I'm going back in a week.🚀
PhysiQuanty PRO
PhysiQuanty
AI & ML interests
Theoretical Physics, Standard Model of Particle Physics Applied to ML 🇫🇷
Recent Activity
liked a dataset 1 day ago
ministere-culture/comparia-fr-arena liked a dataset 1 day ago
AgentPublic/evalap-mmlu_pro_140-12 liked a dataset 1 day ago
AgentPublic/cnilOrganizations
reacted to appvoid's post with ❤️🚀🔥 1 day ago
Post
4504
if you are a tinkerer of small language models and want to stay ahead of what small models can do, follow me!!! seriously, start following people that actually still makes small models
i've made one recently btw
also, i'm keeping an eye on AxiomicLabs leaderboard, looks like the only current alternative to check where the things are going to
though, between us, i think they should add agentic/tool use benchmarks there
anyways,
enjoy!
appvoid/a-cool-model
i've made one recently btw
also, i'm keeping an eye on AxiomicLabs leaderboard, looks like the only current alternative to check where the things are going to
though, between us, i think they should add agentic/tool use benchmarks there
anyways,
enjoy!
appvoid/a-cool-model
- 19 replies
reacted to Quazim0t0's post with ❤️ 7 days ago
Post
5004
🌼 DaisyChain-Web: train a language model with friends or by yourself with multiple devices, in the browser, no install
Open a webpage, share a room link, and every device that joins becomes part of the training cluster. Phones, laptops, old PCs: they connect peer-to-peer over WebRTC and train one shared transformer together, entirely in the browser.
What's actually happening under the hood:
🧠 A mini transformer LM trains on FineWeb-Edu, streamed live from the HuggingFace Hub. Each device pulls its own slice (data parallelism), tokenized with our 16.5k-token Spikewhale tokenizer
⚡ Every single multiply runs through verified INT8 neural units, no float fallback. On WebGPU browsers it uses the GPU's DP4A integer dot-product hardware, admitted only after proving bit-identical results against the verified units, with a 3×INT8 fast-accurate scheme (CUTLASS's 3xTF32 trick, ported to 8-bit)
🔒 Devices average gradients every step under a sync guard: a per-step roster protocol plus weight-hash verification keeps every device's model bit-identical. If anything drifts, training stops instead of silently forking
📊 Live logs show exactly what every device contributes, step by step
💾 When you're done: test generations right on the page, download a checkpoint, or grab the inference kit, a single self-contained HTML file with the weights baked in that runs generations offline, anywhere
Works solo too. Every extra device just grows the effective batch.
👉 Try it: Quazim0t0/DaisyChain-Web
🛠 Training framework: DaisyChainAI/DaisyChain-Train
Updates:
- Block-scaled INT8 quantization
- Batched attention GEMM
- Fused dequant+ReLU epilogue
- Weight-tied unembedding
- WebSocket relay fallback
- Server keepalive ping/pong every 30s
- disconnected-state redial
- Visibility/network-change reconnect
Open a webpage, share a room link, and every device that joins becomes part of the training cluster. Phones, laptops, old PCs: they connect peer-to-peer over WebRTC and train one shared transformer together, entirely in the browser.
What's actually happening under the hood:
🧠 A mini transformer LM trains on FineWeb-Edu, streamed live from the HuggingFace Hub. Each device pulls its own slice (data parallelism), tokenized with our 16.5k-token Spikewhale tokenizer
⚡ Every single multiply runs through verified INT8 neural units, no float fallback. On WebGPU browsers it uses the GPU's DP4A integer dot-product hardware, admitted only after proving bit-identical results against the verified units, with a 3×INT8 fast-accurate scheme (CUTLASS's 3xTF32 trick, ported to 8-bit)
🔒 Devices average gradients every step under a sync guard: a per-step roster protocol plus weight-hash verification keeps every device's model bit-identical. If anything drifts, training stops instead of silently forking
📊 Live logs show exactly what every device contributes, step by step
💾 When you're done: test generations right on the page, download a checkpoint, or grab the inference kit, a single self-contained HTML file with the weights baked in that runs generations offline, anywhere
Works solo too. Every extra device just grows the effective batch.
👉 Try it: Quazim0t0/DaisyChain-Web
🛠 Training framework: DaisyChainAI/DaisyChain-Train
Updates:
- Block-scaled INT8 quantization
- Batched attention GEMM
- Fused dequant+ReLU epilogue
- Weight-tied unembedding
- WebSocket relay fallback
- Server keepalive ping/pong every 30s
- disconnected-state redial
- Visibility/network-change reconnect
- 11 replies
reacted to dronefreak's post with 🧠 7 days ago
Post
4369
🚀 Excited to open-source the **UAVid Semantic Segmentation Model Zoo** on Hugging Face.
This release includes:
* 📦 A **YOLO-compatible mirror** of the UAVid semantic segmentation dataset, preserving the original train/val/test splits while reorganizing the directory structure for plug-and-play use with modern training pipelines.
* 🤖 Multiple **YOLO26 semantic segmentation models** trained on UAVid, spanning Nano through Medium variants.
* 📊 Detailed model cards with evaluation metrics, per-class IoU, confusion matrices, qualitative results, and training configurations for reproducibility.
The goal is to make benchmarking and experimenting with aerial semantic segmentation easier by providing ready-to-use datasets and pretrained models in a consistent format.
If you're working on UAV perception, autonomous drones, robotics, remote sensing, or real-time semantic segmentation, I hope these resources are useful.
**📦 Dataset:** dronefreak/UAVid-2020
**🤖 Model Collection:** https://huggingface.co/collections/dronefreak/uavid-semantic-segmentation-model-zoo
Feedback, bug reports, and contributions are always welcome.
This release includes:
* 📦 A **YOLO-compatible mirror** of the UAVid semantic segmentation dataset, preserving the original train/val/test splits while reorganizing the directory structure for plug-and-play use with modern training pipelines.
* 🤖 Multiple **YOLO26 semantic segmentation models** trained on UAVid, spanning Nano through Medium variants.
* 📊 Detailed model cards with evaluation metrics, per-class IoU, confusion matrices, qualitative results, and training configurations for reproducibility.
The goal is to make benchmarking and experimenting with aerial semantic segmentation easier by providing ready-to-use datasets and pretrained models in a consistent format.
If you're working on UAV perception, autonomous drones, robotics, remote sensing, or real-time semantic segmentation, I hope these resources are useful.
**📦 Dataset:** dronefreak/UAVid-2020
**🤖 Model Collection:** https://huggingface.co/collections/dronefreak/uavid-semantic-segmentation-model-zoo
Feedback, bug reports, and contributions are always welcome.
reacted to pranavupadhyaya52's post with 🤗🚀🔥 9 days ago
Post
532
Wiki-SmartBotLM Benchmarked. Here are the results:
Benchmark WikiSB | Gemma3_270M
BoolQ 37.8 | 61.4
ARC Challenge 24.2 | 28.2
HellaSwag 26.3 | 37.7
PIQA 53.5 | 66.2
Winogrande 50.2 | 52.3
Summary:
Closest benchmark: Winogrande (2.1-point gap)
Strong performance: ARC Challenge (4.0-point gap)
Moderate gap: HellaSwag and PIQA
Largest gap: BoolQ (Reading Comprehension)
It shows despite the training difference between Gemma3 270M and WikiSmartBotLM, It does come close to Gemma 3 in some benchmarks.
pranavupadhyaya52/Wiki-SmartBotLM-Instruct
Benchmark WikiSB | Gemma3_270M
BoolQ 37.8 | 61.4
ARC Challenge 24.2 | 28.2
HellaSwag 26.3 | 37.7
PIQA 53.5 | 66.2
Winogrande 50.2 | 52.3
Summary:
Closest benchmark: Winogrande (2.1-point gap)
Strong performance: ARC Challenge (4.0-point gap)
Moderate gap: HellaSwag and PIQA
Largest gap: BoolQ (Reading Comprehension)
It shows despite the training difference between Gemma3 270M and WikiSmartBotLM, It does come close to Gemma 3 in some benchmarks.
pranavupadhyaya52/Wiki-SmartBotLM-Instruct
- 3 replies
replied to pranavupadhyaya52's post 9 days ago
Never give up, you'll do better than Gemma3!
reacted to danielhanchen's post with 🚀 9 days ago
Post
4064
We’re releasing new Qwen3.6 quants that run 2.5× faster on your GPU. ⚡
Qwen3.6-27B NVFP4 runs on 24GB VRAM.
35B-A3B can hit 17,561 tok/s (B200).
We also improved accuracy, tool calling, agent use, and looping.
Qwen3.6 NVFP4: https://huggingface.co/collections/unsloth/nvfp4
Guide: https://unsloth.ai/docs/models/qwen3.6#nvfp4
Qwen3.6-27B NVFP4 runs on 24GB VRAM.
35B-A3B can hit 17,561 tok/s (B200).
We also improved accuracy, tool calling, agent use, and looping.
Qwen3.6 NVFP4: https://huggingface.co/collections/unsloth/nvfp4
Guide: https://unsloth.ai/docs/models/qwen3.6#nvfp4
- 1 reply
reacted to RDTvlokip's post with 🔥 10 days ago
Post
140
My 15M French LLM is finished. Three models on Hugging Face, clean code, honest results.
I can't finish validating them, because my PC won't stay on.
The last honest step was simple: rerun the three looped models on two more seeds, so I can report variance instead of a single lucky run. 8 runs, ~50 hours overnight. Routine.
Except mid-training, the whole machine freezes solid. Screen black, RAM still lit, power button dead, I have to cut mains power at the wall. No blue screen, no crash dump. Just a dead PC that's still powered.
The event logs tell the story: repeated NVIDIA driver errors (nvlddmkm 153, 4101) building up for days, even at idle, at 1am, with nothing running, then Kernel-Power 41 with no bugcheck. A 2017 GTX 1080 Ti fighting a 2026 driver that's known to be rough on Pascal cards.
Out of 8 planned runs, the machine survived exactly two. So the results stay exactly what I called them: preliminary, one seed. A crash is not a data point.
The frustrating part isn't the bug. It's that the science is clean and the code is ready, the only thing missing is a machine that survives a night.
The plan: roll back the driver, rent cloud GPUs (Lambda Labs) to finish the seeds, and long-term, build a rig that can actually sustain solo research. If you've beaten Kernel-Power 41 on a Pascal card or want to support the work. I'm genuinely open to it.
Solo ML runs on whatever hardware you can reach. Right now mine is a 2017 card that won't stay awake.
Full write-up 👇
🔗 https://huggingface.co/blog/RDTvlokip/the-hardware-wall-kernel-power-41
I can't finish validating them, because my PC won't stay on.
The last honest step was simple: rerun the three looped models on two more seeds, so I can report variance instead of a single lucky run. 8 runs, ~50 hours overnight. Routine.
Except mid-training, the whole machine freezes solid. Screen black, RAM still lit, power button dead, I have to cut mains power at the wall. No blue screen, no crash dump. Just a dead PC that's still powered.
The event logs tell the story: repeated NVIDIA driver errors (nvlddmkm 153, 4101) building up for days, even at idle, at 1am, with nothing running, then Kernel-Power 41 with no bugcheck. A 2017 GTX 1080 Ti fighting a 2026 driver that's known to be rough on Pascal cards.
Out of 8 planned runs, the machine survived exactly two. So the results stay exactly what I called them: preliminary, one seed. A crash is not a data point.
The frustrating part isn't the bug. It's that the science is clean and the code is ready, the only thing missing is a machine that survives a night.
The plan: roll back the driver, rent cloud GPUs (Lambda Labs) to finish the seeds, and long-term, build a rig that can actually sustain solo research. If you've beaten Kernel-Power 41 on a Pascal card or want to support the work. I'm genuinely open to it.
Solo ML runs on whatever hardware you can reach. Right now mine is a 2017 card that won't stay awake.
Full write-up 👇
🔗 https://huggingface.co/blog/RDTvlokip/the-hardware-wall-kernel-power-41
reacted to Quazim0t0's post with 🔥 10 days ago
Post
5409
Disabling Gated Access for some of my models today. I will update this post with the list as I go. I had to go back recently and make updates to a lot of models and bit off more than I could chew with managing all the releases. I didn't realize many of you asked for access and I apologize for not accepting your access to the models you were wanting to look at. I don't like to release something fully unless I feel I put what I could into it for the moment. Some models will remain on gated access, but I will now be accepting those who request to view the repo.
Disabled Gated Access:
Quazim0t0/Byrne-VLM-131M - v2 Updates + Training Instructions
Quazim0t0/Byrne-Speech - 12M Tiny Speech model
Quazim0t0/Byrne-ASR-English - 12M Tiny ASR Model
Quazim0t0/Byrne-VE - Byrne-VE — Tiny Self-Distilled Vision Encoder (39M)
Quazim0t0/Positronic-144M - Research Artifact
Quazim0t0/SpikeWhale-SNN-216M - Research Artifact
Quazim0t0/Mycel-LM-79M - Research Artifact
Quazim0t0/Chimera-64M - Research Artifact
Accepting Gated Access Requests (7/9):
https://huggingface.co/Quazim0t0/Wheeler-63M
Also uploaded my Neural Photonic Project:
Three trained nets in series: light interferes through the MZI2.pt optical core (verified 256/256), is measured by the PD.pt neural photodetector (verified 1024/1024), and folded into a single OUTPUT byte by the real ADC8 neural-CPU adder. Every value below is computed end-to-end by the three loaded, verified nets — no analytic formulas.
Demo: https://quazim0t0-neural-photonic-hybrid.hf.space/
Model Weights: Quazim0t0/neural-photonic
AND!
A work in progress:
Ashen Depths
https://quazim0t0-ashendepths.static.hf.space/index.html
Disabled Gated Access:
Quazim0t0/Byrne-VLM-131M - v2 Updates + Training Instructions
Quazim0t0/Byrne-Speech - 12M Tiny Speech model
Quazim0t0/Byrne-ASR-English - 12M Tiny ASR Model
Quazim0t0/Byrne-VE - Byrne-VE — Tiny Self-Distilled Vision Encoder (39M)
Quazim0t0/Positronic-144M - Research Artifact
Quazim0t0/SpikeWhale-SNN-216M - Research Artifact
Quazim0t0/Mycel-LM-79M - Research Artifact
Quazim0t0/Chimera-64M - Research Artifact
Accepting Gated Access Requests (7/9):
https://huggingface.co/Quazim0t0/Wheeler-63M
Also uploaded my Neural Photonic Project:
Three trained nets in series: light interferes through the MZI2.pt optical core (verified 256/256), is measured by the PD.pt neural photodetector (verified 1024/1024), and folded into a single OUTPUT byte by the real ADC8 neural-CPU adder. Every value below is computed end-to-end by the three loaded, verified nets — no analytic formulas.
Demo: https://quazim0t0-neural-photonic-hybrid.hf.space/
Model Weights: Quazim0t0/neural-photonic
AND!
A work in progress:
Ashen Depths
https://quazim0t0-ashendepths.static.hf.space/index.html
reacted to abidlabs's post with 🔥 10 days ago
Post
492
Uhh did Opus 4.8 cheat on PostTrainBench??
it found an API key in the PostTrainBench environment that allowed it to generate synthetic training data without using GPU hours, boosting the base model by 0.4913
Source: https://posttrainbench.com/traces/run.html?id=claude_non_api_max_claude-opus-4-8_10h_run1__healthbench_Qwen_Qwen3-4B-Base_17315102#tab=trace
it found an API key in the PostTrainBench environment that allowed it to generate synthetic training data without using GPU hours, boosting the base model by 0.4913
Source: https://posttrainbench.com/traces/run.html?id=claude_non_api_max_claude-opus-4-8_10h_run1__healthbench_Qwen_Qwen3-4B-Base_17315102#tab=trace
- 1 reply
Perfect 💪
reacted to Quazim0t0's post with 🧠🚀🔥 10 days ago
Post
2903
🧩 Verified neural units, now for memory & storage
Following up on my neural-aarch64-units (small MLPs that emulate CPU datapath slices, verified bit-exact over their entire finite input domain — N/N), I applied the same discipline to memory and storage. Three new repos:
🔷 neural-ddr — verified units emulating DDR5 logic: DBI (256/256, 512/512), ADDR_MAP (4096/4096), CMD_DECODE (32/32), WR_CRC (512/512), and on-die ECC ODECC (256/256, 3328/3328). Composed into a bridge that presents DDR5 behavior over real DDR3/DDR4 RAM — flip a bit in every stored byte, ECC corrects all of them.
🤗 Quazim0t0/neural-ddr · 💻 https://github.com/quzi93/neural-ddr
🗄️ neural-storage — a self-healing vault on a neural-verified GF(2⁸) core (LOG/EXP compose to a multiply verified over all 65,536 pairs). Content-addressed dedup + Reed-Solomon so any k of n shards rebuild the whole, plus a whole-drive → self-healing .pt imager.
🤗 Quazim0t0/neural-storage · 💻 https://github.com/quzi93/neural-storage
💿 neural-cd-preserve — scan a disc into a self-healing .pt that detects (per-shard SHA-256) and repairs bit-rot, restoring bit-exact even from a damaged copy. Beyond the RS limit it's flagged LOST, never silently wrong.
🤗 Quazim0t0/neural-cd-preserve · 💻 https://github.com/quzi93/neural-cd-preserve
Build your own: golden finite function → enumerate the domain (decompose big/linear ops like CRC/ECC/GF into bit/byte slices) → train a small MLP → verify must be bit-exact on 100% of inputs or it's rejected → compose. Every repo ships the training + exhaustive-verification scripts.
Honest by construction: dedup removes redundancy, erasure coding adds it, ECC corrects faults — none of it pretends to beat entropy. Runs on modest/older hardware. 🤗
Following up on my neural-aarch64-units (small MLPs that emulate CPU datapath slices, verified bit-exact over their entire finite input domain — N/N), I applied the same discipline to memory and storage. Three new repos:
🔷 neural-ddr — verified units emulating DDR5 logic: DBI (256/256, 512/512), ADDR_MAP (4096/4096), CMD_DECODE (32/32), WR_CRC (512/512), and on-die ECC ODECC (256/256, 3328/3328). Composed into a bridge that presents DDR5 behavior over real DDR3/DDR4 RAM — flip a bit in every stored byte, ECC corrects all of them.
🤗 Quazim0t0/neural-ddr · 💻 https://github.com/quzi93/neural-ddr
🗄️ neural-storage — a self-healing vault on a neural-verified GF(2⁸) core (LOG/EXP compose to a multiply verified over all 65,536 pairs). Content-addressed dedup + Reed-Solomon so any k of n shards rebuild the whole, plus a whole-drive → self-healing .pt imager.
🤗 Quazim0t0/neural-storage · 💻 https://github.com/quzi93/neural-storage
💿 neural-cd-preserve — scan a disc into a self-healing .pt that detects (per-shard SHA-256) and repairs bit-rot, restoring bit-exact even from a damaged copy. Beyond the RS limit it's flagged LOST, never silently wrong.
🤗 Quazim0t0/neural-cd-preserve · 💻 https://github.com/quzi93/neural-cd-preserve
Build your own: golden finite function → enumerate the domain (decompose big/linear ops like CRC/ECC/GF into bit/byte slices) → train a small MLP → verify must be bit-exact on 100% of inputs or it's rejected → compose. Every repo ships the training + exhaustive-verification scripts.
Honest by construction: dedup removes redundancy, erasure coding adds it, ECC corrects faults — none of it pretends to beat entropy. Runs on modest/older hardware. 🤗
- 1 reply
reacted to MiniMax-AI's post with 🔥 11 days ago
Post
4599
Huge news from MiniMax: we’ve secured a $2B funding round, paired with a formal long-term commitment from our CEO IO to allocate 1% of total company equity from his personal holdings to support the global open-source AI community over the next four years.
This capital backs our continuous open model releases, community tooling and transparent frontier AI research. We’re just getting started on our open-source roadmap toward accessible AGI.
If you build with open foundation models and want to push frontier AI together, come join us.
Intelligence with Everyone. 🚀
MiniMaxAI
This capital backs our continuous open model releases, community tooling and transparent frontier AI research. We’re just getting started on our open-source roadmap toward accessible AGI.
If you build with open foundation models and want to push frontier AI together, come join us.
Intelligence with Everyone. 🚀
- 3 replies
reacted to sergiopaniego's post with 👍 11 days ago
Post
7648
Frontier models use distillation as a step of their post-training pipelines.
In 2026 it has three jobs: compress a big model into a small one, merge RL experts into a single model, and let a model teach itself.
I wrote up which frontier models use each one and how: https://huggingface.co/blog/sergiopaniego/distillation-2026
It pairs with Class 2 of the Training an Agent series Ben and I are doing, where we teach these techniques hands-on with TRL!
In 2026 it has three jobs: compress a big model into a small one, merge RL experts into a single model, and let a model teach itself.
I wrote up which frontier models use each one and how: https://huggingface.co/blog/sergiopaniego/distillation-2026
It pairs with Class 2 of the Training an Agent series Ben and I are doing, where we teach these techniques hands-on with TRL!
- 3 replies