Feature flags of AxonML crate // Lib.rs

Cargo Features

[dependencies]
axonml = { version = "0.6.0", default-features = false, features = ["core", "nn", "data", "vision", "text", "audio", "distributed", "profile", "llm", "jit", "onnx", "nccl", "cuda", "cudnn", "wgpu", "full"] }

default = full

The full feature is set by default whenever axonml is added without default-features = false somewhere in the dependency tree.

core data? full jit? nn? onnx? profile?

Core features (always included)

Enables axonml-autograd, axonml-core, and axonml-tensor

Affects trainer::compute_accuracy, benchmark::warmup_model, benchmark::benchmark_model, benchmark::benchmark_model_named, benchmark::compare_models, benchmark::throughput_test…

nn audio? distributed? full llm? text? vision? = core

Neural network features

Enables axonml-nn and axonml-optim

Affects adversarial::AdversarialTrainer, adversarial::fgsm_attack, adversarial::pgd_attack, adversarial::adversarial_training_step, benchmark::profile_model_memory, axonml::hvac, trainer::clip_grad_norm, benchmark::warmup_model, benchmark::benchmark_model, benchmark::benchmark_model_named, benchmark::compare_models, benchmark::throughput_test…

data audio? full text? vision? = core

Data loading

Enables axonml-data

vision full = data, nn

Vision processing

Enables axonml-vision

Affects hub::list_all_models, hub::search_models, hub::models_by_category, hub::models_by_max_size_mb, hub::models_by_max_params, hub::recommended_models…

text full = data, nn

Text processing

Enables axonml-text

audio full = data, nn

Audio processing

Enables axonml-audio

distributed full nccl? = nn

Distributed training

Enables axonml-distributed

profile full = core

Profiling tools

Enables axonml-profile

llm full = nn

LLM architectures (BERT, GPT-2)

Enables axonml-llm

Affects hub::list_all_models, hub::search_models, hub::models_by_category, hub::models_by_max_size_mb, hub::models_by_max_params, hub::recommended_models…

jit full = core