Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation (NeurIPS 2025)
-
Updated
Sep 26, 2025 - Python
#
adaptive-computation
Here are 23 public repositories matching this topic...
A curated reading list of research in Adaptive Computation, Inference-Time Computation & Mixture of Experts (MoE).
nlp machine-learning computer-vision tensorflow transformers pytorch mixture-of-experts adaptive-computation
-
Updated
Jan 1, 2025
[ICLR 2025] Dynamic Mixture of Experts: An Auto-Tuning Approach for Efficient Transformer Models
-
Updated
Jul 9, 2025 - Python
Exploration into the proposed "Self Reasoning Tokens" by Felipe Bonetto
-
Updated
May 17, 2024 - Python
Yet another random morning idea to be quickly tried and architecture shared if it works; to allow the transformer to pause for any amount of time on any token
-
Updated
Oct 22, 2023 - Python
Dynamic weight generation for recursive transformers via input-conditioned LoRA modulation
-
Updated
Apr 2, 2026 - Python
(NeurIPS-2019 MicroNet Challenge - 3rd Winner) Open source code for "SIPA: A simple framework for efficient networks"
pruning model-compression micronet model-acceleration neurips-2019 micronet-challenge early-exiting adaptive-computation compact-neural-network
-
Updated
Dec 18, 2022 - Python
Temporalmesh-transformer. It is the first architecture to simultaneously fuse dynamic graph topology, token-level adaptive compute, and temporal semantic decay into a single unified model. No prior work does all three together.
nlp machine-learning natural-language-processing research deep-learning pytorch artificial-intelligence transformer language-model attention-mechanism preprint graph-neural-network adaptive-computation efficient-transformers sparse-attention
-
Updated
Aug 29, 2026 - Python
[Under review] To Think or Not to Think: Pre-Decisional Reasoning Budgets for Referring Audio-Visual Segmentation
pytorch interpretability adaptive-computation audio-visual-segmentation chain-of-thought multimodal-llm
-
Updated
Aug 27, 2026 - Python
The ARL Hierarchical MultiScale Framework (ARL-HMS) is a software library for development of multiscale models on heterogeneous high-performance computing systems.
-
Updated
Jul 18, 2024 - C++
Frozen KV Context for Mixture-of-Recursions on a Modernized BERT
research representation-learning bert encoder-model adaptive-computation masked-language-modeling efficient-ai mixture-of-recursions frozen-kv
-
Updated
Mar 31, 2026 - Jupyter Notebook
Token-level early-exit Transformer with Gumbel-Softmax routing and a latency-aware joint loss — skips redundant layers for "easy" tokens while going deep on hard ones.
nlp research deep-learning transformers pytorch gumbel-softmax model-compression efficient-inference early-exit adaptive-computation mixture-of-depths pareto-frontier
-
Updated
Aug 19, 2026 - Python
Curated papers on dynamic neural networks, pruning, growing architectures, sparse training, and adaptive computation.
deep-learning pruning adaptive-computation conditional-computation dynamic-neural-network topology-adaptation dynamic-architectures
-
Updated
Jun 6, 2026
Recursive Convergent Inference — dynamic MoE with convergence-gated stopping. Unexpected finding: model-relative complexity diverges from human difficulty labels
-
Updated
Mar 29, 2026 - Jupyter Notebook
Lightweight PyTorch implementation of Mixture-of-Recursions with Expert-Choice & Token-Choice routing | Runs on your laptop!
router neural-network python3 pytorch mor mixture-of-experts early-exiting adaptive-computation kv-cache llm recursive-transformers
-
Updated
Nov 25, 2025 - Python
42 — adaptive-depth inference: match accuracy, cut compute and energy 20-83%.
machine-learning deep-learning gpu tabular-data efficient-inference inference-optimization green-ai adaptive-computation energy-efficient-ai
-
Updated
Jul 2, 2026
42 — adaptive-depth inference
machine-learning deep-learning gpu tabular-data efficient-inference inference-optimization green-ai adaptive-computation energy-efficient-ai
-
Updated
Jul 2, 2026
Research code for testing whether causal token surprisal can guide adaptive computation, sparse refinement, and learned compute allocation in byte-level language models.
-
Updated
Aug 1, 2026 - Python
PyTorch benchmark for CTM-style adaptive computation, sparse-retrieval failure analysis, adaptive halting, and attention-supervised recovery.
benchmark machine-learning pytorch interpretability research-engineering adaptive-computation continuous-thought-machines
-
Updated
May 17, 2026 - Python
Technical report on deep supervision, recursive Transformers, and adaptive inference over large-scale process-chain telemetry.
machine-learning transformers process-mining adaptive-computation deep-supervision research-report security-telemetry
-
Updated
Jul 16, 2026
Add this topic to your repo
To associate your repository with the adaptive-computation topic, visit your repo's landing page and select "manage topics."