slexisvn/mlfw

How slexisvn/mlfw works

MLFW is a TypeScript machine-learning framework with tensor operations, neural-network and training APIs, and a multi-target compiler/runtime; the tree also contains a native Tera compiler and a compiler visualizer. The main execution path is tensor work through graph and tensor-level compilation, backend code generation, and runtime execution. Eager/autograd, training/data, and higher-level ML utilities are distinct capabilities. README content was absent, and source excerpts were selective, so uncertain call wiring is omitted; paths for unsampled subsystems identify only their existing directories.

Main entry points: ops.ts Tensor Operations · index.ts Public Entry · graph_pipeline.ts Graph Pipeline · tir_pipeline.ts TIR Pipeline

Components 21

ML APIs

Public Entry src/index.ts
Neural Networks src/nn/index.ts
ML Utilities src/ml/index.ts

Tensor Runtime

Tensor Core src/tensor/core/tensor.ts
Tensor Storage src/tensor/core/storage.ts
Tensor Operations src/tensor/ops/ops.ts
Operator Dispatch src/dispatcher/dispatcher.ts
Backward Engine src/autograd/engine.ts

Compiler

Program Tracing src/tracing/trace_core.ts
Graph Pipeline src/compiler/pipeline/graph_pipeline.ts
TIR Pipeline src/compiler/pipeline/tir_pipeline.ts
Compiler Analysis src/compiler/analysis/analysis_manager.ts
Compiler Visualizer tools/visualizer/src/App.tsx

Execution Backends

Backend Codegen src/backend/pipeline.ts
Runtime Selection src/runtime/runtime.ts
Device Kernels src/kernels/defs/host_ops.ts
Tera Native Backend terac/lib/Tera/Pipelines/Pipelines.cpp

Training And ML

Data Loading src/data/index.ts
Trainer src/lightning/core/trainer.ts
Optimizers src/optim/index.ts

Other components

Framework User

Connections 23

Framework User → Public Entry: uses
Public Entry → Neural Networks: exposes
Public Entry → ML Utilities: exposes
Framework User → Tensor Core: creates tensors
Neural Networks → Tensor Operations: uses operations
ML Utilities → Tensor Operations: uses operations
Tensor Core → Tensor Storage: owns storage
Tensor Operations → Operator Dispatch: dispatches operators
Backward Engine → Tensor Operations: computes gradients
Framework User → Program Tracing: traces programs
Program Tracing → Graph Pipeline: builds graph IR
Graph Pipeline → TIR Pipeline: lowers IR
Graph Pipeline → Compiler Analysis: queries analyses
TIR Pipeline → Compiler Analysis: uses analyses
TIR Pipeline → Backend Codegen: produces functions
Backend Codegen → Runtime Selection: provides kernels
Runtime Selection → Device Kernels: selects implementations
Backend Codegen → Tera Native Backend: optional native path
Framework User → Data Loading: loads batches
Data Loading → Trainer: supplies data
Trainer → Neural Networks: runs modules
Trainer → Optimizers: updates parameters
Framework User → Compiler Visualizer: inspects compilation

More TypeScript diagrams: imagekit-developer/imagekit-video-player · jeremyarancio/invoice-reader-app · kelvinkn17/skills · lebbitheplow/open-game-master · londonstrategicedge/lse-terminal

Drawn by GitDiagram on September 26, 2026 from the repository's file tree, README and source files. Read as Markdown ·