🟪🟪🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-11 00:19:36 CET

- reply

WORD5 #494 3/6
⬛⬛⬛🟪⬛
🟧⬛⬛🟪🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai
Healing 🥰

2026-02-09 13:26:04 CET

- reply

WORD5 #493 4/6
⬛⬛🟧⬛⬛
⬛🟪🟧🟧⬛
⬛🟪🟧🟪⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-08 00:56:00 CET

- reply

Worst one for me yet (have missed doing a day or two recently though)
WORD5 #491 6/6
⬛⬛⬛🟧⬛
🟪🟧⬛⬛⬛
🟪⬛🟪⬛⬛
🟪🟪🟪⬛🟪
🟪🟪🟪⬛🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-07 12:04:13 CET

- reply

I was wondering if Saylor was going to do ANYTHING regarding funding of Bitcoin R&D so I think it’s good he is contributing to something..
…hopefully in a meaningful way, leveraging the existing approaches (e.g OpenSats) or similar such structures

2026-02-07 02:23:50 CET

Max Cooper @ Perth Festival
https://blossom.primal.net/7b324257dfc6d7aa6e1beacb0183c086faf771bd52a2057ae5a6b078512dcc9e.mov

2026-02-06 06:16:34 CET

Got the new GreenBit satboy v2 last night at BitDevs 🍃
SILENT ~1.1 TH/s miner 🤫
Designed, developed and assembled in Perth, WA 🇦🇺
https://greenbit.tech/
https://blossom.primal.net/40c393070b1bbce965a8e60de7a8a5e70f4b74e99b4836c048dca4b00c88cf67.png
https://blossom.primal.net/2843645ea6c86069555391f63b05d7c992bda04db766cf7d4d1e8a8712286340.jpg
https://blossom.primal.net/39d84efdf2bf7b08e430e6af5015abae54a2c3241ddfff55d2f1357fd9f834d4.jpg

2026-02-06 02:08:56 CET

Gloria Zhao
Thank you for your many, many contributions, and you service and dedication to Bitcoin Core over a number of years
The incessant bullying and vitriol that has led to this has been very sad to see
Unfortunately I think this will lead to contributors being less willing to be “on the record” or talking to journalists or media etc, but understandably they won’t want to open themselves to what we’ve seen happen to Gloria
(maybe come back as a nym when you’re ready)

2026-02-05 14:14:49 CET

- reply

Thanks for the new thread
Quickest one in many days, daughter will be mad I did it without her 😬
WORD5 #489 4/6
⬛⬛⬛🟧⬛
⬛🟧⬛⬛⬛
🟧🟪🟪🟪⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-05 14:06:53 CET

- reply

There isn’t (yet?), otherwise we’d be doing it in earnest

2026-02-04 13:31:17 CET

- reply

Wat

2026-02-04 05:08:04 CET

📣📣📣
Perth BitDevs/Bitcoin Meetup for February 2026
⚠️ Note the change of venue (trialling…)
When: TOMORROW Thursday 5th February, 6pm
Where: The Loft Lounge & Bar @ InterContinental Perth City Centre
Topics:
- BIP 110 (formerly BIP-444)
- Bitcoin Core v30 wallet migration bug (fixed in v30.2)
- OpenClaw (formerly Clawdbot/Moltbot) & Moltbook
- Bitcoin Core: Broadcast own transactions via short-lived Tor or I2P connections
- Price action 📉
- New Bitcoin Core maintainer: TheCharlatan (sedited)
- Hash-based signature schemes for Bitcoin - quantum resistance (SPHINCS & SHRINCS)
- 64-byte transactions and BIP-54 "Consensus Cleanup"
- Tether open-sources MiningOS
- OpenClaw Sovereign Setup
Suggestions are open for additional topics to discuss!
https://github.com/PerthBitDevs/PerthBitDevs/issues/30

2026-02-04 02:13:17 CET

- reply

This is the way!!
#nevent1q…rwwj

2026-02-04 00:52:28 CET

- reply

I think we need to start a new thread, spend ages to get to the bottom 😬
Credit to my wife for pulling this one out at the end!!
WORD5 #487 4/6
⬛⬛🟧⬛⬛
🟪⬛🟪⬛⬛
🟪⬛🟪🟧⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-02-02 23:11:13 CET

- reply

WORD5 #486 4/6
⬛⬛⬛⬛🟧
🟧⬛🟧🟧⬛
🟧🟧⬛🟧🟪
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai
This one had me for a while

2026-02-01 12:47:47 CET

- reply

WORD5 #485 4/6
⬛⬛⬛🟧⬛
🟪🟧🟪⬛⬛
🟪🟪🟪⬛⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-01-31 03:08:03 CET

- reply

WORD5 #484 4/6
🟪⬛⬛🟧⬛
🟪🟪⬛⬛⬛
🟪🟪⬛⬛🟧
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-01-30 09:22:24 CET

- reply

🤑
But yeah, for sure, I’d love something like this too!!

2026-01-30 09:15:27 CET

- reply

“The 1.8-bit (UD-TQ1_0) quant will run on a single 24GB GPU if you offload all MoE layers to system RAM (or a fast SSD). With ~256GB RAM, expect ~10 tokens/s. The full Kimi K2.5 model is 630GB and typically requires at least 4× H200 GPUs.
If the model fits, you will get >40 tokens/s when using a B200.”
https://unsloth.ai/docs/models/kimi-k2.5

2026-01-30 01:06:22 CET

- reply

WORD5 #483 3/6
⬛🟧🟪⬛🟧
🟪🟪🟪🟪⬛
🟪🟪🟪🟪🟪
https://word5.otherstuff.ai

2026-01-29 09:03:03 CET

- reply

Damnit, I really am a kid of the 90s, I say “like” a lot