the-void / aether-server.mjs

Commit History

perf: cannon-rotated WASM SIMD kernels (14KB→22KB)
8f4ed48

Taylor commited on

perf: raise token limits to 256 (both PyTorch and Aether)
ec694f7

Taylor commited on

feat: Aether v2 with RoPE fix -- PyTorch vs Aether side by side
fcac5c7

Taylor commited on

fix: remove missing WASM flashAttention + stream results independently
e5e1d2b

Taylor commited on

fix: handle WASM OOM for LM head + show error details
90bf42d

Taylor commited on

fix: use Q8_0 GGUF + type-aware dequantization
5b5a680

Taylor commited on

perf: add WASM SIMD kernels + use Q4_K_M for faster inference
7336fde

Taylor commited on

feat: PyTorch vs Aether side-by-side inference
c92238b

Taylor commited on