Skip to content
View sehaxe's full-sized avatar
2 followers · 2 following
  • Minsk
  • 07:39 (UTC +03:00)

Block or report sehaxe

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
sehaxe/README.md

👋 sehaxe

Building Aria - a 1.58-bit ternary LLM in Rust/Burn.

🔥 Burn ecosystem

Crate Description
burn-bitnet b1.58 ternary weights + v2 Hadamard activations
burn-sct Spectral Compact Training - permanent truncated SVD
burn-gdn2 Gated DeltaNet 2 - linear recurrent token mixer
burn-msa MiniMax Sparse Attention - block-sparse top-K
burn-rope Rotary Position Embedding + YaRN extrapolation
burn-swiglu SiLU-gated FFN - the standard Transformer feed-forward

🛠 Stack

📫 Connect

Pinned Loading

  1. burn-gdn2 burn-gdn2 Public

    Gated DeltaNet 2 for Burn — linear recurrent token mixer with channel-wise erase/write gates. 2–5.5× faster than PyTorch chunk mode.

    Rust 1

  2. burn-msa burn-msa Public

    MSA for Burn - top-K KV block selection with 5D batched matmul. Up to 8.2× faster than PyTorch.

    Rust

  3. burn-sct burn-sct Public

    Spectral Compact Training for Burn — permanent truncated SVD with Stiefel QR retraction. Up to 199× memory reduction, 25–68× faster than dense.

    Rust

  4. burn-bitnet burn-bitnet Public

    BitNet quantization family for Burn - b1.58 ternary weights + v2 Hadamard activations with fast Walsh-Hadamard transform

    Rust