ReadyArt/Wicked-Tutu-31B-v0.2-GGUF
Text Generation • 31B • Updated • 3
jaithree PRO
jaithree
AI & ML interests
Language, lingo, jargon, parlance...
Recent Activity
liked a model about 2 hours ago
ReadyArt/Wicked-Tutu-31B-v0.2-GGUF liked a model 1 day ago
ReadyArt/Wicked-Tutu-31B-v0.1Organizations
liked a
model about 2 hours ago
liked a
model 1 day ago
reacted to branikita's post with 🔥 1 day ago
Post
4,238
A step-by-step guide to assembling the SO-ARM101 Follower Arm, from parts to a pick-up test.
https://youtu.be/EhWyX_KcCuk
https://youtu.be/EhWyX_KcCuk
reacted to SeaWolf-AI's post with 🔥 2 days ago
Post
2,733
🚀 We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
🤗 Model: FINAL-Bench/VKORE-46M-NPU 📝 Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE · SwiGLU · RMSNorm · GQA), char-level, TinyStories.
📏 Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. · Loss trajectories correlate at 0.99 · Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
🧩 One accelerator now covers pretraining · LoRA · inference · quantization. We call this layer VKORE: "CUDA for NPUs."
⚖️ Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. 🙌
🤗 Model: FINAL-Bench/VKORE-46M-NPU 📝 Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE · SwiGLU · RMSNorm · GQA), char-level, TinyStories.
📏 Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. · Loss trajectories correlate at 0.99 · Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
🧩 One accelerator now covers pretraining · LoRA · inference · quantization. We call this layer VKORE: "CUDA for NPUs."
⚖️ Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. 🙌
- 10 replies
reacted to yanghaojin's post with 🔥 2 days ago
Post
2,962
Dear all,
we are happy to share that gbx v0.8.0, our local inference engine for Apple Silicon, runs the 105 GB Qwen3.8-Flash-Next on a 48 GB MacBook Pro by streaming MoE experts from the SSD. When the model fits in memory, it decodes up to 50% faster than omlx, Ollama and mlx-lm on the same Mac, with output identical token for token.
Full write-up: https://medium.com/@semamediadata/running-a-105-gb-qwen-on-a-48-gb-macbook-pro-f116fe493a49
Download gbx: https://github.com/GreenBitAI/gbx-lm/releases
Models: https://huggingface.co/collections/GreenBitAI/greenbitai-for-apple-silicon
we are happy to share that gbx v0.8.0, our local inference engine for Apple Silicon, runs the 105 GB Qwen3.8-Flash-Next on a 48 GB MacBook Pro by streaming MoE experts from the SSD. When the model fits in memory, it decodes up to 50% faster than omlx, Ollama and mlx-lm on the same Mac, with output identical token for token.
Full write-up: https://medium.com/@semamediadata/running-a-105-gb-qwen-on-a-48-gb-macbook-pro-f116fe493a49
Download gbx: https://github.com/GreenBitAI/gbx-lm/releases
Models: https://huggingface.co/collections/GreenBitAI/greenbitai-for-apple-silicon
- 2 replies
liked a
model 4 days ago
Permission to use Mag Mell in a free game (donations + Patreon early access)
❤️ 1
8
#21 opened 6 days ago
by 
grognar
upvoted a changelog 5 days ago
Hugging Face Changelog
Contributor permission for access tokens
• 86
liked a
model 8 days ago
reacted to tardellirs's post with 🔥 10 days ago
Post
3,921
Robotics is now the second most downloaded dataset category on the Hub, after text generation.
Robotics datasets got 13.7M downloads in September, ahead of text classification and question answering. Two years ago the category ranked 23rd. One in 7 new datasets is now robotics, mostly LeRobot recordings: typically a few dozen demos, about half of them on low-cost SO-100/SO-101 arms.
I found this after adding datasets and Spaces to Model Pulse, which rebuilds daily history from @cfahlgren1 's hub-stats snapshots. Two more findings:
- In 2022, 36% of authors who list training data cited classic NLP sets like IMDb, SQuAD and GLUE. In 2026 it's 2.4%. Reasoning traces distilled from models like DeepSeek-R1 and Claude are now the most cited kind.
- In October 2025, 122K Spaces were created, 71K of them websites built with DeepSite. That's about 6x the monthly pace of late 2024, while likes given per month fell from about 35K to about 20K.
New in the app: a page for every dataset, with daily downloads and the models trained on it (636 list FineWeb), a page for every Space, and rankings for both.
Thank you to everyone who liked Model Pulse this week: it made Spaces of the Week and is #7 on trending. Thanks also to @dipankarsarkar , whose comments on the last post fixed three data issues. If a number looks wrong, tell me.
Spaces: tardellirs/model-pulse
Robotics datasets got 13.7M downloads in September, ahead of text classification and question answering. Two years ago the category ranked 23rd. One in 7 new datasets is now robotics, mostly LeRobot recordings: typically a few dozen demos, about half of them on low-cost SO-100/SO-101 arms.
I found this after adding datasets and Spaces to Model Pulse, which rebuilds daily history from @cfahlgren1 's hub-stats snapshots. Two more findings:
- In 2022, 36% of authors who list training data cited classic NLP sets like IMDb, SQuAD and GLUE. In 2026 it's 2.4%. Reasoning traces distilled from models like DeepSeek-R1 and Claude are now the most cited kind.
- In October 2025, 122K Spaces were created, 71K of them websites built with DeepSite. That's about 6x the monthly pace of late 2024, while likes given per month fell from about 35K to about 20K.
New in the app: a page for every dataset, with daily downloads and the models trained on it (636 list FineWeb), a page for every Space, and rankings for both.
Thank you to everyone who liked Model Pulse this week: it made Spaces of the Week and is #7 on trending. Thanks also to @dipankarsarkar , whose comments on the last post fixed three data issues. If a number looks wrong, tell me.
Spaces: tardellirs/model-pulse
- 24 replies
reacted to Undi95's post with 🔥 12 days ago
Post
4,382
Yo, I'm back, and I'm currently trying to teach a local LLM to stop waiting for a prompt kek.
I'm building a small proof of concept: can an open-weight model (Qwen3.8-27B, running locally on 2 RTX 5090 GPUs) learn to direct itself, then improve from its own exploration, without a human in the loop and without breaking it for normal use?
No user, no task. The model only gets observations from its environment. Each turn, it writes its own agenda (goal/open questions/next step), then picks an action: search the web, read a page, or take a note.
The environment is the judge, not another LLM. A note is accepted only if it quotes the page it read word for word. Facts are checked by exact match.
Later, code will be checked by actually running tests.
The best episodes become fine-tuning data (LoRA). The helper system prompt is removed at training time, so the behavior has to live in the weights.
Each new model goes through a fixed benchmark gate: math, general knowledge, "does it still answer humans normally?", autonomy, and learned facts on held-out sources. It's kept only if nothing regresses, otherwise it's discarded. Then the loop starts again.
The full pipeline works end to end: collect, train, merge, deploy, benchmark. The baseline is clear. Without any instructions, the base model's real autonomy is zero: it behaves like a chatbot waiting for a question. That's the number this small project is trying to move.
I haven't found a public tool that runs this whole loop (self-directed exploration, verifiable rewards, continual fine-tuning and a regression gate) on home hardware. The goal isn't AGI in a bedroom. It's to show that anyone can try it, measure it honestly, and see where it breaks.
Code and results will be released once the first real iterations are done. At the moment the code is... running, but made with scotch and stick, still only a PoC I want to try.
Did you already tried something like that? What was your result? I'm curious!
I'm building a small proof of concept: can an open-weight model (Qwen3.8-27B, running locally on 2 RTX 5090 GPUs) learn to direct itself, then improve from its own exploration, without a human in the loop and without breaking it for normal use?
No user, no task. The model only gets observations from its environment. Each turn, it writes its own agenda (goal/open questions/next step), then picks an action: search the web, read a page, or take a note.
The environment is the judge, not another LLM. A note is accepted only if it quotes the page it read word for word. Facts are checked by exact match.
Later, code will be checked by actually running tests.
The best episodes become fine-tuning data (LoRA). The helper system prompt is removed at training time, so the behavior has to live in the weights.
Each new model goes through a fixed benchmark gate: math, general knowledge, "does it still answer humans normally?", autonomy, and learned facts on held-out sources. It's kept only if nothing regresses, otherwise it's discarded. Then the loop starts again.
The full pipeline works end to end: collect, train, merge, deploy, benchmark. The baseline is clear. Without any instructions, the base model's real autonomy is zero: it behaves like a chatbot waiting for a question. That's the number this small project is trying to move.
I haven't found a public tool that runs this whole loop (self-directed exploration, verifiable rewards, continual fine-tuning and a regression gate) on home hardware. The goal isn't AGI in a bedroom. It's to show that anyone can try it, measure it honestly, and see where it breaks.
Code and results will be released once the first real iterations are done. At the moment the code is... running, but made with scotch and stick, still only a PoC I want to try.
Did you already tried something like that? What was your result? I'm curious!
- 22 replies
liked a
model 12 days ago
liked a
model 13 days ago
reacted to pollix's post with 🚀 17 days ago
Post
6,451
stuntd 0.1.2 is out 🎉
stuntd sits in front of your LLM, learns its typed decisions and answers the confident ones locally with a small head on the Laya encoder by @convaiinnovations . About 20ms on GPU and 60ms on CPU, and anything it isn't sure about still goes to the big model.
New in 0.1.2:
- decisions with several fields, like category + urgency + needs_human in one call, answered locally only when every field is sure
- the Anthropic Messages API learns too, not only OpenAI
- auto_retrain: the daemon retrains a site in the background once enough new traffic comes in, so collect, train, shadow and live run on their own
- serve --lazy loads the checkpoint on the first request
Try it in the browser: pollix/stuntd
Code: https://github.com/bladedevoff/stuntd
stuntd sits in front of your LLM, learns its typed decisions and answers the confident ones locally with a small head on the Laya encoder by @convaiinnovations . About 20ms on GPU and 60ms on CPU, and anything it isn't sure about still goes to the big model.
New in 0.1.2:
- decisions with several fields, like category + urgency + needs_human in one call, answered locally only when every field is sure
- the Anthropic Messages API learns too, not only OpenAI
- auto_retrain: the daemon retrains a site in the background once enough new traffic comes in, so collect, train, shadow and live run on their own
- serve --lazy loads the checkpoint on the first request
Try it in the browser: pollix/stuntd
Code: https://github.com/bladedevoff/stuntd
pip install -U stuntd
- 1 reply
liked a
Space 17 days ago
liked 2
models 18 days ago
reacted to Banaxi-Tech's post with 🔥 20 days ago
Post
7,629
We're releasing a MAJOR update to the BananaAll SLM Super App.
If you want to use a custom architecture, previously you had to go trough reviewing the code yourself, now add an Openrouter API key and review it with GPT 6 Luna in one button. A review cost be half a cent so anyone can try it. This is one of the main features.
Now ROCm, AMD and Windows, Mac support.
Colab and Molab support.
Detailed list of features:
Get improved Windows Python detection and support paths for compatible AMD ROCm, Intel XPU, and Apple MPS setups.
Choose local training or export a self-contained Python script for Colab or Molab. Notebook runs produce a downloadable model ZIP.
Start pretraining with an existing model’s tokenizer, or train a new one from your datasets.
Try experimental 1.58-bit Ternary fake-quantized training on NVIDIA GPUs.
Watch live tokens per second. Model compilation is on by default and falls back automatically if it fails.
Build custom architectures with separate configuration and modeling files, then review the training code manually or with optional OpenRouter AI Review.
Install from source with the new coding-agent instructions.
This release also fixes inflated loss reporting for custom models.
And for those users who didn't want to try it out just because installation would be so hard, it isnt now.
Go to any coding agent (Pi, Claude Code, Codex, OpenCode, basically all work), and just paste "Install BananaAll for me. Fetch and follow https://raw.githubusercontent.com/BananaMind/BananaAll/main/agent_install.txt."
That's it.
Check it out at https://github.com/BananaMind/BananaAll/
Also on SAICR, we're currently training a new major model (NACR v2) and ACR 1.0 is in the finishing.
If you want to use a custom architecture, previously you had to go trough reviewing the code yourself, now add an Openrouter API key and review it with GPT 6 Luna in one button. A review cost be half a cent so anyone can try it. This is one of the main features.
Now ROCm, AMD and Windows, Mac support.
Colab and Molab support.
Detailed list of features:
Get improved Windows Python detection and support paths for compatible AMD ROCm, Intel XPU, and Apple MPS setups.
Choose local training or export a self-contained Python script for Colab or Molab. Notebook runs produce a downloadable model ZIP.
Start pretraining with an existing model’s tokenizer, or train a new one from your datasets.
Try experimental 1.58-bit Ternary fake-quantized training on NVIDIA GPUs.
Watch live tokens per second. Model compilation is on by default and falls back automatically if it fails.
Build custom architectures with separate configuration and modeling files, then review the training code manually or with optional OpenRouter AI Review.
Install from source with the new coding-agent instructions.
This release also fixes inflated loss reporting for custom models.
And for those users who didn't want to try it out just because installation would be so hard, it isnt now.
Go to any coding agent (Pi, Claude Code, Codex, OpenCode, basically all work), and just paste "Install BananaAll for me. Fetch and follow https://raw.githubusercontent.com/BananaMind/BananaAll/main/agent_install.txt."
That's it.
Check it out at https://github.com/BananaMind/BananaAll/
Also on SAICR, we're currently training a new major model (NACR v2) and ACR 1.0 is in the finishing.
- 11 replies
reacted to SeaWolf-AI's post with 🔥 21 days ago
Post
3,808
🖼️ NO GPU, Only CPU : Z-Image model
Zero graphics cards. 46 seconds. Photoreal.
That laptop you're reading this on. No graphics card, right? It generates images.
No CUDA install. No Python environment. No driver changes. One binary, three model files. Done.
📊 Measured — GPU count used: zero
512×512 : 46.4 s
Korean prompt : 45.3 s (faster than English)
1024×1024 : 192.7 s
Peak RAM : 6.42 GB
GPUs used : 0
(Intel Xeon Gold 6526Y ×2, 48 threads, Q4_0, 3 steps)
⚡ From 244 seconds to 46 — 5.3×
Run it on defaults and it takes 244 s. Switch to 3 steps and it's 48.6 s. Add VAE tiling and it's 46.4 s.
The biggest culprit was the default. Z-Image Turbo is distilled to paint in few strokes, but the tool's default is 20. We were throwing away 5× for no reason. So were we, at first.
3 is the floor. Put 4 and 3 side by side and you cannot tell them apart. At 2 it collapses — water droplets and wood grain vanish, and the surface turns cloth-like.
🔗 Links
Model
FINAL-Bench/POCKET-Zimage-CPU
Live demo Space (runs on CPU)
FINAL-Bench/POCKET-Zimage-CPU
POCKET collection
FINAL-Bench/pocket-models-6a618ee5d23eafb7e185a5c6
Zero graphics cards. 46 seconds. Photoreal.
That laptop you're reading this on. No graphics card, right? It generates images.
No CUDA install. No Python environment. No driver changes. One binary, three model files. Done.
📊 Measured — GPU count used: zero
512×512 : 46.4 s
Korean prompt : 45.3 s (faster than English)
1024×1024 : 192.7 s
Peak RAM : 6.42 GB
GPUs used : 0
(Intel Xeon Gold 6526Y ×2, 48 threads, Q4_0, 3 steps)
⚡ From 244 seconds to 46 — 5.3×
Run it on defaults and it takes 244 s. Switch to 3 steps and it's 48.6 s. Add VAE tiling and it's 46.4 s.
The biggest culprit was the default. Z-Image Turbo is distilled to paint in few strokes, but the tool's default is 20. We were throwing away 5× for no reason. So were we, at first.
3 is the floor. Put 4 and 3 side by side and you cannot tell them apart. At 2 it collapses — water droplets and wood grain vanish, and the surface turns cloth-like.
🔗 Links
Model
FINAL-Bench/POCKET-Zimage-CPU
Live demo Space (runs on CPU)
FINAL-Bench/POCKET-Zimage-CPU
POCKET collection
FINAL-Bench/pocket-models-6a618ee5d23eafb7e185a5c6
- 2 replies
liked a
model 21 days ago
