corpus.blog
41554 blogs · [ { "id": "01a08799-abe2-72b8-80c8-eae8d873edf9", "title": "LoRA-Muon-OGD: Spectral Orthogonal Gradient Projection on the Low-Rank Manifold for LLM Continual Learning", "url": "https://leloykun.github.io/ponder/lora-muon-ogd/", "published_at": "2026-06-25T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8d8dffe84", "title": "LoRA-Muon: Spectral Steepest Descent on the Low-Rank Manifold", "url": "https://leloykun.github.io/papers/lora-muon/", "published_at": "2026-06-11T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8d9214d22", "title": "Lean4-TileLang Tensor Program Superoptimizer [WIP]", "url": "https://leloykun.github.io/ponder/lean4-tilelang/", "published_at": "2026-05-11T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8d95ac8a8", "title": "Frequency Domain Muon for Convolutional Neural Networks: Simplified", "url": "https://leloykun.github.io/ponder/freqmuon/", "published_at": "2026-03-06T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8d9ba47c7", "title": "LUCID-MoE: Mixture of Experts with Preconditioned Routing", "url": "https://leloykun.github.io/ponder/lucid-moe/", "published_at": "2026-02-17T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8d9e9191e", "title": "Error-Compensating Optimizers: ECO-AdamW, ECO-Muon, and Beyond", "url": "https://leloykun.github.io/ponder/eco/", "published_at": "2026-02-10T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8da5b9ef1", "title": "Bidirectional-PRISM: Kronecker-Factored Optimization via Anisotropic Spectral Shaping", "url": "https://leloykun.github.io/ponder/shampoo-prism/", "published_at": "2026-02-04T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8db4f737f", "title": "Steepest Descent on Affine-Conic Representable Manifolds with Boundary via Dual Ascent", "url": "https://leloykun.github.io/ponder/steepest-descent-affine-conic/", "published_at": "2026-01-09T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8db560849", "title": "Steepest Descent on the Birkhoff Polytope Equipped with the Spectral Norm", "url": "https://leloykun.github.io/ponder/steepest-descent-doubly-stochastic/", "published_at": "2026-01-04T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dbf18782", "title": "Convergence Bounds for Steepest Descent Under Arbitrary Norms", "url": "https://leloykun.github.io/ponder/steepest-descent-convergence/", "published_at": "2025-12-11T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dc24341c", "title": "Critical Batch Size for Steepest Descent Under Arbitrary Norms", "url": "https://leloykun.github.io/ponder/steepest-descent-crit-bz/", "published_at": "2025-11-22T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dc973425", "title": "Rethinking Maximal Update Parametrization: Steepest Descent on Finsler-Structured (Matrix) Geometries via Dual Ascent", "url": "https://leloykun.github.io/ponder/steepest-descent-finsler-dual-ascent/", "published_at": "2025-10-29T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dd64a111", "title": "Rethinking Maximal Update Parametrization: Steepest Descent on the Spectral Ball", "url": "https://leloykun.github.io/ponder/rethinking-mup-spectral-ball/", "published_at": "2025-10-15T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dd6a6680", "title": "Steepest Descent on Finsler-Structured (Matrix) Manifolds", "url": "https://leloykun.github.io/ponder/steepest-descent-finsler/", "published_at": "2025-08-20T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8de507c2c", "title": "Heuristic Solutions for Steepest Descent on the Stiefel Manifold", "url": "https://leloykun.github.io/ponder/steepest-descent-stiefel/", "published_at": "2025-07-18T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8df182044", "title": "Training Transformers with Enforced Lipschitz Bounds", "url": "https://leloykun.github.io/papers/lipschitz-transformers/", "published_at": "2025-07-17T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dfc83b17", "title": "Sensitivity and Sharpness of n-Simplicial Attention", "url": "https://leloykun.github.io/ponder/lipschitz-n-simplical-transformer/", "published_at": "2025-07-06T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8dfdd76f9", "title": "Adam with Aggressive Gradient Clipping ≈ Smoothed SignSGD/NormSGD", "url": "https://leloykun.github.io/ponder/adam-aggressive-clipping/", "published_at": "2025-07-03T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e07b67f2", "title": "Fast, Numerically Stable, and Auto-Differentiable Spectral Clipping via Newton-Schulz Iteration", "url": "https://leloykun.github.io/ponder/spectral-clipping/", "published_at": "2025-06-23T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e148c57b", "title": "Muon and a Selective Survey on Steepest Descent in Riemannian and Non-Riemannian Manifolds", "url": "https://leloykun.github.io/ponder/steepest-descent-non-riemannian/", "published_at": "2025-04-03T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e1ed4f4e", "title": "Napkin Math on Non-Euclidean Trust Region Optimization", "url": "https://leloykun.github.io/ponder/napkin-math-trust-region-opt/", "published_at": "2025-03-24T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e2198908", "title": "Block Matrix Formulation of Linear Attention Mechanisms", "url": "https://leloykun.github.io/ponder/blockmat-linear-attn/", "published_at": "2025-03-16T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e2abb018", "title": "Steepest Descent Under Schatten-p Norms", "url": "https://leloykun.github.io/ponder/steepest-descent-schatten-p/", "published_at": "2025-02-27T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e2d132f7", "title": "Squeezing 1-2% Efficiency Gains Out of Muon by Optimizing the Newton-Schulz Coefficients", "url": "https://leloykun.github.io/ponder/muon-opt-coeffs/", "published_at": "2025-02-21T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e349f853", "title": "CASPR Without Accumulation is Muon", "url": "https://leloykun.github.io/ponder/caspr-wo-accum-is-muon/", "published_at": "2025-02-13T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e3ad8420", "title": "GRPO's Main Flaw", "url": "https://leloykun.github.io/ponder/grpo-flaw/", "published_at": "2025-02-11T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e411e3bd", "title": "(Linear) Attention as Test-Time Regression", "url": "https://leloykun.github.io/ponder/test-time-regression/", "published_at": "2025-01-27T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e4c3fd63", "title": "Deep Learning Optimizers as Steepest Descent in Normed Spaces", "url": "https://leloykun.github.io/ponder/steepest-descent-opt/", "published_at": "2024-10-20T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e4f09932", "title": "Multimodal Structured Generation", "url": "https://leloykun.github.io/personal-projects/mmsg/", "published_at": "2024-07-14T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e53d9214", "title": "Multimodal Structured Generation: CVPR's 2nd MMFM Challenge Technical Report", "url": "https://leloykun.github.io/papers/mmsg/", "published_at": "2024-06-17T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e61ba89c", "title": "Flash Attention Minimal", "url": "https://leloykun.github.io/personal-projects/flash-attention-minimal/", "published_at": "2024-04-16T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e68377e3", "title": "Retrieval Augmented Structured Generation: Business Document Information Extraction As Tool Use", "url": "https://leloykun.github.io/papers/rasg/", "published_at": "2024-04-15T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e6b9e841", "title": "Llama.cpp", "url": "https://leloykun.github.io/personal-projects/llama.cpp/", "published_at": "2023-07-25T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e7627fd7", "title": "Expedock Assistant: ChatGPT Applied to Logistics Data", "url": "https://leloykun.github.io/personal-projects/expedock-assistant/", "published_at": "2023-01-31T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e7d04dbe", "title": "Expedock AutoML", "url": "https://leloykun.github.io/personal-projects/expedock-automl/", "published_at": "2022-07-25T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e84e8896", "title": "Booking Demand Prediction for Grab SEA", "url": "https://leloykun.github.io/personal-projects/grab-booking-demand-prediction/", "published_at": "2019-06-16T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e8b5997f", "title": "Codeball 2018", "url": "https://leloykun.github.io/personal-projects/codeball/", "published_at": "2019-01-24T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e9030687", "title": "Codewars 2017", "url": "https://leloykun.github.io/personal-projects/codewars/", "published_at": "2018-02-12T00:00:00+00:00" }, { "id": "01a08799-abe2-72b8-80c8-eae8e9d0ddb1", "title": "Ateneo's Competitive Programming Varsity's Code Library", "url": "https://leloykun.github.io/personal-projects/progvar-library/", "published_at": null } ] posts Claim your blog
Back to leloykun.github.io
Blog · corpus.blog/blogs/leloykun.github.io/posts

leloykun.github.io

leloykun.github.io

2026

LoRA-Muon-OGD: Spectral Orthogonal Gradient Projection on the Low-Rank Manifold for LLM Continual Learningoriginal ↗
25 Jun 2026
LoRA-Muon: Spectral Steepest Descent on the Low-Rank Manifoldoriginal ↗
11 Jun 2026
Lean4-TileLang Tensor Program Superoptimizer [WIP]original ↗
11 May 2026
Frequency Domain Muon for Convolutional Neural Networks: Simplifiedoriginal ↗
6 Mar 2026
LUCID-MoE: Mixture of Experts with Preconditioned Routingoriginal ↗
17 Feb 2026
Error-Compensating Optimizers: ECO-AdamW, ECO-Muon, and Beyondoriginal ↗
10 Feb 2026
Bidirectional-PRISM: Kronecker-Factored Optimization via Anisotropic Spectral Shapingoriginal ↗
4 Feb 2026
Steepest Descent on Affine-Conic Representable Manifolds with Boundary via Dual Ascentoriginal ↗
9 Jan 2026
Steepest Descent on the Birkhoff Polytope Equipped with the Spectral Normoriginal ↗
4 Jan 2026

2025

Convergence Bounds for Steepest Descent Under Arbitrary Normsoriginal ↗
11 Dec 2025
Critical Batch Size for Steepest Descent Under Arbitrary Normsoriginal ↗
22 Nov 2025
Rethinking Maximal Update Parametrization: Steepest Descent on Finsler-Structured (Matrix) Geometries via Dual Ascentoriginal ↗
29 Oct 2025
Rethinking Maximal Update Parametrization: Steepest Descent on the Spectral Balloriginal ↗
15 Oct 2025
Steepest Descent on Finsler-Structured (Matrix) Manifoldsoriginal ↗
20 Aug 2025
Heuristic Solutions for Steepest Descent on the Stiefel Manifoldoriginal ↗
18 Jul 2025
Training Transformers with Enforced Lipschitz Boundsoriginal ↗
17 Jul 2025
Sensitivity and Sharpness of n-Simplicial Attentionoriginal ↗
6 Jul 2025
Adam with Aggressive Gradient Clipping ≈ Smoothed SignSGD/NormSGDoriginal ↗
3 Jul 2025
Fast, Numerically Stable, and Auto-Differentiable Spectral Clipping via Newton-Schulz Iterationoriginal ↗
23 Jun 2025
Muon and a Selective Survey on Steepest Descent in Riemannian and Non-Riemannian Manifoldsoriginal ↗
3 Apr 2025
Napkin Math on Non-Euclidean Trust Region Optimizationoriginal ↗
24 Mar 2025
Block Matrix Formulation of Linear Attention Mechanismsoriginal ↗
16 Mar 2025
Steepest Descent Under Schatten-p Normsoriginal ↗
27 Feb 2025
Squeezing 1-2% Efficiency Gains Out of Muon by Optimizing the Newton-Schulz Coefficientsoriginal ↗
21 Feb 2025
CASPR Without Accumulation is Muonoriginal ↗
13 Feb 2025
GRPO's Main Flaworiginal ↗
11 Feb 2025
(Linear) Attention as Test-Time Regressionoriginal ↗
27 Jan 2025

2024

Deep Learning Optimizers as Steepest Descent in Normed Spacesoriginal ↗
20 Oct 2024
Multimodal Structured Generationoriginal ↗
14 Jul 2024
Multimodal Structured Generation: CVPR's 2nd MMFM Challenge Technical Reportoriginal ↗
17 Jun 2024
Flash Attention Minimaloriginal ↗
16 Apr 2024
Retrieval Augmented Structured Generation: Business Document Information Extraction As Tool Useoriginal ↗
15 Apr 2024

2023

Llama.cpporiginal ↗
25 Jul 2023
Expedock Assistant: ChatGPT Applied to Logistics Dataoriginal ↗
31 Jan 2023

2022

Expedock AutoMLoriginal ↗
25 Jul 2022

2019

Booking Demand Prediction for Grab SEAoriginal ↗
16 Jun 2019
Codeball 2018original ↗
24 Jan 2019

2018

Codewars 2017original ↗
12 Feb 2018

Undated

Ateneo's Competitive Programming Varsity's Code Libraryoriginal ↗
—