Disentangling Gradients Recursive Reasoning

--- license: mit language: - en tags: - recursive-reasoning - sudoku - transformer - reasoning - gradient-methods - bptt pretty_name: Disentangling Gradients in Recursive Reasoning --- # Disentangling Gradient Quality from Architecture in Recursive Reasoning Models Two TRM model variants trained on Sudoku-Extreme to isolate the effect of gradient method from architecture design. ## Models | Variant | Gradient Method | Token Accuracy | Exact Accuracy | |---------|----------------|---------------|----------------| | `trm_fullbp` | Full BPTT (O(T) memory) | 71.6% | **18.9%** | | `trm_1step` | 1-step gradient (O(1) memory) | 65.9% | **2.2%** | Both share the same flat TRM architecture. The only difference is the gradient method. ## Architecture - **Base:** [TinyRecursiveModels](https://github.com/SamsungSAILMontreal/TinyRecursiveModels) (TRM) - **Hidden size:** 384 - **Layers:** 2 - **Attention heads:** 8 - **H-cycles:** 3, L-cycles: 6 - **Parameters:** ~7M ## Training - **Dataset:** [Sudoku-Extreme](https://huggingface.co/datasets/sapientinc/sudoku-extreme) — 500 puzzles + 500 augmentations - **Epochs:** 10,000 - **Batch size:** 512 (256 per GPU) - **Optimizer:** AdamW (lr=1e-4, β1=0.9, β2=0.95) - **Hardware:** 2× NVIDIA Tesla T4 (16GB) ## Usage This model requires the TRM codebase from [SamsungSAILMontreal/TinyRecursiveModels](https://github.com/SamsungSAILMontreal/TinyRecursiveModels). ```python import torch # Load checkpoint ckpt = torch.load("trm_fullbp/step_9764", map_location="cpu") # Load into TRM model (requires TinyRecursiveModels codebase) # model = TinyRecursiveReasoningModel_ACTV1(config) # model.load_state_dict(ckpt, strict=False) ``` ## Paper [Disentangling Gradient Quality from Architecture in Recursive Reasoning Models](https://doi.org/10.5281/zenodo.20712090) | [Code](https://github.com/heyiamjj/disentangling-gradients-recursive-reasoning) **Abstract:** We disentangle gradient quality from architecture design in recursive reasoning models by running TRM's flat architecture with HRM's 1-step gradient approximation. Our results show gradient quality, not architecture, is the dominant factor explaining the performance gap. ## Citation ```bibtex @article{jj2026disentangling, title={Disentangling Gradient Quality from Architecture in Recursive Reasoning Models}, author={Jatin (JJ)}, doi={10.5281/zenodo.20712090}, year={2026} } ```