mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-20 23:08:29 +00:00
Decode 8.5->24.5 t/s (0.98x llama, at the BW wall via lane-strided unified qmatvec_core<WTYPE>); unified quant compute core (6 native + IQ/ternary/FP4 bit-exact, one core); native MoE ~2x; ROCm decode-graph machinery; Metal build fixes (offset kernels, exhaustive match, borrow, dep).
4 lines
262 B
Rust
4 lines
262 B
Rust
// Relocated to hanzo-cli/tests/qmmq_numeric.rs — that crate links the HIP runtime via its build.rs
|
|
// and the engine workspace resolves cleanly (the ml workspace has wasm-only members that break
|
|
// native cargo resolution). Kept here as an empty placeholder.
|