RDTvlokip/InfiniQA
Viewer • Updated • 101k • 140 • 3
RDTvlokip PRO
RDTvlokip
AI & ML interests
I'm builds complete AI systems from the ground up, tokenizers, search engines, and small-scale language models to investigate, through exact mathematics and public experimentation, how language and intelligence emerge from minimal resources and pure signal.
Recent Activity
repliedto their post 2 days ago
I published an article about training a network to write from reward alone. Code on GitHub with it.
Then someone read the code.
Dipankar Sarkar commented four times in a day. Each time he had run something first. He rebuilt my statistics in numpy because he had no torch installed.
He found a bound I had missed. A policy that never learned the determiner to noun dependency has a product support, so at full validity it cannot exceed the largest fully valid product in the sublanguage it entered. That is 12 on one side and 24 on the other, computable before any training. Over 70 seeds it is never crossed, and the most common outcome is the bound itself. I had published one of those numbers as an interesting coincidence.
Then four of my published numbers came apart. Three were a single seed. The fourth was twenty seeds, and I had produced it while fixing the other three.
And the test I built to validate his bound tested nothing. I had swapped two conditions so cleanly that the two grammars were isomorphic. Seventy seeds would have returned the mirror image by construction.
The real lesson:
A relabelling can permute, but it cannot change a ratio. A perfectly symmetric control is often a perfectly empty one. None of my errors were in the reasoning. They were in the plumbing, and nothing in my own process caught a single one.
Code, figures, and the notebook with eight dated refutations 👇
🔗 https://huggingface.co/blog/RDTvlokip/i-published-my-rl-experiments
💻 https://github.com/RDTvlokip/RDTRL
📦 https://doi.org/10.5281/zenodo.21726216 repliedto their post 3 days ago
I published an article about training a network to write from reward alone. Code on GitHub with it.
Then someone read the code.
Dipankar Sarkar commented four times in a day. Each time he had run something first. He rebuilt my statistics in numpy because he had no torch installed.
He found a bound I had missed. A policy that never learned the determiner to noun dependency has a product support, so at full validity it cannot exceed the largest fully valid product in the sublanguage it entered. That is 12 on one side and 24 on the other, computable before any training. Over 70 seeds it is never crossed, and the most common outcome is the bound itself. I had published one of those numbers as an interesting coincidence.
Then four of my published numbers came apart. Three were a single seed. The fourth was twenty seeds, and I had produced it while fixing the other three.
And the test I built to validate his bound tested nothing. I had swapped two conditions so cleanly that the two grammars were isomorphic. Seventy seeds would have returned the mirror image by construction.
The real lesson:
A relabelling can permute, but it cannot change a ratio. A perfectly symmetric control is often a perfectly empty one. None of my errors were in the reasoning. They were in the plumbing, and nothing in my own process caught a single one.
Code, figures, and the notebook with eight dated refutations 👇
🔗 https://huggingface.co/blog/RDTvlokip/i-published-my-rl-experiments
💻 https://github.com/RDTvlokip/RDTRL
📦 https://doi.org/10.5281/zenodo.21726216 posted an update 3 days ago
My PSU fan started making noise. Turns out that's not nothing either.
Same rig, different failure mode from the GPU freeze post. This one didn't kill a training run — it just wouldn't go away.
The pattern: silent at cold boot, "grr + zzz" cyclical noise kicking in after ~2 minutes of runtime, tracking the fan's own accelerate/decelerate cycle. Ear-to-case testing ruled out the HDD and the GTX 1080 Ti. It's the PSU fan. No crackling, no smell, no smoke — so no panic, but no ignoring it either.
The PSU documents its own fan curve: idle speed until 50% load at 25°C ambient. My CPU alone never gets close to that threshold, which is exactly why the noise never showed up under light tasks — only once thermal buildup pushed the fan into the RPM band where a worn bearing becomes audible. Classic cold-silent, warm-loud signature of bearing wear, not an electrical fault.
I'm not opening the unit. The PSU's own safety documentation says never to, even unplugged — internal capacitors can hold a dangerous charge well after disconnection. Some repairs aren't worth the corner-cut.
Plan: keep GPU load light for now, full replacement within 1-2 months before wear turns into failure. The old unit doesn't go in the bin — it sits next to the 1080 Ti. Both trophies from the same stretch of work.
Same discipline as the GPU wall: hear the symptom, isolate the cause, don't assume it's nothing just because nothing crashed yet.
Full write-up 👇
🔗 https://huggingface.co/blog/RDTvlokip/the-other-kind-of-hardware-wall-when-its-not-the-gOrganizations
