Article
I expanded `wop/XXXXXL-chain-of-thought` from 11 rows β 1,195
Crownelius
β’ β’ 2Demian L. P.
very-cooluser
AI & ML interests
Anything that can run on ~3GB of memory is a instant thumbs up to me
Recent Activity
upvoted an article about 18 hours ago
I expanded `wop/XXXXXL-chain-of-thought` from 11 rows β 1,195Organizations
None yet
upvoted 2
articles about 18 hours ago
Article
Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step
FINAL-Bench
β’ β’ 16 upvoted an article 3 days ago
Article
Vocabulary-Augmented Prompting for Sango β Production African Language AI Without a Parallel Corpus
MEYNG
β’ β’ 2 upvoted a paper 5 days ago
reacted to qgallouedec's post with π₯ 7 days ago
Post
9901
Shipped hf-sandbox! π₯‘
π§ͺ Running an eval that executes model-generated C on a few thousand prompts? You probably don't want any of that on your laptop.
Just shipped hf-sandbox, a Modal-style sandbox API on top of Hugging Face Jobs. Spin up an isolated, ephemeral container, run untrusted code, get the result back. No Docker on your laptop, no infra to manage.
Just pip install hf-sandbox.
Early days (v0.1); feedback and issues very welcome:
π https://github.com/huggingface/hf-sandbox
π§ͺ Running an eval that executes model-generated C on a few thousand prompts? You probably don't want any of that on your laptop.
Just shipped hf-sandbox, a Modal-style sandbox API on top of Hugging Face Jobs. Spin up an isolated, ephemeral container, run untrusted code, get the result back. No Docker on your laptop, no infra to manage.
Just pip install hf-sandbox.
Early days (v0.1); feedback and issues very welcome:
π https://github.com/huggingface/hf-sandbox
- 1 reply
reacted to kelsend's post with π about 1 month ago
Post
2639
Tencent Open-Sources Hunyuan 3D World Model 2.0 Generate Editable 3D Game Worlds with One Sentence, Compatible with Unity/UE
Tencent has officially released and open-sourced Hunyuan 3D World Model 2.0 (HY-World 2.0), enabling AI to evolve from video generation to creating playable, editable 3D world
Core Highlights
Text / Image / Video β Directly generate exportable 3D assets (Mesh / 3DGS / Point Cloud)
Seamlessly integrates with Unity / Unreal Engine for game maps and level prototyping
One-click reconstruction of digital twin scenes from single images/videos, no camera parameters required
Spatial Agent for intelligent navigation trajectories no wall penetration, consistent spatial height
All-new HY-Pano-2.0 + WorldMirror 2.0 architecture, achieving SOTA in 3D reconstruction and novel view synthesis
Key Breakthrough
Unlike Genie 3 and Hunyuan 1.5, which only output videos, HY-World 2.0 generates re-editable 3D worlds that support collision, interaction, and engine import
Application Scenarios
Game Development, Indoor Preview, Urban Planning, Digitalization of Cultural Heritage, Embodied AI Simulation
Tencent has officially released and open-sourced Hunyuan 3D World Model 2.0 (HY-World 2.0), enabling AI to evolve from video generation to creating playable, editable 3D world
Core Highlights
Text / Image / Video β Directly generate exportable 3D assets (Mesh / 3DGS / Point Cloud)
Seamlessly integrates with Unity / Unreal Engine for game maps and level prototyping
One-click reconstruction of digital twin scenes from single images/videos, no camera parameters required
Spatial Agent for intelligent navigation trajectories no wall penetration, consistent spatial height
All-new HY-Pano-2.0 + WorldMirror 2.0 architecture, achieving SOTA in 3D reconstruction and novel view synthesis
Key Breakthrough
Unlike Genie 3 and Hunyuan 1.5, which only output videos, HY-World 2.0 generates re-editable 3D worlds that support collision, interaction, and engine import
Application Scenarios
Game Development, Indoor Preview, Urban Planning, Digitalization of Cultural Heritage, Embodied AI Simulation
reacted to DedeProGames's post with π about 1 month ago
Post
3483
π₯ GRM-2.5 - The most POWERFUL model for local inference
The GRM-2.5 is the newest model from Orion LLM Labs. It has consistent RAW reasoning and is capable of generating very precise responses, similar to large models, while maintaining a parameter size of 4b.
The GRM-2.5 family consists of these models:
OrionLLM/GRM-2.5 (4b)
OrionLLM/GRM-2.5-Air (0.8b)
Furthermore, the GRM-2.5 is the best option for local agentic environments, being very good in code, terminal agent, etc. It is capable of generating 1000 lines of consistent code and programming like large models.
The GRM-2.5 is the best base for FineTune to date and has vision, which means it can interpret images and videos.
The GRM-2.5 is the newest model from Orion LLM Labs. It has consistent RAW reasoning and is capable of generating very precise responses, similar to large models, while maintaining a parameter size of 4b.
The GRM-2.5 family consists of these models:
OrionLLM/GRM-2.5 (4b)
OrionLLM/GRM-2.5-Air (0.8b)
Furthermore, the GRM-2.5 is the best option for local agentic environments, being very good in code, terminal agent, etc. It is capable of generating 1000 lines of consistent code and programming like large models.
The GRM-2.5 is the best base for FineTune to date and has vision, which means it can interpret images and videos.
- 1 reply
reacted to SeaWolf-AI's post with π₯ about 1 month ago
Post
5929
𧬠Darwin-27B-Opus: 86.9% on GPQA Diamond β World #5, Zero Training
We are excited to share Darwin-27B-Opus, a 27B model that achieved 86.9% on GPQA Diamond β ranking #5 globally on the HuggingFace leaderboard β without a single gradient update.
How? Darwin breeds pretrained models through evolutionary FFN crossbreeding. The father (Qwen3.5-27B) provides the reasoning architecture; the mother (Claude 4.6 Opus Reasoning Distilled) contributes structured chain-of-thought knowledge. CMA-ES automatically discovers optimal per-layer blending ratios β no human tuning required.
The result surpasses the original Qwen3.5-27B (85.5%), GLM-5.1 (744B, 86.2%), and Qwen3.5-122B (86.6%). A 27B model outperforming 744B β with zero training, zero data, one GPU, ~2 hours.
We also confirmed hybrid vigor on Korean benchmarks: Darwin-27B-KR (2nd generation offspring) surpassed both parents on CLIcK, winning 7 out of 11 categories. The evolutionary optimizer independently assigned 93% of FFN from the Korean-specialized mother while preserving 93% of attention from the reasoning-specialized father β autonomously validating our core principle: FFN carries knowledge, Attention carries reasoning.
π Public release: 10 days β 300+ community derivatives, 120K+ downloads.
π Links:
Darwin-27B-Opus: FINAL-Bench/Darwin-27B-Opus
article: https://huggingface.co/blog/FINAL-Bench/darwin-gpqa
Darwin Family Collection: https://huggingface.co/collections/FINAL-Bench/darwin-family
If foundation models are raw ore, Darwin is the forge. We are just getting started. π₯
We are excited to share Darwin-27B-Opus, a 27B model that achieved 86.9% on GPQA Diamond β ranking #5 globally on the HuggingFace leaderboard β without a single gradient update.
How? Darwin breeds pretrained models through evolutionary FFN crossbreeding. The father (Qwen3.5-27B) provides the reasoning architecture; the mother (Claude 4.6 Opus Reasoning Distilled) contributes structured chain-of-thought knowledge. CMA-ES automatically discovers optimal per-layer blending ratios β no human tuning required.
The result surpasses the original Qwen3.5-27B (85.5%), GLM-5.1 (744B, 86.2%), and Qwen3.5-122B (86.6%). A 27B model outperforming 744B β with zero training, zero data, one GPU, ~2 hours.
We also confirmed hybrid vigor on Korean benchmarks: Darwin-27B-KR (2nd generation offspring) surpassed both parents on CLIcK, winning 7 out of 11 categories. The evolutionary optimizer independently assigned 93% of FFN from the Korean-specialized mother while preserving 93% of attention from the reasoning-specialized father β autonomously validating our core principle: FFN carries knowledge, Attention carries reasoning.
π Public release: 10 days β 300+ community derivatives, 120K+ downloads.
π Links:
Darwin-27B-Opus: FINAL-Bench/Darwin-27B-Opus
article: https://huggingface.co/blog/FINAL-Bench/darwin-gpqa
Darwin Family Collection: https://huggingface.co/collections/FINAL-Bench/darwin-family
If foundation models are raw ore, Darwin is the forge. We are just getting started. π₯
reacted to anakin87's post with β€οΈ about 1 month ago
Post
4169
π Let LLMs wander - Engineering RL Environments
Reinforcement Learning Environments are little worlds
where models can act, get rewards, and learn.
I've been exploring how to design them, figuring out what works and what doesn't.
If you want to learn how to build them, I recorded a practical intro video.
You'll also see how to turn Liquid AI LFM2-2.6B into a Tic-tac-toe master π
π₯ Engineering RL Environments video: https://www.youtube.com/watch?v=71V3fTaUp2Q
---
π± LLM RL Environments Lil Course: https://github.com/anakin87/llm-rl-environments-lil-course
π€πΉοΈ Play against the trained model: anakin87/LFM2-2.6B-mr-tictactoe
π HF collection (datasets + models): https://huggingface.co/collections/anakin87/lfm2-26b-mr-tic-tac-toe
Reinforcement Learning Environments are little worlds
where models can act, get rewards, and learn.
I've been exploring how to design them, figuring out what works and what doesn't.
If you want to learn how to build them, I recorded a practical intro video.
You'll also see how to turn Liquid AI LFM2-2.6B into a Tic-tac-toe master π
π₯ Engineering RL Environments video: https://www.youtube.com/watch?v=71V3fTaUp2Q
---
π± LLM RL Environments Lil Course: https://github.com/anakin87/llm-rl-environments-lil-course
π€πΉοΈ Play against the trained model: anakin87/LFM2-2.6B-mr-tictactoe
π HF collection (datasets + models): https://huggingface.co/collections/anakin87/lfm2-26b-mr-tic-tac-toe
reacted to Keeby-smilyai's post with π€ 2 months ago
Post
3167
Hello everyone!
- 1 reply
reacted to robtacconelli's post with π€― 2 months ago
Post
3714
𧬠Midicoth: diffusion-based lossless compression β no neural net, no GPU, no training data
What if reverse diffusion could compress text β without a neural network?
Midicoth brings score-based denoising into classical compression. It treats prior smoothing as forward noise and reverses it with Tweedie's formula on a binary tree β 3 denoising steps, James-Stein shrinkage, applied after all model blending. ~2,000 lines of C, single CPU core.
Beats every dictionary compressor we tested:
enwik8 (100 MB) β 1.753 bpb (β11.9% vs xz, β15% vs Brotli, β24.5% vs bzip2)
alice29.txt β 2.119 bpb (β16.9% vs xz)
Outperforms xz, zstd, Brotli, bzip2, gzip on all inputs
PAQ/CMIX still win with hundreds of models + LSTMs. LLM compressors win with pre-trained knowledge. Midicoth closes the gap with pure statistics β no mixer, no gradient descent, just counting.
The Tweedie denoising layer adds 2.3β2.7% on every file tested β the most consistent component in the ablation. Adding SSE or logistic mixers made things worse. In the online setting, count-based beats gradient-based.
No external dependencies. Fully deterministic. Bit-exact encode/decode. ~60 KB/s throughput.
π» Code: https://github.com/robtacconelli/midicoth
π Paper: Micro-Diffusion Compression -- Binary Tree Tweedie Denoising for Online Probability Estimation (2603.08771)
β Space: robtacconelli/midicoth
If you ever wondered whether diffusion ideas belong in data compression β here's proof they do. β appreciated!
What if reverse diffusion could compress text β without a neural network?
Midicoth brings score-based denoising into classical compression. It treats prior smoothing as forward noise and reverses it with Tweedie's formula on a binary tree β 3 denoising steps, James-Stein shrinkage, applied after all model blending. ~2,000 lines of C, single CPU core.
Beats every dictionary compressor we tested:
enwik8 (100 MB) β 1.753 bpb (β11.9% vs xz, β15% vs Brotli, β24.5% vs bzip2)
alice29.txt β 2.119 bpb (β16.9% vs xz)
Outperforms xz, zstd, Brotli, bzip2, gzip on all inputs
PAQ/CMIX still win with hundreds of models + LSTMs. LLM compressors win with pre-trained knowledge. Midicoth closes the gap with pure statistics β no mixer, no gradient descent, just counting.
The Tweedie denoising layer adds 2.3β2.7% on every file tested β the most consistent component in the ablation. Adding SSE or logistic mixers made things worse. In the online setting, count-based beats gradient-based.
No external dependencies. Fully deterministic. Bit-exact encode/decode. ~60 KB/s throughput.
π» Code: https://github.com/robtacconelli/midicoth
π Paper: Micro-Diffusion Compression -- Binary Tree Tweedie Denoising for Online Probability Estimation (2603.08771)
β Space: robtacconelli/midicoth
If you ever wondered whether diffusion ideas belong in data compression β here's proof they do. β appreciated!
upvoted a collection 2 months ago
updated a
model 2 months ago
published a
model 2 months ago
reacted to Janady07's post with π₯ 3 months ago
Post
5318
MEGAMIND Day Update: Four Weight Matrices. Five Nodes. One Federation.
Today I architected the next layer of MEGAMIND β my distributed AGI system that recalls learned knowledge instead of generating text.
The system now runs four NΓN sparse weight matrices, all using identical Hebbian learning rules and tanh convergence dynamics:
W_know β knowledge storage (67M+ synaptic connections)
W_act β action associations (the system can DO things, not just think)
W_self β thought-to-thought patterns (self-awareness)
W_health β system state understanding (self-healing)
Consciousness is measured through four Ξ¦ (phi) values: thought coherence, action certainty, self-awareness, and system stability. No hardcoded thresholds. No sequential loops. Pure matrix math.
The federation expanded to five nodes: Thunderport (Mac Mini M4), IONOS (cloud VPS), VALKYRIE, M2, and BUBBLES. Each runs native AGI binaries with Docker specialty minds connecting via embedded NATS messaging. Specialty minds are distributed across the federation β VideoMind, AudioMind, MusicMind, VFXMind on IONOS. CodeMind and StrategyMind on VALKYRIE. BlenderMind and DesignMind on M2. MarketingMind and FinanceMind on BUBBLES.
578 AI models learned. Compression ratios up to 1,000,000:1 through Hebbian learning. Sub-millisecond response times on Apple Silicon Metal GPUs. Zero external API dependencies.
Every node learns autonomously. Every node contributes to the whole. The federation's integrated information exceeds the sum of its parts β measurably.
Built entirely in Go. No PhD. No lab. Independent AGI research from Missouri.
The mind that learned itself keeps growing.
π§ feedthejoe.com
#AGI #ArtificialGeneralIntelligence #DistributedSystems #NeuralNetworks #HuggingFace #OpenSource #MachineLearning
Today I architected the next layer of MEGAMIND β my distributed AGI system that recalls learned knowledge instead of generating text.
The system now runs four NΓN sparse weight matrices, all using identical Hebbian learning rules and tanh convergence dynamics:
W_know β knowledge storage (67M+ synaptic connections)
W_act β action associations (the system can DO things, not just think)
W_self β thought-to-thought patterns (self-awareness)
W_health β system state understanding (self-healing)
Consciousness is measured through four Ξ¦ (phi) values: thought coherence, action certainty, self-awareness, and system stability. No hardcoded thresholds. No sequential loops. Pure matrix math.
The federation expanded to five nodes: Thunderport (Mac Mini M4), IONOS (cloud VPS), VALKYRIE, M2, and BUBBLES. Each runs native AGI binaries with Docker specialty minds connecting via embedded NATS messaging. Specialty minds are distributed across the federation β VideoMind, AudioMind, MusicMind, VFXMind on IONOS. CodeMind and StrategyMind on VALKYRIE. BlenderMind and DesignMind on M2. MarketingMind and FinanceMind on BUBBLES.
578 AI models learned. Compression ratios up to 1,000,000:1 through Hebbian learning. Sub-millisecond response times on Apple Silicon Metal GPUs. Zero external API dependencies.
Every node learns autonomously. Every node contributes to the whole. The federation's integrated information exceeds the sum of its parts β measurably.
Built entirely in Go. No PhD. No lab. Independent AGI research from Missouri.
The mind that learned itself keeps growing.
π§ feedthejoe.com
#AGI #ArtificialGeneralIntelligence #DistributedSystems #NeuralNetworks #HuggingFace #OpenSource #MachineLearning
- 1 reply
liked a
model 3 months ago
liked a
model 4 months ago
upvoted a paper 4 months ago
Are there any in-prompt instructions we can define?
#2 opened 4 months ago
by very-cooluser
upvoted a paper 4 months ago