Hugging Face Changelog
Introducing Buckets: S3-like storage on the Hub
• 126
Bambuu
bambuuai
AI & ML interests
None yet
Recent Activity
upvoted a changelog 1 day ago
Introducing Buckets: S3-like storage on the Hub upvoted an article 3 days ago
Introducing Storage Buckets on the Hugging Face Hub reacted
to
SeaWolf-AI's
post with 🔥 3 days ago
🏟️ Smol AI WorldCup: A 4B Model Just Beat 8B — Here's the Data
We evaluated 18 small language models from 12 makers on 125 questions across 7 languages. The results challenge the assumption that bigger is always better.
Community Article: https://huggingface.co/blog/FINAL-Bench/smol-worldcup
Live Leaderboard: https://huggingface.co/spaces/ginigen-ai/smol-worldcup
Dataset: https://huggingface.co/datasets/ginigen-ai/smol-worldcup
What we found:
→ Gemma-3n-E4B (4B, 2GB RAM) outscores Qwen3-8B (8B, 5.5GB). Doubling parameters gained only 0.4 points. RAM cost: 2.75x more.
→ GPT-OSS-20B fits in 1.5GB yet matches Champions-league dense models requiring 8.5GB. MoE architecture is the edge AI game-changer.
→ Thinking models hurt structured output. DeepSeek-R1-7B scores 8.7 points below same-size Qwen3-8B and runs 2.7x slower.
→ A 1.3B model fabricates confident fake content 80% of the time when prompted with nonexistent entities. Qwen3 family hits 100% trap detection across all sizes.
→ Qwen3-1.7B (1.2GB) outscores Mistral-7B, Llama-3.1-8B, and DeepSeek-R1-14B. Latest architecture at 1.7B beats older architecture at 14B.
What makes this benchmark different?
Most benchmarks ask "how smart?" — we measure five axes simultaneously: Size, Honesty, Intelligence, Fast, Thrift (SHIFT). Our ranking metric WCS = sqrt(SHIFT x PIR_norm) rewards models that are both high-quality AND efficient. Smart but massive? Low rank. Tiny but poor? Also low.
Top 5 by WCS:
1. GPT-OSS-20B — WCS 82.6 — 1.5GB — Raspberry Pi tier
2. Gemma-3n-E4B — WCS 81.8 — 2.0GB — Smartphone tier
3. Llama-4-Scout — WCS 79.3 — 240 tok/s — Fastest model
4. Qwen3-4B — WCS 76.6 — 2.8GB — Smartphone tier
5. Qwen3-1.7B — WCS 76.1 — 1.2GB — IoT tier
Built in collaboration with the FINAL Bench research team. Interoperable with ALL Bench Leaderboard for full small-to-large model comparison.
Dataset is open under Apache 2.0 (125 questions, 7 languages). We welcome new model submissions. Organizations
upvoted a changelog 1 day ago
upvoted an article 3 days ago
Article
Introducing Storage Buckets on the Hugging Face Hub
- +10
•
160
reacted to SeaWolf-AI's post with 🔥 3 days ago
Post
11002
🏟️ Smol AI WorldCup: A 4B Model Just Beat 8B — Here's the Data
We evaluated 18 small language models from 12 makers on 125 questions across 7 languages. The results challenge the assumption that bigger is always better.
Community Article: https://huggingface.co/blog/FINAL-Bench/smol-worldcup
Live Leaderboard: ginigen-ai/smol-worldcup
Dataset: ginigen-ai/smol-worldcup
What we found:
→ Gemma-3n-E4B (4B, 2GB RAM) outscores Qwen3-8B (8B, 5.5GB). Doubling parameters gained only 0.4 points. RAM cost: 2.75x more.
→ GPT-OSS-20B fits in 1.5GB yet matches Champions-league dense models requiring 8.5GB. MoE architecture is the edge AI game-changer.
→ Thinking models hurt structured output. DeepSeek-R1-7B scores 8.7 points below same-size Qwen3-8B and runs 2.7x slower.
→ A 1.3B model fabricates confident fake content 80% of the time when prompted with nonexistent entities. Qwen3 family hits 100% trap detection across all sizes.
→ Qwen3-1.7B (1.2GB) outscores Mistral-7B, Llama-3.1-8B, and DeepSeek-R1-14B. Latest architecture at 1.7B beats older architecture at 14B.
What makes this benchmark different?
Most benchmarks ask "how smart?" — we measure five axes simultaneously: Size, Honesty, Intelligence, Fast, Thrift (SHIFT). Our ranking metric WCS = sqrt(SHIFT x PIR_norm) rewards models that are both high-quality AND efficient. Smart but massive? Low rank. Tiny but poor? Also low.
Top 5 by WCS:
1. GPT-OSS-20B — WCS 82.6 — 1.5GB — Raspberry Pi tier
2. Gemma-3n-E4B — WCS 81.8 — 2.0GB — Smartphone tier
3. Llama-4-Scout — WCS 79.3 — 240 tok/s — Fastest model
4. Qwen3-4B — WCS 76.6 — 2.8GB — Smartphone tier
5. Qwen3-1.7B — WCS 76.1 — 1.2GB — IoT tier
Built in collaboration with the FINAL Bench research team. Interoperable with ALL Bench Leaderboard for full small-to-large model comparison.
Dataset is open under Apache 2.0 (125 questions, 7 languages). We welcome new model submissions.
We evaluated 18 small language models from 12 makers on 125 questions across 7 languages. The results challenge the assumption that bigger is always better.
Community Article: https://huggingface.co/blog/FINAL-Bench/smol-worldcup
Live Leaderboard: ginigen-ai/smol-worldcup
Dataset: ginigen-ai/smol-worldcup
What we found:
→ Gemma-3n-E4B (4B, 2GB RAM) outscores Qwen3-8B (8B, 5.5GB). Doubling parameters gained only 0.4 points. RAM cost: 2.75x more.
→ GPT-OSS-20B fits in 1.5GB yet matches Champions-league dense models requiring 8.5GB. MoE architecture is the edge AI game-changer.
→ Thinking models hurt structured output. DeepSeek-R1-7B scores 8.7 points below same-size Qwen3-8B and runs 2.7x slower.
→ A 1.3B model fabricates confident fake content 80% of the time when prompted with nonexistent entities. Qwen3 family hits 100% trap detection across all sizes.
→ Qwen3-1.7B (1.2GB) outscores Mistral-7B, Llama-3.1-8B, and DeepSeek-R1-14B. Latest architecture at 1.7B beats older architecture at 14B.
What makes this benchmark different?
Most benchmarks ask "how smart?" — we measure five axes simultaneously: Size, Honesty, Intelligence, Fast, Thrift (SHIFT). Our ranking metric WCS = sqrt(SHIFT x PIR_norm) rewards models that are both high-quality AND efficient. Smart but massive? Low rank. Tiny but poor? Also low.
Top 5 by WCS:
1. GPT-OSS-20B — WCS 82.6 — 1.5GB — Raspberry Pi tier
2. Gemma-3n-E4B — WCS 81.8 — 2.0GB — Smartphone tier
3. Llama-4-Scout — WCS 79.3 — 240 tok/s — Fastest model
4. Qwen3-4B — WCS 76.6 — 2.8GB — Smartphone tier
5. Qwen3-1.7B — WCS 76.1 — 1.2GB — IoT tier
Built in collaboration with the FINAL Bench research team. Interoperable with ALL Bench Leaderboard for full small-to-large model comparison.
Dataset is open under Apache 2.0 (125 questions, 7 languages). We welcome new model submissions.
- 1 reply
liked a dataset 4 days ago
commented on Mixture of Experts (MoEs) in Transformers 6 days ago
The enable_expert_parallel flag hiding the complexity of GroupedGemmParallel + RouterParallel behind a single config is a great DX win — distributing experts across devices used to require a lot of custom plumbing.
upvoted an article 6 days ago
Article
Mixture of Experts (MoEs) in Transformers
- +5
•
132
reacted to danielhanchen's post with 🤗 8 days ago
Post
2810
You can now fine-tune Qwen3.5 for free with our notebook! 🔥
You just need 5GB VRAM to train Qwen3.5-2B LoRA locally!
Unsloth trains Qwen3.5 1.5x faster with 50% less VRAM.
GitHub: https://github.com/unslothai/unsloth
Guide: https://unsloth.ai/docs/models/qwen3.5/fine-tune
Qwen3.5-4B Colab: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Qwen3_5_(4B)_Vision.ipynb
You just need 5GB VRAM to train Qwen3.5-2B LoRA locally!
Unsloth trains Qwen3.5 1.5x faster with 50% less VRAM.
GitHub: https://github.com/unslothai/unsloth
Guide: https://unsloth.ai/docs/models/qwen3.5/fine-tune
Qwen3.5-4B Colab: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Qwen3_5_(4B)_Vision.ipynb
updated a model 9 days ago
liked 2 models 9 days ago
published a model 9 days ago
updated a model 10 days ago
liked a model 10 days ago
reacted to imnotkitty's post with 🚀🚀 10 days ago
Post
2258
The most popular OpenClaw tool has been released!
🥇 Claw for All: The ultimate all-rounder. Simplifies deployment for both devs & pros with a seamless web/mobile experience.
🥈 OpenClaw Launch: Speed is king. Deploy your apps in under 30 seconds with a single click.
🥉 ClawTeam: Skip the setup. Get pre-configured AI agent blueprints built specifically for OpenClaw.
4️⃣ vibeclaw: Local-first. Run OpenClaw in your browser sandbox in literally 1 second.
5️⃣ Tinkerclaw: The startup favorite. Zero-code platform to deploy, manage, and scale AI assistants.
6️⃣ ClawWrapper: The "last mile" tool. Simplifies the entire packaging and launch process.
Which one are you adding to your stack? 🛠️
(Source: OpenClaw Directory)
🥇 Claw for All: The ultimate all-rounder. Simplifies deployment for both devs & pros with a seamless web/mobile experience.
🥈 OpenClaw Launch: Speed is king. Deploy your apps in under 30 seconds with a single click.
🥉 ClawTeam: Skip the setup. Get pre-configured AI agent blueprints built specifically for OpenClaw.
4️⃣ vibeclaw: Local-first. Run OpenClaw in your browser sandbox in literally 1 second.
5️⃣ Tinkerclaw: The startup favorite. Zero-code platform to deploy, manage, and scale AI assistants.
6️⃣ ClawWrapper: The "last mile" tool. Simplifies the entire packaging and launch process.
Which one are you adding to your stack? 🛠️
(Source: OpenClaw Directory)
- 2 replies
updated a model 10 days ago
published a model 10 days ago
updated a dataset 11 days ago
reacted to OzTianlu's post with 🔥 11 days ago
Post
4770
🔥 UPGRADE in Kai: 30B Scaling! 🔥
NoesisLab/Kai-30B-Instruct
NoesisLab/Kai-30B-Instruct
We are incredibly excited to announce that the Kai-30B-Instruct model and its official Space are now LIVE! 🚀
If you've been following the journey from Kai-0.35B to Kai-3B, you know we're rethinking how models reason. Tired of verbose, slow Chain-of-Thought (CoT) outputs that flood your screen with self-talk? So are we.
Kai-30B-Instruct scales up our Adaptive Dual-Search Distillation (ADS) framework. By bridging classical A* heuristic search with continuous gradient descent , we use an information-theoretic log-barrier to physically prune high-entropy reasoning paths during training.
The result? Pure implicit reasoning. The model executes structured logic, arithmetic carries, and branch selections as a reflex in a single forward pass—no external scaffolding required.
At 3B, we observed a phase transition where the model achieved "logical crystallization". Now, at 30B, we are giving the ADS regularizer the massive representational capacity it needs to tackle higher-order symbolic abstractions and complex reasoning tasks.
🧪 Test Kai yourself in our new Space:
NoesisLab/Kai-30B-Instruct
📦 Model Weights:
NoesisLab/Kai-30B-Instruct
Bring your hardest math, logic, and coding benchmarks. We invite the community to stress-test the limits of the penalty wall! 🧱💥
NoesisLab/Kai-30B-Instruct
NoesisLab/Kai-30B-Instruct
We are incredibly excited to announce that the Kai-30B-Instruct model and its official Space are now LIVE! 🚀
If you've been following the journey from Kai-0.35B to Kai-3B, you know we're rethinking how models reason. Tired of verbose, slow Chain-of-Thought (CoT) outputs that flood your screen with self-talk? So are we.
Kai-30B-Instruct scales up our Adaptive Dual-Search Distillation (ADS) framework. By bridging classical A* heuristic search with continuous gradient descent , we use an information-theoretic log-barrier to physically prune high-entropy reasoning paths during training.
The result? Pure implicit reasoning. The model executes structured logic, arithmetic carries, and branch selections as a reflex in a single forward pass—no external scaffolding required.
At 3B, we observed a phase transition where the model achieved "logical crystallization". Now, at 30B, we are giving the ADS regularizer the massive representational capacity it needs to tackle higher-order symbolic abstractions and complex reasoning tasks.
🧪 Test Kai yourself in our new Space:
NoesisLab/Kai-30B-Instruct
📦 Model Weights:
NoesisLab/Kai-30B-Instruct
Bring your hardest math, logic, and coding benchmarks. We invite the community to stress-test the limits of the penalty wall! 🧱💥
- 1 reply
published a dataset 11 days ago