How labmlai/annotated_deep_learning_paper_implementations works
labml_nn is a collection of annotated PyTorch implementations of neural networks and related algorithms, rather than one application with a single runtime. The principal workflow is to choose an implementation for a learning or generation task, provide task data or prompts, and use its model or algorithm to produce predictions, generated content, or learned strategies. The README documents broad capability areas; the tree supports their module locations, but sampled excerpts do not establish most cross-module call relationships.
Main entry points: transformers Transformer Variants · neox GPT-NeoX · lora LoRA · sampling Text Sampling
Components 21
Language Models
- Transformer Variants
labml_nn/transformers - GPT-NeoX
labml_nn/neox - LoRA
labml_nn/lora - Text Sampling
labml_nn/sampling
Generative Models
- Diffusion Models
labml_nn/diffusion - GAN Variants
labml_nn/gan - Sketch RNN
labml_nn/sketch_rnn/__init__.py
Learning Algorithms
- Reinforcement Learning
labml_nn/rl - Game Strategy CFR
labml_nn/cfr - Model Distillation
labml_nn/distillation - Scalable Training
labml_nn/scaling
Model Building Blocks
- Vision Models
labml_nn - Graph Networks
labml_nn/graphs - Recurrent Networks
labml_nn - Activations
labml_nn/activations - Normalization Layers
labml_nn/normalization
Shared Support
- Optimizers
labml_nn/optimizers - Training Helpers
labml_nn/helpers
Other components
- Researcher
- Task Data or Prompt
- Model or Algorithm Results
Connections 33
- Researcher → Task Data or Prompt: defines task
- Task Data or Prompt → Transformer Variants: selects implementation
- Task Data or Prompt → GPT-NeoX: selects implementation
- Task Data or Prompt → LoRA: selects implementation
- Task Data or Prompt → Text Sampling: selects method
- Task Data or Prompt → Diffusion Models: selects implementation
- Task Data or Prompt → GAN Variants: selects implementation
- Task Data or Prompt → Sketch RNN: selects implementation
- Task Data or Prompt → Reinforcement Learning: selects algorithm
- Task Data or Prompt → Game Strategy CFR: selects algorithm
- Task Data or Prompt → Model Distillation: selects method
- Task Data or Prompt → Scalable Training: selects technique
- Task Data or Prompt → Vision Models: selects implementation
- Task Data or Prompt → Graph Networks: selects implementation
- Task Data or Prompt → Recurrent Networks: selects implementation
- Task Data or Prompt → Activations: selects component
- Task Data or Prompt → Normalization Layers: selects component
- Transformer Variants → Model or Algorithm Results: produces predictions
- GPT-NeoX → Model or Algorithm Results: generates or evaluates
- LoRA → Model or Algorithm Results: adapts model
- Text Sampling → Model or Algorithm Results: selects tokens
- Diffusion Models → Model or Algorithm Results: generates images
- GAN Variants → Model or Algorithm Results: generates images
- Sketch RNN → Model or Algorithm Results: generates sketches
- Reinforcement Learning → Model or Algorithm Results: learns policies
- Game Strategy CFR → Model or Algorithm Results: learns strategies
- Model Distillation → Model or Algorithm Results: trains smaller models
- Scalable Training → Model or Algorithm Results: enables scaled runs
- Vision Models → Model or Algorithm Results: produces predictions
- Graph Networks → Model or Algorithm Results: produces predictions
- Recurrent Networks → Model or Algorithm Results: produces predictions
- Activations → Model or Algorithm Results: transforms features
- Normalization Layers → Model or Algorithm Results: normalizes features