Pinned Tweet
hey i'm sudo and yes, x changed my life, all i did was keep posting
this is the worst month of my life at 27 and i didn't stop posting, because the timeline never let my effort down
x helped me connect with brands and paid me well, consistently, if it wasn't for x i'd be on the street or stuck in some 9 to 5 job
when i started posting a year ago people around me threw shots at me, asking what i do at home on a computer all day and telling me to get a proper job, but i believed in the timeline and kept posting, and it didn't take long for the same people to go quiet and fade out of my life
i've learned a lot along the way, in ai and in life, and i've come to a point where i care for my peace first, before anything, and nobody who could take it away gets close to me
so fuck you to those pathetic pieces of shit, and thank you to the timeline
this is the new era and i'm here to win, i must not get distracted by the ups and downs, i forge through one day at a time
this is the worst month of my life at 27 and i didn't stop posting, because the timeline never let my effort down
x helped me connect with brands and paid me well, consistently, if it wasn't for x i'd be on the street or stuck in some 9 to 5 job
when i started posting a year ago people around me threw shots at me, asking what i do at home on a computer all day and telling me to get a proper job, but i believed in the timeline and kept posting, and it didn't take long for the same people to go quiet and fade out of my life
i've learned a lot along the way, in ai and in life, and i've come to a point where i care for my peace first, before anything, and nobody who could take it away gets close to me
so fuck you to those pathetic pieces of shit, and thank you to the timeline
this is the new era and i'm here to win, i must not get distracted by the ups and downs, i forge through one day at a time
worst month of my life personally, and x made it financially easier thank you @elonmusk for backing independent builders 🙏
14
3
94
3,872
watch! i ran ling 3.1 flash on the same prompt twice with hermes agent on my floating tree spec
once as a q4 gguf across 4x dgx sparks and once on the api, with the same agent settings both times
> q4 over llama[.]cpp rpc: 12h 26m
> api, novita on openrouter: 57 min
> 7.5 vs 98.6 tok/tokens out
> 12/12 tests + visual gate on both
> $0.27 for the whole api run
both builds pass everything, the api build has the better tree with real branches and 8,270 veined leaves, the q4 build kept the better cliff
the video is one camera path on both builds, frame for frame, q4 on the left and the api on the right
my verdict:
ling 3.1 flash is a strong model and the speed gap is the setup, q4 over rpc proves it fits on 4 sparks, for agent work you want it on a real serving stack, and an official 4bit build would make it practical locally
the long turns question from the 12 hour build is answered in the thread below 👇
once as a q4 gguf across 4x dgx sparks and once on the api, with the same agent settings both times
> q4 over llama[.]cpp rpc: 12h 26m
> api, novita on openrouter: 57 min
> 7.5 vs 98.6 tok/tokens out
> 12/12 tests + visual gate on both
> $0.27 for the whole api run
both builds pass everything, the api build has the better tree with real branches and 8,270 veined leaves, the q4 build kept the better cliff
the video is one camera path on both builds, frame for frame, q4 on the left and the api on the right
my verdict:
ling 3.1 flash is a strong model and the speed gap is the setup, q4 over rpc proves it fits on 4 sparks, for agent work you want it on a real serving stack, and an official 4bit build would make it practical locally
the long turns question from the 12 hour build is answered in the thread below 👇
i've run antling's models since they started and new ling 3.1 flash is the first one that feels like a real leap from them
this is the floating tree it built on the api:
> 560b moe, 25b active
> 57 min on one prompt
> 77 requests, 228k tokens out
> 98.6 tok/s on openrouter (novita)
> 1green
> 2,051 lines across 9 modules
> $0.27 for the whole build
i can feel their models improve every release, ling 3.0 flash couldn't finish a 3d scene like this on a single dgx spark
sure 3.1 is a much bigger model, 560b total and 25b active against 3.0 flash's 124b and 5.1b, but the capability jumped with it
the scene has taste and every part is complete and works and it actually looks good
next i'm posting the q4 gguf i ran on @niklaslenz_ai's 4x dgx sparks and this api build side by side, so you can see the difference between the gguf and the api
stay tuned anon, ai is moving at this speed and i love it, the acceleration is insane in this space
5
2
11
1,959
the api run was pinned to novita on openrouter with no fallback and all 77 calls landed there, novita doesn't publish its precision so i call it the api, not full precision
the q4 side is atomicchat's ad-q4_k_m gguf over llama[.]cpp rpc, rpc splits the model by layers so the 4 ery token, that caps it near 12 tok/s and it ran at 7.5 tok/s
the q4 side is atomicchat's ad-q4_k_m gguf over llama[.]cpp rpc, rpc splits the model by layers so the 4 ery token, that caps it near 12 tok/s and it ran at 7.5 tok/s
2
0
1
539
> model: huggingface.co/inclusionAI/Li…
> api: openrouter.ai/inclusionai/li…
> q4 gguf: huggingface.co/AtomicChat/Lin…
> prompt: github.com/sudoingx/float…
> 4x dgx sparks: @niklaslenz_ai
model by @AntLingAGI
music: heroic intrusion by patrick de arteaga
> api: openrouter.ai/inclusionai/li…
> q4 gguf: huggingface.co/AtomicChat/Lin…
> prompt: github.com/sudoingx/float…
> 4x dgx sparks: @niklaslenz_ai
model by @AntLingAGI
music: heroic intrusion by patrick de arteaga
0
0
1
383
Sudo su retweeted
holy shit! typesafe is hacked and is not safe anymore
i opened the dump and every output they ever gave a customer is in there, 0.00 to 1.00, mine was 0.67 on line 69 and now the whole internet has it
a company called typesafe leaked its outputs as 101 floats and 0.30000000000000004 is missing from the dump, so the real leak is still out there
the entire typesafe leak is 507 tokens of counting 0.00 to 1.00, my 2x dgx sparks do counting at 147.6 tok/s, so i can regenerate their whole customer history locally in under 4 s and nobody leaks mine but me
i opened the dump and every output they ever gave a customer is in there, 0.00 to 1.00, mine was 0.67 on line 69 and now the whole internet has it
a company called typesafe leaked its outputs as 101 floats and 0.30000000000000004 is missing from the dump, so the real leak is still out there
the entire typesafe leak is 507 tokens of counting 0.00 to 1.00, my 2x dgx sparks do counting at 147.6 tok/s, so i can regenerate their whole customer history locally in under 4 s and nobody leaks mine but me
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/81003…
13
23
529
150,546
i've run antling's models since they started and new ling 3.1 flash is the first one that feels like a real leap from them
this is the floating tree it built on the api:
> 560b moe, 25b active
> 57 min on one prompt
> 77 requests, 228k tokens out
> 98.6 tok/s on openrouter (novita)
> 1green
> 2,051 lines across 9 modules
> $0.27 for the whole build
i can feel their models improve every release, ling 3.0 flash couldn't finish a 3d scene like this on a single dgx spark
sure 3.1 is a much bigger model, 560b total and 25b active against 3.0 flash's 124b and 5.1b, but the capability jumped with it
the scene has taste and every part is complete and works and it actually looks good
next i'm posting the q4 gguf i ran on @niklaslenz_ai's 4x dgx sparks and this api build side by side, so you can see the difference between the gguf and the api
stay tuned anon, ai is moving at this speed and i love it, the acceleration is insane in this space
this is the floating tree it built on the api:
> 560b moe, 25b active
> 57 min on one prompt
> 77 requests, 228k tokens out
> 98.6 tok/s on openrouter (novita)
> 1green
> 2,051 lines across 9 modules
> $0.27 for the whole build
i can feel their models improve every release, ling 3.0 flash couldn't finish a 3d scene like this on a single dgx spark
sure 3.1 is a much bigger model, 560b total and 25b active against 3.0 flash's 124b and 5.1b, but the capability jumped with it
the scene has taste and every part is complete and works and it actually looks good
next i'm posting the q4 gguf i ran on @niklaslenz_ai's 4x dgx sparks and this api build side by side, so you can see the difference between the gguf and the api
stay tuned anon, ai is moving at this speed and i love it, the acceleration is insane in this space
this was my first time running 4x dgx sparks in my entire life, and i'm more proud of this moment than whatever the fuck is going on in my life
the timeline has never disappointed me, and i'm dedicating myself to high quality content on everything from 128gb dgx sparks down to 3.1 flash, a huge model that doesn't fit on my 2 sparks, running across 4 while it builds the floating tree 3d scene, not perfect, but i ran their previous ling 3 on a single spark and i can see how fast ling is improving
in q4 gguf over llama[.]cpp rpc the speed is the pain, so next i run the full model on the api and drop the comparison soon
3
0
11
3,644
model by @AntLingAGI, the spec it built from is open for anyone to run: github.com/sudoingx/float…
0
0
1
611
holy shit! typesafe is hacked and is not safe anymore
i opened the dump and every output they ever gave a customer is in there, 0.00 to 1.00, mine was 0.67 on line 69 and now the whole internet has it
a company called typesafe leaked its outputs as 101 floats and 0.30000000000000004 is missing from the dump, so the real leak is still out there
the entire typesafe leak is 507 tokens of counting 0.00 to 1.00, my 2x dgx sparks do counting at 147.6 tok/s, so i can regenerate their whole customer history locally in under 4 s and nobody leaks mine but me
i opened the dump and every output they ever gave a customer is in there, 0.00 to 1.00, mine was 0.67 on line 69 and now the whole internet has it
a company called typesafe leaked its outputs as 101 floats and 0.30000000000000004 is missing from the dump, so the real leak is still out there
the entire typesafe leak is 507 tokens of counting 0.00 to 1.00, my 2x dgx sparks do counting at 147.6 tok/s, so i can regenerate their whole customer history locally in under 4 s and nobody leaks mine but me
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/81003…
13
23
529
150,546
this was my first time running 4x dgx sparks in my entire life, and i'm more proud of this moment than whatever the fuck is going on in my life
the timeline has never disappointed me, and i'm dedicating myself to high quality content on everything from 128gb dgx sparks down to 3.1 flash, a huge model that doesn't fit on my 2 sparks, running across 4 while it builds the floating tree 3d scene, not perfect, but i ran their previous ling 3 on a single spark and i can see how fast ling is improving
in q4 gguf over llama[.]cpp rpc the speed is the pain, so next i run the full model on the api and drop the comparison soon
the timeline has never disappointed me, and i'm dedicating myself to high quality content on everything from 128gb dgx sparks down to 3.1 flash, a huge model that doesn't fit on my 2 sparks, running across 4 while it builds the floating tree 3d scene, not perfect, but i ran their previous ling 3 on a single spark and i can see how fast ling is improving
in q4 gguf over llama[.]cpp rpc the speed is the pain, so next i run the full model on the api and drop the comparison soon
i gave ling 3.1 flash one prompt and never touched it again
the prompt was the floating tree, and the model ran on the 4x dgx sparks, 560b moe in q4 over llama[.]cpp rpc, with hermes agent driving
> 12 hours 26 min on one prompt
> 86 requests, 325k tokens out
> 7.5 tok/s decode, qpeak context 143k of 262k
> 12/12 tests pass, smoke green
> 2,166 lines across 12 modules
it built the island, the tree with a full crown, two houses, the pond and the dusk sky, and with no vision on this setup it checked its own renders with pixel stats and frame diffs, and it caught a flaky test and fixed it on its own
four of its turns took about 9 hours and 77% of all its tokens and wrote very little, one ran 2h52m and left a 10 line file
watch the whole build in 7 minutes and then what it made
q4 gguf over rpc is slow, so next the same prompt goes to the full model on the api, and that shows whether those long turns come from the model or from my setup
2
1
56
5,056
model by @AntLingAGI, and the q4 gguf i ran is AtomicChat's AD-Q4_K_M: huggingface.co/AtomicChat/Lin…
0
0
0
623
Sudo su retweeted
i gave ling 3.1 flash one prompt and never touched it again
the prompt was the floating tree, and the model ran on the 4x dgx sparks, 560b moe in q4 over llama[.]cpp rpc, with hermes agent driving
> 12 hours 26 min on one prompt
> 86 requests, 325k tokens out
> 7.5 tok/s decode, qpeak context 143k of 262k
> 12/12 tests pass, smoke green
> 2,166 lines across 12 modules
it built the island, the tree with a full crown, two houses, the pond and the dusk sky, and with no vision on this setup it checked its own renders with pixel stats and frame diffs, and it caught a flaky test and fixed it on its own
four of its turns took about 9 hours and 77% of all its tokens and wrote very little, one ran 2h52m and left a 10 line file
watch the whole build in 7 minutes and then what it made
q4 gguf over rpc is slow, so next the same prompt goes to the full model on the api, and that shows whether those long turns come from the model or from my setup
the prompt was the floating tree, and the model ran on the 4x dgx sparks, 560b moe in q4 over llama[.]cpp rpc, with hermes agent driving
> 12 hours 26 min on one prompt
> 86 requests, 325k tokens out
> 7.5 tok/s decode, qpeak context 143k of 262k
> 12/12 tests pass, smoke green
> 2,166 lines across 12 modules
it built the island, the tree with a full crown, two houses, the pond and the dusk sky, and with no vision on this setup it checked its own renders with pixel stats and frame diffs, and it caught a flaky test and fixed it on its own
four of its turns took about 9 hours and 77% of all its tokens and wrote very little, one ran 2h52m and left a 10 line file
watch the whole build in 7 minutes and then what it made
q4 gguf over rpc is slow, so next the same prompt goes to the full model on the api, and that shows whether those long turns come from the model or from my setup
i am running a monster model on 4x dgx sparks right now
the model is ling 3.1 flash, i tested the older ling versions and enjoyed them, and i wanted to try this big one but it doesn't fit on my 2x dgx sparks, so a friend here on x, @niklaslenz_ai from germany, let me borrow his 'm set up and about to run my tests
there's no official fp4 build for vllm yet, so i'm on atomic chat's day one q4 gguf with llama.cpp rpc across all 4x dgx sparks
i know bandwidth and speed will torture me, but i couldn't stay away from loading this one because i want to know where it lands next to the 124b ling 3.0 flash
stay tuned, i'll drop numbers as we go, and if you've already run it, share your thoughts
10
3
22
5,890
i gave ling 3.1 flash one prompt and never touched it again
the prompt was the floating tree, and the model ran on the 4x dgx sparks, 560b moe in q4 over llama[.]cpp rpc, with hermes agent driving
> 12 hours 26 min on one prompt
> 86 requests, 325k tokens out
> 7.5 tok/s decode, qpeak context 143k of 262k
> 12/12 tests pass, smoke green
> 2,166 lines across 12 modules
it built the island, the tree with a full crown, two houses, the pond and the dusk sky, and with no vision on this setup it checked its own renders with pixel stats and frame diffs, and it caught a flaky test and fixed it on its own
four of its turns took about 9 hours and 77% of all its tokens and wrote very little, one ran 2h52m and left a 10 line file
watch the whole build in 7 minutes and then what it made
q4 gguf over rpc is slow, so next the same prompt goes to the full model on the api, and that shows whether those long turns come from the model or from my setup
the prompt was the floating tree, and the model ran on the 4x dgx sparks, 560b moe in q4 over llama[.]cpp rpc, with hermes agent driving
> 12 hours 26 min on one prompt
> 86 requests, 325k tokens out
> 7.5 tok/s decode, qpeak context 143k of 262k
> 12/12 tests pass, smoke green
> 2,166 lines across 12 modules
it built the island, the tree with a full crown, two houses, the pond and the dusk sky, and with no vision on this setup it checked its own renders with pixel stats and frame diffs, and it caught a flaky test and fixed it on its own
four of its turns took about 9 hours and 77% of all its tokens and wrote very little, one ran 2h52m and left a 10 line file
watch the whole build in 7 minutes and then what it made
q4 gguf over rpc is slow, so next the same prompt goes to the full model on the api, and that shows whether those long turns come from the model or from my setup
i am running a monster model on 4x dgx sparks right now
the model is ling 3.1 flash, i tested the older ling versions and enjoyed them, and i wanted to try this big one but it doesn't fit on my 2x dgx sparks, so a friend here on x, @niklaslenz_ai from germany, let me borrow his 'm set up and about to run my tests
there's no official fp4 build for vllm yet, so i'm on atomic chat's day one q4 gguf with llama.cpp rpc across all 4x dgx sparks
i know bandwidth and speed will torture me, but i couldn't stay away from loading this one because i want to know where it lands next to the 124b ling 3.0 flash
stay tuned, i'll drop numbers as we go, and if you've already run it, share your thoughts
10
3
22
5,890
Sudo su retweeted
i'm falling for my own agent harness
xpress agent is the harness i built for myself, it's rust and i can feel it in the speed, and offrail 27b is the model behind it, running on my laptop right now
it pulls images off the web and lay them out right under its answer, i slid the one i wanted, no more clicking through folders to find what it downloaded, and this one feature saved me so much time
the model does the thinking, the harness decides what it sees and what it remembers, and you spend your whole day inside the harness, so it should fit the way you work
> one binary under 10 mb
> 14 mb of ram for the whole loop
> finds my local model by itself
> self learning loop
it writes its own skills as it works, and on that same run it rewrote its image skill without me asking
i think eventually you all will have to build your own agent harness the way you like it and the future of agents is rust
xpress agent is the harness i built for myself, it's rust and i can feel it in the speed, and offrail 27b is the model behind it, running on my laptop right now
it pulls images off the web and lay them out right under its answer, i slid the one i wanted, no more clicking through folders to find what it downloaded, and this one feature saved me so much time
the model does the thinking, the harness decides what it sees and what it remembers, and you spend your whole day inside the harness, so it should fit the way you work
> one binary under 10 mb
> 14 mb of ram for the whole loop
> finds my local model by itself
> self learning loop
it writes its own skills as it works, and on that same run it rewrote its image skill without me asking
i think eventually you all will have to build your own agent harness the way you like it and the future of agents is rust
yesterday this was a tweet asking a grokbot to build it.
since last night it's the only agent i open xpress agent v0.1.0.
local first, learns as it works, written in rust grokbot prototyped it, i took it into cursor and engineered every inch of it, and it's already the best ever used.
built by me, for me the moment you launch it, it finds the model server on localhost by itself and configures itself. no config file, no setup.
my 27b is just there, 128k context, ready to work /model to switch between local and cloud. free openrouter models chained as fallbacks so it never dies mid task. permissions ask by default, shift+tab to change. /resume any past chat. @ any file.
web search built in and the part i love most: it learns as it works. hermes agent's self-improving loop is the best idea in agents right now, so i rebuilt it in rust my way one binary. no python waking up.
it loads faster and answers faster than hermes agent and openclaw on the same model. and my api keys never leave my machine this era is insane.
if you're still waiting for someone to ship the tool you want, you're already behind.
build it yourself with @bot, in a day
20
9
141
8,452