All writing
WRITING ARCHIVE
Browse ShivasNotes from newest to oldest. Public posts remain open; member previews still follow their existing access rules.
Show
entries
| Article | Published | Note |
|---|---|---|
| How Kokoro Uses Dilated Convolution | Oct 16, 2026 | 5993 |
| Backpropagation Through Convolution: Shared Weights, Overlapping Gradients | Oct 15, 2026 | 5992 |
| Convolution Is Wx+b Moving Across a Signal | Oct 14, 2026 | 5991 |
| From Generated C to Useful Agents: CKE Serving Qwen Code and Gemma4 | Oct 13, 2026 | 5990 |
| CKE Vision: Find the First Bad Edge | Oct 11, 2026 | 5989 |
| CKE's First Kokoro WAV Is Not Ready Yet | Oct 11, 2026 | 5988 |
| What Two Requests Actually Mean for CKE Serving | Oct 11, 2026 | 5987 |
| What Changed in C-Kernel-Engine This September | Oct 3, 2026 | 5986 |
| ALBERT: How Kokoro's Phoneme Encoder Works, and Where CKE Has Reached It | Oct 2, 2026 | 5985 |
| Kokoro Is Reaching Generated C One Boundary at a Time | Sep 29, 2026 | 5984 |
| How Kokoro TTS Works, and Where CKE Stands | Sep 27, 2026 | 5983 |
| I Want To Use My Linux Workstation Without Looking At It | Sep 27, 2026 | 5982 |
| From Native Kernels to Generated Audio Circuits: Parakeet and Cohere in CKE | Sep 23, 2026 | 5981 |
| How Ericsson Edge Routers Shaped CKE | Sep 19, 2026 | 5980 |
| How to Get Started With C-Kernel-Engine on Linux | Sep 13, 2026 | 5979 |
| Beyond Whisper: CKE Runs Parakeet and Cohere Across 42 Minutes on CPUs | Sep 13, 2026 | 5978 |
| What CKE Fixed Today: Honest Evidence and Faster CPU Whisper | Sep 9, 2026 | 5977 |
| How CKE Brought Up Qwen3.8 Flash Next | Sep 6, 2026 | 5976 |
| Qwen3.8 Broke In CKE. Our Regression Tests Needed To Catch It. | Sep 5, 2026 | 5975 |
| I Used CKE To Transcribe My Video. It Found A Bug In My Compiler. | Sep 5, 2026 | 5974 |
| How Audio Becomes Words: Whisper's Kernels, Step By Step, On A CPU | Sep 2, 2026 | 5973 |
| Today I Made My Drone Simulation Take Off From A Gamepad | Aug 29, 2026 | 5972 |
| CKE Now Runs More Than Qwen: Cohere, Laguna, Instella, Nemotron And More | Aug 26, 2026 | 5971 |
| One Ryzen, 128K Context, And A 27B Model: What Qwen3.8 Generated With CKE | Aug 25, 2026 | 5970 |
| CKE Added Qwen3.8 In About An Hour. The Architecture Is Starting To Pay Off | Aug 23, 2026 | 5969 |
| CKE Zip: From Splitting Tokens To Splitting Qwen3.5 Weights | Aug 22, 2026 | 5968 |
| I Spent CAD $9,575 Building CKE's CPU AI Lab | Aug 22, 2026 | 5967 |
| Training vs Inference in CKE: One Interface Discipline For Both Lanes | Aug 21, 2026 | 5966 |
| The Qwen3-VL CPU Gap vs llama.cpp: A Build Log That Keeps Getting Measured | Aug 20, 2026 | 5965 |
| How CKE Picks a Kernel Without a Single if-Statement: v8 Provider Selection | Aug 12, 2026 | 5964 |
| Antsand vs WordPress, Webflow, Ghost, And Substack: Five Different Platform Contracts | Aug 9, 2026 | 5963 |
| What Is Antsand? Why It Took 14 Years and Where It Is Going | Aug 8, 2026 | 5962 |
| Linux Memory Allocation: malloc, mmap, Pages, TLBs, and Huge Pages | Aug 6, 2026 | 5961 |
| Training vs Inference in CKE: One Interface Discipline For Both Lanes | Aug 5, 2026 | 5960 |
| A Green Checkmark Should Name Its Machine: How CKE's Nightly Report Became Evidence | Aug 4, 2026 | 5959 |
| LinuxUtilities: My Personal Linux Control Layer For AI, Publishing, And C-Kernel-Engine Work | Aug 3, 2026 | 5958 |
| How CKE Made Whisper Faster In Two Days — And The Evidence That Makes It Believable | Aug 2, 2026 | 5957 |
| What CKE Built In July 2026: Vision, Audio, X-Ray And Faster CPU Inference | Aug 1, 2026 | 5956 |
| How CKE Made Qwen3.6 Prefill 2.6x Faster | Jul 31, 2026 | 5955 |
| Modelling A Quadcopter: From Four Rotor Speeds To Motion | Jul 30, 2026 | 5954 |
| How CKE X-Ray Found Qwen3.6's First Bad Circuit | Jul 29, 2026 | 5953 |
| Kimi K3 Open Weights: What Is Actually New In Its 2.8-Trillion-Parameter Architecture? | Jul 28, 2026 | 5952 |
| Memory Planning in C-Kernel-Engine | Jul 27, 2026 | 5951 |
| Teaching C-Kernel-Engine To Listen: Whisper Tiny, End To End On v8 | Jul 26, 2026 | 5950 |
| Why I Am Testing Kimi And ChatGPT Together For Real Work | Jul 25, 2026 | 5949 |
| Hugging Face Got Hacked By A Closed Model. A Closed Model Also Refused To Help Fix It. | Jul 24, 2026 | 5948 |
| What Numerical Parity Actually Requires: BF16, Quantization, ISA Drift And CKE X-Ray | Jul 23, 2026 | 5947 |
| Softmax vs Softplus: Competition, Positive Gates, and Backpropagation | Jul 22, 2026 | 5946 |
| GLM4 Bring-Up In CKE: What A New Model Family Actually Requires | Jul 21, 2026 | 5945 |
| Qwen3.5 Hybrid Recurrent Attention: How DeltaNet State And Full Attention Share One Decoder | Jul 20, 2026 | 5944 |
| Prefill vs Decode in CKE: Why One Model Needs GEMM, GEMV, and Two Execution Plans | Jul 19, 2026 | 5943 |
| CKE Kernel Registry: How Exact C Functions And Numerical Contracts Connect | Jul 18, 2026 | 5942 |
| Where Does LLM Cross-Entropy Loss Come From? Compression, Shannon Entropy And Generalization | Jul 17, 2026 | 5941 |
| What Is A CPU PMU? How perf, VTune And Advisor Measure What A Processor Actually Did | Jul 16, 2026 | 5940 |
| What kernel.kptr_restrict=0 Actually Does: Linux Profiling, Kernel Symbols, VTune And Advisor | Jul 15, 2026 | 5939 |
| Nemotron Kernels From Inference To Backpropagation | Jul 14, 2026 | 5938 |
| Nemotron Is Not One Circuit: Mamba, Attention, And MoE Routing | Jul 13, 2026 | 5937 |
| Gemma4-VL In CKE: How Images Become Decoder Tokens | Jul 12, 2026 | 5936 |
| Gemma4 In CKE: Four Attention Paths, Shared KV, And Sliding Windows | Jul 11, 2026 | 5935 |
| Templates Are Circuit Maps: How CKE Describes A Model Family | Jul 10, 2026 | 5934 |
| GGUF: The File Format That Made Local LLMs Practical | Jul 9, 2026 | 5933 |
| Amdahl's Law, Theory Of Constraints, And C-Kernel-Engine Optimization | Jul 8, 2026 | 5932 |
| DeepSeek-V4 Architecture From Kernels: mHC, CSA, HCA, DSA, and C-Kernel-Engine | Jul 7, 2026 | 5931 |
| MOSFET DUT Characterization Pipeline: Rebuilding My Drone ESC Work From Measurements | Jul 7, 2026 | 5930 |
| How the Jacobian Matrix Works | Jul 6, 2026 | 5929 |
| How Audio Transformers Work: The Encoder Path, Whisper, Timestamps, And Why Audio Is Not A VLM Patch | Jul 5, 2026 | 5928 |
| DeltaNet vs SSM: Two Fixed-Size Memory Systems Beyond Attention | Jul 4, 2026 | 5927 |
| Nemotron Architecture From A C-Kernel-Engine Runtime Perspective | Jul 3, 2026 | 5926 |
| K-Quants Deep Dive: Q4_K, Q5_K, Q6_K, Q8_K And Mixed Dot Products | Jul 2, 2026 | 5925 |
| Threadpools And Memory Pools: Why CKE Needs Runtime Ownership For CPU AI Kernels | Jul 1, 2026 | 5924 |
| Pipeline vs Tensor Parallelism: How CKE Splits AI Across CPU Nodes | Jun 30, 2026 | 5923 |
| Distributed CPU AI: MPI, RDMA, NUMA, and C-Kernel-Engine | Jun 29, 2026 | 5922 |
| Muon Optimizer: SGD vs AdamW vs Matrix-Aware Training Updates | Jun 28, 2026 | 5921 |
| Kernel Engineers Are Worth Billions In 2026: Chris Lattner, Mojo, LLVM, and the Compiler Layer of AI | Jun 27, 2026 | 5920 |
| Linux System Programming For AI Kernels: Core Pinning, Huge Pages, TLBs, NUMA, and Memory Discipline | Jun 26, 2026 | 5919 |
| KV Cache Memory: The Hidden State That Makes LLM Decode Work | Jun 25, 2026 | 5918 |
| v8 IR Pipeline Codegen: How CKE Hardens Pure-C Inference | Jun 24, 2026 | 5917 |
| AI Kernels You Should Know To Learn Deep Neural Networks | Jun 23, 2026 | 5916 |
| CPU Performance Engineering for AI: Rooflines, Flamegraphs, VTune, and Perf Gates | Jun 22, 2026 | 5915 |
| Flash Attention on CPU: Online Softmax, Cache Discipline, and C-Kernel-Engine | Jun 21, 2026 | 5914 |
| Quantization Deep Dive: How CPU Kernels Compress Weights and Preserve Accuracy | Jun 20, 2026 | 5913 |
| ARM NEON in C-Kernel-Engine: Real Quantized Kernels on ARM | Jun 19, 2026 | 5912 |
| SIMD Deep Dive: How AI Kernels Use SSE, AVX, AVX-512, and VNNI | Jun 18, 2026 | 5911 |
| How Qwen3-VL Vision Works: Header Body Footer | Jun 17, 2026 | 5910 |
| Gated DeltaNet: Recurrent Attention With Error-Driven Memory | Jun 16, 2026 | 5909 |
| Optimizers: From SGD to AdamW | Jun 15, 2026 | 5908 |
| AI Kernel Engineer Beginner Guide: Math, Linear Algebra, C/Linux | Jun 14, 2026 | 5907 |
| Thread Pools in C: How CPU Runtimes Dispatch Work Across Cores | Jun 13, 2026 | 5906 |
| dL/d(LLM): The Full Backward Pass | Jun 12, 2026 | 5905 |
| Residual Connections: The Gradient Highway | Jun 11, 2026 | 5904 |
| Attention: The Core Of The Transformer | Jun 10, 2026 | 5903 |
| Tokenization: The First Decision That Shapes Everything | Jun 9, 2026 | 5902 |
| Positional Encoding: Teaching Transformers Where To Look | Jun 8, 2026 | 5901 |
| LayerNorm And RMSNorm: Stabilizing The Signal | Jun 7, 2026 | 5900 |
| Activation Functions: The Nonlinearity Inside Neural Networks | Jun 6, 2026 | 5899 |
| Softmax: The Probability Engine | Jun 5, 2026 | 5898 |
| Matrix Wx+b: From Scalars To Transformers | Jun 4, 2026 | 5897 |
| Backpropagation From wx+b: The Smallest Training Loop | Jun 3, 2026 | 5896 |
| Derivatives And The Chain Rule: The Math Of Tiny Changes | Jun 2, 2026 | 5895 |
| Why I Build Things From Scratch | May 31, 2026 | 5894 |
| Training SVGs With C Kernel Engine: A Research Report | May 31, 2026 | 5893 |
| Hardening My Linux Engineering Workflow | May 29, 2026 | 5892 |
| Quaternions, Axis-Angle, And Vector Rotation | May 28, 2026 | 5891 |
| Flight Controller for Drones | May 27, 2026 | 5890 |
| What Is the C Kernel Engine? | May 26, 2026 | 5889 |
| 14 Ways React Brain Rot Infuriates Me and the Cost of SPA Overuse | May 25, 2026 | 5888 |
| Is the Future of My Work Writing, Typing, Drawing, and Orchestrating? | May 24, 2026 | 5887 |
| AI Is Teaching Everyone the Command Line. So Why Are We Still Ignoring Linux? | May 23, 2026 | 5886 |
| Your Job May Change. That Still Does Not Make Optionality Easy. | May 22, 2026 | 5885 |
| Are AI Agents Good for Your Brain? | May 21, 2026 | 5884 |
| Qwen3 From a C Runtime Perspective | May 20, 2026 | 5883 |
| HTML Was Always the Interface. SVG Was Always the Diagram. | May 20, 2026 | 5882 |
| Hardening Long Projects in the Age of AI | May 15, 2026 | 5881 |
| Why I Replaced QuickBooks with Antsand — and Why I Think AI + Open Source Changes the Software Sub... | May 6, 2026 | 5880 |
| I Think Like a Compiler - But not in the way you think. | Apr 30, 2026 | 5879 |
| Why I Stopped Getting High on the Newer AI Models (And Why My Strategic Bet Is Still Consistent: CPU... | Apr 21, 2026 | 5878 |
| My two-tier A.I workflow: frontier thinks, MiniMax (open weights) does the heavy lifting | Feb 5, 2026 | 5877 |
| Why I’m Doubling Down on C in 2026 | Jan 2, 2026 | 5876 |
| 2025: Fighting Banks, Writing C, and The Great Untangling | Dec 31, 2025 | 5875 |
| The Great Untangling: Why I Ditched Vue, Node, and Microservices for Pure JS, PHP, and AI | Dec 31, 2025 | 5874 |
| Unimporting PyTorch: How Constraint and Curiosity Built a C Kernel Engine | Dec 23, 2025 | 5872 |
| My 2025 Roadmap: Building Next-Generation Drones and Robotics | Jan 1, 2025 | 5870 |
| Reflecting on 2024 | Dec 31, 2024 | 5869 |
| Reflecting on 2023: A Year in Review | Jan 1, 2024 | 5866 |
| Year end review 2022 | Jan 2, 2023 | 5854 |
| Year end review 2021 | Jan 1, 2022 | 5843 |
| The 3rd Week of January | May 21, 2021 | 5841 |
| Client work and system identification | Jan 13, 2021 | 5839 |
| Understaning the Kalman Filter | Jan 11, 2021 | 5838 |
| Headache and eye hurt | Jan 7, 2021 | 5837 |
| The importance of priority #1. What is ANTSAND and why is it being built? | Jan 6, 2021 | 5836 |
| Progress on Drones | Jan 6, 2021 | 5835 |
| 3rd day of the year | Jan 4, 2021 | 5834 |
| The 2nd day of the year | Jan 3, 2021 | 5833 |
| 1st Jan 2021 | Jan 2, 2021 | 5832 |
| My year end review 2020 | Dec 31, 2020 | 5831 |
| A cooking day | Dec 31, 2020 | 5830 |
| Working on ANTSAND | Dec 30, 2020 | 5829 |
| A lazy day | Dec 29, 2020 | 5828 |
| Root Vegetables | Dec 28, 2020 | 5827 |
| Getting the acclerometer working | Dec 7, 2020 | 5823 |
| I2C and learn how to troubleshoot | Dec 5, 2020 | 5822 |
| Learning Kalman Filtering ... | Dec 4, 2020 | 5821 |
| Recovering ... but bloated | Dec 3, 2020 | 5820 |
| Recovering ... | Dec 2, 2020 | 5818 |
| Little Tension and a christmas tree | Nov 30, 2020 | 5817 |
| let the progress begin ... | Nov 29, 2020 | 5816 |
| Another slow day | Nov 28, 2020 | 5815 |
| The day of ups and downs | Nov 27, 2020 | 5814 |
| Let us get moving | Nov 26, 2020 | 5813 |
| More work .. but progressing | Nov 25, 2020 | 5812 |
| Drone Specification ... | Nov 24, 2020 | 5811 |
| Getting the radio to work ... | Nov 23, 2020 | 5810 |
| Another Day ... | Nov 22, 2020 | 5809 |
| Let's get the drones sketched | Nov 21, 2020 | 5808 |
| A dumb A.I. | Nov 20, 2020 | 5806 |
| Getting Closer | Nov 19, 2020 | 5805 |
| Getting the radio working - Work in progress | Nov 18, 2020 | 5804 |
| Let the training Begin | Nov 17, 2020 | 5803 |
| Back Propogation | Nov 16, 2020 | 5802 |
| Neural Networks and a so so day ... | Nov 14, 2020 | 5801 |
| The Radio | Nov 14, 2020 | 5800 |
| Client work | Nov 13, 2020 | 5799 |
| Machine Learning progress | Nov 12, 2020 | 5798 |
| The Starry night | Nov 11, 2020 | 5797 |
| logistic regression | Nov 10, 2020 | 5796 |
| The pivot | Nov 9, 2020 | 5795 |
| Headaches and a dud day | Nov 8, 2020 | 5794 |
| the Dilemma | Nov 7, 2020 | 5793 |
| Just another day | Nov 6, 2020 | 5792 |
| A sore body and a semi-productive day | Nov 5, 2020 | 5791 |
| Let the exercise begin | Nov 3, 2020 | 5790 |
| Day 2 of my Journaling | Nov 3, 2020 | 5789 |
| Blog writing and habit tracking | Nov 2, 2020 | 5785 |
| 39+ mental tools I use to make better decisions | Oct 1, 2018 | 5644 |
| 13 things I learnt after doing the 10-week wim-hof method | Apr 2, 2017 | 5591 |
Showing 1 to 25 of 176 entries
ShivasNotes
Independent notes on software, systems, AI workflows, robotics, and building in public.
