Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance
Paper • 2606.19195 • Published • 127
Lapo Luchini
lapo
AI & ML interests
None yet
Recent Activity
upvoted a paper 1 day ago
Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance reacted to matteospanio's post with 🚀 16 days ago
🎶 Released mule-torch — an unofficial PyTorch port of MULE (SF-NFNet-F0), SiriusXM/Pandora's music-audio embedding model (McCallum et al., ISMIR 2022).
No retraining: I re-implemented the architecture in pure PyTorch and transferred the original TensorFlow weights, then checked it layer by layer against the genuine TF pipeline.
✅ End-to-end clip-embedding cosine 0.9999999 vs the original
✅ ONNX backbone parity < 1e-6
✅ 62.35M params (paper: ~62.4M)
✅ Batched, GPU-native, ONNX-exportable — none of which the original `Analysis` pipeline does
```python
pip install mule-torch
```
```python
from mule_torch import MuleModel
emb = MuleModel.from_pretrained()(waveform) # (B, T)@16kHz -> (B, 1728)
```
🤗 Weights: https://huggingface.co/matteospanio/mule
💻 Code: https://github.com/matteospanio/mule-torch
📦 PyPI: https://pypi.org/project/mule-torch/
The fun bug: parity was perfect through every conv but the block output was anti-correlated (cos = −1). Cause: the learnable skip-init gains couldn't be mapped by layer name (Keras scrambles the order) — they had to be recovered from the graph.
⚠️ Unofficial, community port — not affiliated with or endorsed by the original authors. All credit to them; please cite the paper. Weights inherit CC-BY-NC-4.0. reacted to HannesVonEssen's post with 👍 about 2 months ago
📣 I made a visualizer for Hugging Face models: https://hfviewer.com
✨ Simply paste a Hugging Face URL to get an interactive visualization of the architecture!
🔗 The recent Qwen3.6-27B model as an example: https://hfviewer.com/Qwen/Qwen3.6-27B
Feel free to try it out and give me feedback on how it can be improved! ❤️
Organizations
None yet
upvoted a paper 1 day ago
reacted to matteospanio's post with 🚀 16 days ago
Post
7051
🎶 Released mule-torch — an unofficial PyTorch port of MULE (SF-NFNet-F0), SiriusXM/Pandora's music-audio embedding model (McCallum et al., ISMIR 2022).
No retraining: I re-implemented the architecture in pure PyTorch and transferred the original TensorFlow weights, then checked it layer by layer against the genuine TF pipeline.
✅ End-to-end clip-embedding cosine 0.9999999 vs the original
✅ ONNX backbone parity < 1e-6
✅ 62.35M params (paper: ~62.4M)
✅ Batched, GPU-native, ONNX-exportable — none of which the original
🤗 Weights: matteospanio/mule
💻 Code: https://github.com/matteospanio/mule-torch
📦 PyPI: https://pypi.org/project/mule-torch/
The fun bug: parity was perfect through every conv but the block output was anti-correlated (cos = −1). Cause: the learnable skip-init gains couldn't be mapped by layer name (Keras scrambles the order) — they had to be recovered from the graph.
⚠️ Unofficial, community port — not affiliated with or endorsed by the original authors. All credit to them; please cite the paper. Weights inherit CC-BY-NC-4.0.
No retraining: I re-implemented the architecture in pure PyTorch and transferred the original TensorFlow weights, then checked it layer by layer against the genuine TF pipeline.
✅ End-to-end clip-embedding cosine 0.9999999 vs the original
✅ ONNX backbone parity < 1e-6
✅ 62.35M params (paper: ~62.4M)
✅ Batched, GPU-native, ONNX-exportable — none of which the original
Analysis pipeline doespip install mule-torchfrom mule_torch import MuleModel
emb = MuleModel.from_pretrained()(waveform) # (B, T)@16kHz -> (B, 1728)🤗 Weights: matteospanio/mule
💻 Code: https://github.com/matteospanio/mule-torch
📦 PyPI: https://pypi.org/project/mule-torch/
The fun bug: parity was perfect through every conv but the block output was anti-correlated (cos = −1). Cause: the learnable skip-init gains couldn't be mapped by layer name (Keras scrambles the order) — they had to be recovered from the graph.
⚠️ Unofficial, community port — not affiliated with or endorsed by the original authors. All credit to them; please cite the paper. Weights inherit CC-BY-NC-4.0.
reacted to HannesVonEssen's post with 👍 about 2 months ago
Post
246
📣 I made a visualizer for Hugging Face models: https://hfviewer.com
✨ Simply paste a Hugging Face URL to get an interactive visualization of the architecture!
🔗 The recent Qwen3.6-27B model as an example: https://hfviewer.com/Qwen/Qwen3.6-27B
Feel free to try it out and give me feedback on how it can be improved! ❤️
✨ Simply paste a Hugging Face URL to get an interactive visualization of the architecture!
🔗 The recent Qwen3.6-27B model as an example: https://hfviewer.com/Qwen/Qwen3.6-27B
Feel free to try it out and give me feedback on how it can be improved! ❤️
- 1 reply
reacted to OzTianlu's post with 👍 3 months ago
Post
5414
Arcade-3B — SmolReasoner
NoesisLab/Arcade-3B
Arcade-3B is a 3B instruction-following and reasoning model built on SmolLM3-3B. It is the public release from the ARCADE project at NoesisLab, which investigates the State–Constraint Orthogonality Hypothesis: standard Transformer hidden states conflate factual content and reasoning structure in the same subspace, and explicitly decoupling them improves generalization.
NoesisLab/Arcade-3B
Arcade-3B is a 3B instruction-following and reasoning model built on SmolLM3-3B. It is the public release from the ARCADE project at NoesisLab, which investigates the State–Constraint Orthogonality Hypothesis: standard Transformer hidden states conflate factual content and reasoning structure in the same subspace, and explicitly decoupling them improves generalization.
- 5 replies
liked 2
Spaces 3 months ago
upvoted an article 3 months ago
Article
🏟️ Smol AI WorldCup: A 5-Axis Benchmark That Reveals What Small Language Models Can Really Do
FINAL-Bench
• • 38 reacted to SeaWolf-AI's post with 👍 4 months ago
Post
6135
ALL Bench Leaderboard — Structural Problems in AI Benchmarking and the Case for Unified Evaluation
FINAL-Bench/all-bench-leaderboard
The AI benchmark ecosystem has three structural problems. Major benchmarks like MMLU have surpassed 90%, losing discriminative power. Most leaderboards publish unverified self-reported scores — our cross-verification found Claude Opus 4.6's ARC-AGI-2 listed as 37.6% (actual: 68.8%), Gemini 3.1 Pro as 88.1% (actual: 77.1%). OpenAI's own audit confirmed 59.4% of SWE-bench Verified tasks are defective, yet it remains widely used.
ALL Bench addresses this by comparing 91 models across 6 modalities (LLM · VLM · Agent · Image · Video · Music) with 3-tier confidence badges (✓✓ cross-verified · ✓ single-source · ~ self-reported). Composite scoring uses a 5-Axis Framework and replaces SWE-Verified with contamination-resistant LiveCodeBench.
Key finding: metacognition is the largest blind spot. FINAL Bench shows Error Recovery explains 94.8% of self-correction variance, yet only 9 of 42 models are even measured. The 9.2-point spread (Kimi K2.5: 68.71 → rank 9: 59.5) is 3× the GPQA top-model spread, suggesting metacognition may be the single biggest differentiator among frontier models today.
VLM cross-verification revealed rank reversals — Claude Opus 4.6 leads MMMU-Pro (85.1%) while Gemini 3 Flash leads MMMU (87.6%), producing contradictory rankings between the two benchmarks.
📊 Article: https://huggingface.co/blog/FINAL-Bench/all-bench
📦 Dataset: FINAL-Bench/ALL-Bench-Leaderboard
⚡ GitHub: https://github.com/final-bench/ALL-Bench-Leaderboard
🏆 Leaderboard: FINAL-Bench/all-bench-leaderboard
🧬 FINAL Bench: FINAL-Bench/Metacognitive
FINAL-Bench/all-bench-leaderboard
The AI benchmark ecosystem has three structural problems. Major benchmarks like MMLU have surpassed 90%, losing discriminative power. Most leaderboards publish unverified self-reported scores — our cross-verification found Claude Opus 4.6's ARC-AGI-2 listed as 37.6% (actual: 68.8%), Gemini 3.1 Pro as 88.1% (actual: 77.1%). OpenAI's own audit confirmed 59.4% of SWE-bench Verified tasks are defective, yet it remains widely used.
ALL Bench addresses this by comparing 91 models across 6 modalities (LLM · VLM · Agent · Image · Video · Music) with 3-tier confidence badges (✓✓ cross-verified · ✓ single-source · ~ self-reported). Composite scoring uses a 5-Axis Framework and replaces SWE-Verified with contamination-resistant LiveCodeBench.
Key finding: metacognition is the largest blind spot. FINAL Bench shows Error Recovery explains 94.8% of self-correction variance, yet only 9 of 42 models are even measured. The 9.2-point spread (Kimi K2.5: 68.71 → rank 9: 59.5) is 3× the GPQA top-model spread, suggesting metacognition may be the single biggest differentiator among frontier models today.
VLM cross-verification revealed rank reversals — Claude Opus 4.6 leads MMMU-Pro (85.1%) while Gemini 3 Flash leads MMMU (87.6%), producing contradictory rankings between the two benchmarks.
📊 Article: https://huggingface.co/blog/FINAL-Bench/all-bench
📦 Dataset: FINAL-Bench/ALL-Bench-Leaderboard
⚡ GitHub: https://github.com/final-bench/ALL-Bench-Leaderboard
🏆 Leaderboard: FINAL-Bench/all-bench-leaderboard
🧬 FINAL Bench: FINAL-Bench/Metacognitive
liked a
model 4 months ago
reacted to OzTianlu's post with 👍 4 months ago
Post
1975
We deleted the Embedding Layer -- INTRO Our Collins-Embedding-3M
NoesisLab/Collins-Embedding-3M
Most "small" models are just giant vocab tables in a trench coat. Collins-3M changes that. By using 2-Universal Hashing and Chernoff-bound noise suppression, we’ve collapsed the embedding space into a fixed O(1) hash-map.
* STSB: 0.7114 (Beating many 100M+ models)
* Size: 3M (Edge-ready, IoT-ready)
* Tech: Randomized Sign-Hashing + RoPE positional injection.
Built by NoesisLab
NoesisLab/Collins-Embedding-3M
Most "small" models are just giant vocab tables in a trench coat. Collins-3M changes that. By using 2-Universal Hashing and Chernoff-bound noise suppression, we’ve collapsed the embedding space into a fixed O(1) hash-map.
* STSB: 0.7114 (Beating many 100M+ models)
* Size: 3M (Edge-ready, IoT-ready)
* Tech: Randomized Sign-Hashing + RoPE positional injection.
Built by NoesisLab
liked a
model 4 months ago
updated a
collection 4 months ago
upvoted an article 4 months ago
Article
GGML and llama.cpp join HF to ensure the long-term progress of Local AI
- +4
ggerganov, ngxson, allozaur, lysandre, victor, julien-c
• • 507 liked a
model 4 months ago
upvoted an article about 1 year ago
Article
Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM
- +2
ariG23498, merve, pcuenq, reach-vb
• • 497 liked a
model about 2 years ago
liked a
model over 2 years ago
liked a
Space about 4 years ago