Upload README.md

README.md

@@ -6,7 +6,6 @@ base_model: HuggingFaceTB/SmolLM3-3B
 tags:
 - smollm
 - smolreasoner
-- lora
 - reasoning
 - instruction-tuned
 - arcade
@@ -16,12 +15,18 @@ pipeline_tag: text-generation
 
 # Arcade-3B — SmolReasoner
 
+[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+[![Base Model](https://img.shields.io/badge/Base-SmolLM3--3B-orange)](https://huggingface.co/HuggingFaceTB/SmolLM3-3B)
+[![NoesisLab](https://img.shields.io/badge/Lab-NoesisLab-purple)](https://huggingface.co/NoesisLab)
+[![GSM8K](https://img.shields.io/badge/GSM8K-62.9%25-brightgreen)](https://huggingface.co/NoesisLab/Arcade-3B)
+[![ARC-Easy](https://img.shields.io/badge/ARC--Easy-74.4%25-brightgreen)](https://huggingface.co/NoesisLab/Arcade-3B)
+
 **Arcade-3B** is a 3B instruction-following and reasoning model built on [SmolLM3-3B](https://huggingface.co/HuggingFaceTB/SmolLM3-3B).
-It is the first public release from the **ARCADE** project at [NoesisLab](https://huggingface.co/NoesisLab), which investigates zero-extra-parameter fine-tuning via the *State–Constraint Orthogonality Hypothesis*.
+It is the first public release from the **ARCADE** project at [NoesisLab](https://huggingface.co/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.
 
 ---
 
-## Method: SC-Orthogonal LoRA
+## Method: SC-Orthogonal Training
 
 Standard Transformer hidden states conflate two distinct functions:
 
@@ -30,11 +35,11 @@ Standard Transformer hidden states conflate two distinct functions:
 | `H[..., :D/2]` | **S** (State) | *What* the model knows — factual content |
 | `H[..., D/2:]` | **C** (Constraint) | *How* to retrieve it — reasoning structure |
 
-ARCADE's **SCOrthoTrainer** injects an orthogonality penalty on the final hidden layer during LoRA fine-tuning, encouraging S and C to decouple in representation space without modifying any attention operators:
+ARCADE's **SCOrthoTrainer** injects an orthogonality penalty on the final hidden layer, encouraging S and C to decouple in representation space without modifying any attention operators:
 
 $$\mathcal{L}_{\text{total}} = \mathcal{L}_{\text{CE}} + \frac{\lambda}{B \cdot L} \sum_{b,l} \left( \mathbf{S}_{b,l} \cdot \mathbf{C}_{b,l} \right)^2$$
 
-with **λ = 0.1**.  This "soft logic gate" reduces divergence errors at inference time at zero architectural cost.
+with **λ = 0.1**. This soft regularization reduces divergence errors at inference time at zero architectural cost.
 
 ---
 
@@ -43,9 +48,6 @@ with **λ = 0.1**.  This "soft logic gate" reduces divergence errors at inferenc
 | Setting | Value |
 |---------|-------|
 | Base model | `HuggingFaceTB/SmolLM3-3B` |
-| LoRA rank / alpha | 64 / 128 |
-| LoRA target | all-linear |
-| Dropout | 0.05 |
 | λ (orth penalty) | 0.1 |
 | Max sequence length | 2048 |
 | Learning rate | 2e-4 (cosine) |
@@ -109,7 +111,7 @@ For step-by-step reasoning, the model may emit a `<think>…</think>` block befo
 
 ```bibtex
 @misc{noesislab2025arcade,
-  title        = {ARCADE: State-Constraint Orthogonal LoRA Fine-Tuning},
+  title        = {ARCADE: State-Constraint Orthogonal Training},
   author       = {NoesisLab},
   year         = {2025},
   howpublished = {\url{https://huggingface.co/NoesisLab/Arcade-3B}},