GLOBAL INTENT COMPANY
Interactive technical environment for machine learning, scientific instrumentation, computational systems, experiments, and project evidence.
Developed independently by founder and researcher Cory Tortorici. Source code, controlled experiments, null results, architectural iterations, and public research logs are made inspectable across all project nodes.
LATEST FROM THE LAB
Recent empirical findings, architectural iterations, and null results surfaced directly from active research logs.
Active laboratory research log entries with timestamped records are available in the full research log.
EXPLORE THE RESEARCH
Navigate the connected network of research projects, architectural pipelines, and domain clusters in real-time 3D.
BNLM
GI-Company/BNLMA small decoder-only Transformer initialized, trained, and executed for inference entirely client-side inside the browser. Written in native JavaScript ES modules and WebGPU WGSL compute shaders with CPU fallback, featuring a custom reverse-mode autograd tensor engine, BPE tokenizer, Adam/AdamW optimizers, and non-blocking Web Worker execution with zero server dependencies, Python runtimes, or build steps.
Structural Overview
BNLM implements a complete deep learning training and inference pipeline running directly in the browser runtime. The tensor library provides reverse-mode automatic differentiation, matrix operations are accelerated via a custom WGSL compute shader on WebGPU with CPU fallback, and asynchronous Web Workers prevent UI thread blocking during training passes.
Canonical Source Artifact
Canonical public source repository: Decoder-only Transformer trained in-browser via JavaScript and WebGPU compute shaders.
Design Invariants & Principles
System Components
Custom tensor library supporting dynamic computational graph construction and reverse-mode automatic differentiation (src/tensor.js).
Dedicated WebGPU compute shader dispatching tiled 2D matrix multiplications on GPU hardware with automatic CPU fallback (src/webgpu.js).
Client-side Byte-Pair Encoding tokenizer supporting vocabulary training and byte-level subword segmentation (src/bpe_tokenizer.js).
Samples training windows from TinyStories-formatted text (<|endoftext|>) without cross-story context bleeding (src/dataset.js).
Offloads gradient accumulation and parameter optimization to background threads, maintaining responsive 60 FPS UI rendering (src/worker_train.js).
Implements Adam and AdamW optimizers with decoupled weight decay, learning rate schedules, and gradient clipping (src/optim.js).