burn_dragon 🔥🐉🐣

burn inference and training of the dragon model

features

• cached inference

• training benchmarks and reporting

• pope positional embeddings

• wasm deployments

• sparsity metrics and visualization

• vision dragon cortical column/stack

• recurrent, foveated saccade training and inference

• GDPO

• adaptive tool discovery

• conditional (deep) gating

• document-coherent dataloading and scale mixup

• episodic memory

• fused kernels

• hierarchical, memory-aware recurrent state

• mixture-of-expert routing

• multi-modal architecture

• multi-stream truncated backpropagation through time

• neuromorphic backend

• streaming, sparse synaptic backpropagation

• temporal neuron dampening

Dataset configuration (built-in presets and Hugging Face examples) is documented inline in config/language/base.toml.

training

• cargo run -p burn_dragon_cli --release (defaults to the cuda backend)

inference

• cargo run -p burn_dragon_cli --bin infer -- --max-tokens 2048 --streaming

benchmarks

• cargo bench -p burn_dragon --features train,benchmark (executes both wgpu and cuda benchmarks)

• open target/criterion/report/index.html

compatible burn versions

license

licensed under either of

• Apache License, Version 2.0 (http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (http://opensource.org/licenses/MIT)

at your option.

contribution

unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.