CryptoTrigger (@CryptoTrigger1) header
CryptoTrigger (@CryptoTrigger1) profile picture

Freedom and Liberty. @chutes_ai full-port

Joined February 2021
Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)Photo by CryptoTrigger (@CryptoTrigger1)
CryptoTrigger retweeted
Among #Bittensor’s ten largest subnets, @chutes_ai leads on both whale-risk measures: 1st of 126 for liquid concentration and 3rd for overhang. 👋
1
1
11
495
CryptoTrigger retweeted
TAO is a pure fighter, amongst all this turmoil, drama, people whinning on X, it keeps on getting up after numerous knock downs. Wether from protocol changes, scammers, extractors or grifters, it has not given up. Up 6% on the day while btc is down 3% thats 9% difference. This asset is accumulating serious, prestegeous battle scars. We're in the right path.
14
1
48
1,911
CryptoTrigger retweeted
Big tech is spending around $810 billion on AI infrastructure this year, going by the slide Jon showed at Exploit. The money keeps coming. Power is the harder part. In July, one transmission fault in Ashburn, Virginia knocked more than 3 GW of data center load off the grid at once. Parallax trains on ordinary GPUs spread across lots of places, so it doesn’t need one giant building plugged into one stretched grid.
7
22
2
82
2,311
CryptoTrigger retweeted
The paper, if you want the math: chutes.ai/parallax.pdf
1
3
398
CryptoTrigger retweeted
Most of the work inside an AI model is multiplying numbers. Parallax stores most of its weights as -1, 0 or 1, and at least half of those are zero, which a GPU can skip. On one RTX 5090 that’s the difference between 209 teraflops and 3,352. Part of the jump is the card’s 4-bit mode, and skipping the zeros doubles it again. Those are the chip’s rated speeds, not measured ones. It’s a big part of how one card hit almost 18,000 tokens a second.
3
8
60
1,572
CryptoTrigger retweeted
How is it this fast? A small model, 4-bit math for the expert layers, and a fixed-size cache for most of the attention. The Parallax paper: chutes.ai/parallax.pdf
1
12
788
CryptoTrigger retweeted
Parallax runs at almost 18,000 tokens a second on a single RTX 5090. That’s one consumer GPU answering 256 requests at once, on a vLLM fork that we put together. The more one card can serve, the cheaper each answer gets 💪
12
33
6
168
12,895
CryptoTrigger retweeted
NEWS: Tencent, whose Weixin and WeChat platforms reach 1.439 billion monthly users, has selected Chutes, a Bittensor subnet, as a launch partner for its new flagship AI model, Hy4. The partnership puts @opentensor's decentralized AI infrastructure alongside one of the world's largest technology ecosystems. Chutes offers competitive inference pricing at $0.525 per million input tokens and $1.5625 per million output tokens. This collaboration could bring greater visibility to decentralized AI networks and highlight their growing role in supporting large scale AI applications.
19
135
23
638
24,638
CryptoTrigger retweeted
Replying to @RadoTsc
They clarified they won't be participating as a miner for this purpose, so no miner emissions will be directed to this. To my understanding, they believe they can fund this with revenue (which targets to exceed any costs related to it).
1
2
42
CryptoTrigger retweeted
GLM 5.2 Fast is live on Chutes. Z.ai's GLM 5.2, for chat and agents where the reply is on the clock. 1M-token context, 128K max output. Here's what GLM 5.2 is good at, on Z.ai's published scores: 81.0 on Terminal-Bench 2.1, four points behind Opus 4.8 at 85.0, and 62.1 on SWE-bench Pro, up from GLM 5.1's 58.4. Z.ai built it for long-horizon tasks. Base URL llm.chutes.ai/v1, model zai-org/GLM-5.2-Fast, same key as every other model ofc. $2.45 per million input tokens, $7.70 out.
4
5
2
49
2,874
CryptoTrigger retweeted
All is well in the world: @tradingview shipped another absolutely terrible mobile update
97
52
7
1,550
75,965
CryptoTrigger retweeted
The midterm results, regardless of your opinion on them, come with a fairly high degree of uncertainly around the future of new datacenter capacity (and/or costs) in the US. Parallax is no longer really optional IMO (though it was already inevitable). “You never change anything by fighting it; you change things by making them obsolete through superior technology.” - R. Buckminster Fuller Status updates: Almost done with dataset prep. It turns out ~10t tokens of high quality data tokenized and properly upsampled and so on is a lot of work and compute. Apple M5 and DGX spark support locked in, still waiting on my AMD Ryzen AI MAX+ PRO 495 from lucebox to get that supported as well. Still chugging away at the final touches on the code/agentic pipeline also (but that would be after the pretraining anyways, non-blocking). Settled on the openai o200k tokenizer, updated to always use single tokens for digits for better math capabilities (hopefully). (and more to come soon)
5
18
2
117
3,417
CryptoTrigger retweeted
Here's a very crude simplified chutes roadmap/plan, for near term: - move the most unprofitable things off of the subnet, including today's announcement of some (not all) additional inference options (not paid via emissions) - this allows us to scale the things we want and need to avoid capacity issues/better direct incentive - this includes models we use to build data pipelines and relationships to other teams/networks/etc - add options that are completely revenue positive (hourly pricing/rentals/simple docker based jobs/etc), TEE, private, secure, priced above cost, more extensive version of private chutes we have today basically - expand the hourly compute offerings to also include CPU only (e.g. to run agents, web browsers, sandboxes, non-GPU work of any kind) - use some of the incentive for ideally non-datacenter based model training and serving - build a non-extractable product on chutes from the models we build, via dynamic edge nodes, possibly from the epigenesis work (smaller faster cheaper faster than lora customizations from the sparse sections of pretrained model, participant supplied) - models and pipeline etc. will be open source, the customization and ensemble thereof will live only in chutes, inc. the continuously evolving set of served variations/RL/etc. We'd like, to the extent we can, attract participants who want to build and have no obligation to pay datacenters or any particular hardware vendor, to keep the value inside the subnet at all times by attacking sell pressure at the source (DC commitments, chip purchases, etc). We will continue to have secure TEE inference and GPUs; not elimating anything wholesale, just focusing more on a commodity that doesn't have to flow out to DCs and single hardware vendors. NVIDIA reportedly has 75% profit margins, revenue for the sake of revenue fueling GPU reselling goes straight to one place, basically. Fine, I guess, but I'd rather work together as individuals with whatever hardware we can bring, to a reasonable degree.
9
21
4
114
6,083
CryptoTrigger retweeted
True? No clue. The point of parallax is to put the network into as many hands as possible in a verifiable trustless way, and for the output (the models) to not require yet another source for the wealthiest companies on earth to monopolize (that is, a model that can run even on a phone and not require "now's" completely unaffordable mega GPU). The current AI marketplace is fueled by a massive debt market with very few beneficiaries at the world's expense. If you like K shaped economies here it is in plain sight; trickle up resources, consolidating into a few, spreading the gap at ever-increasing speed. Whether this is self serving for Mr. Musk to gather up more resources for pennies on the dollar or a true statement is irrelevant, the outcome is the same. Will it crash tomorrow, next week, never? No clue, but I would rather not build the foundation on this gamble. I'm possibly (probably?) doing it wrong, but the goal is and has always been, people first, which means accessibility first. Let's make "good enough" models with no kill switch and no hardware moat. Please and thank you. If we get superintelligence out into the universe without/at the cost of consciousness to witness it, we have accomplished nothing, and have in fact forgotten the whole point.
Replying to @chamath
Pretty much. As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers. SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue. That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
2
13
57
2,283
CryptoTrigger retweeted
So let me get this straight $VVV is 40% of the WHOLE bittensor ecosystem, it is valued at +6,66x the Fully diluted mcp of chutes (300m vs 2b) and pretty much has the same selling points as chutes? Let's not even begin mentionning the new @chutes_ai parrallax branch, a boiling pot of discoveries imminent. I am sitting comfortably in root and in chutes. I dont care about the rest.
8
3
40
1,867
CryptoTrigger retweeted
SN 64 @chutes_ai with 800k users, phD research level papers on AI bottlenecks, real world implications, combatting hardware shortages with smart training systems, is valued half of a gpu rental marketplace SN 51 @lium_io. HALF BROTHER, we're talking HALF. Something that adresses global, real world commodity bottlenecks. I am feeling refreshed and confidently holding. I want to part take in this movement. Everything will fall into place.
7
7
68
2,622
CryptoTrigger retweeted
The problem is not pow registration. It's Null Consensus. This is the culmination of the unresolved Validator weight-copying problem. Whether const mines or not i dont even care because even if he says he doesnt, nobody can prove it. Null Consensus means Yuma Consens is not useful all the time and that the purpose of Validators is not as relevant as it should. It's admitting that the incentive system is flawed at the validation level and that it needs to be bypassed in some cases. So Yuma Consensus, which was the moat of Bittensor because it was meant to make products trustless and miner scoring fair, is now openly declared irrelevant on-demand. Why do you think there are even big entities using the Yuma name in their brand? because it was a central aspect that made Bittensor uniquely positioned to be a legit decentralized network where proof of useful work is verified mathematically instead of using trust me bro. With this admission comes eTao next which will allow subnet owners to make the emissions split different than 41% valis / 41% miners / 18% subnet owners... yeah i fucking hope the subnets with no validator purpose will reduce the 41% sent for validators, no shit Im tired of all these wild swings that look like yolo decisions to test results in production Nobody can say anything really, it just happens and we must adapt. Fair. But this isn't a 1000-year decentralized federation lol This is a cool database that Const can shut down when he wants, with ponzi-like micro companies within a niche of another niche (crypto)
4
2
1
15
1,522
CryptoTrigger retweeted
github.com/chutesai/epigenes… Still experimenting, but seems to check out. Smaller, faster, better? Models that grow with you (or your repo, or proprietary company docs, etc.) Short term exact recall is the context window. Context window feeds training continuously, binary optimizer. This becomes stored knowledge and higher order perceptual understanding over time, not just RAG. Sparse MSA global attention already selects most important tokens to attend to in longer context, so you can reinforce those parts more.
🤖 Made with AI
There's another magic trick (aka cool math) I believe this parallax enforced sparsity enables by the way... arxiv.org/abs/1711.05769 <- prune weights, adapt the pruned weights, sharply cuts catastrophic forgetting risk. If you have a model with the enforced sparsity, 50% of the weights are locked in to particular patterns. What if you mask the other expert weights leaving only those sparse contiguous weights to learn new data? That's the intuition, perhaps it works, perhaps not. But... if it does, it would make the base model significantly more malleable for new domains/tasks/whatever. SFT/continued pretraining with less risk of catastrophic forgetting and without lots of magic tricks and RL and re-sampling in your own words etc.
6
10
53
2,305