Correctness

Logit parity with mlx-lm (same weights, Python reference) is the project’s oracle. The test suite holds the forward pass bit-exact against it — including every quantized-KV configuration (kv8, kv4, and the 26B’s mixed per-layer scheme) and the fused quantized-attention prefill, which is bit-exact against optiq’s reference implementation.

Every ported helper follows the reference implementation’s exact op composition, down to constants built at load time. The one latent divergence ever found — rope frequencies computed host-side instead of on-device — was root-caused and fixed (see the findings log in PLAN.md). Golden files are regenerated only by explicit scripts running the Python oracle.

bun test    # fast tier runs everywhere; model-loaded tests auto-skip
            # unless the reference snapshot is in your HF cache

Why this matters

A local model is only useful if it produces the same outputs as the reference implementation it claims to run. “Close enough” quantization or a subtly different attention kernel can silently change behavior — wrong tool calls, drifting reasoning, different refusals. Holding the forward pass bit-exact means mlx-bun’s outputs are the reference’s outputs, not an approximation of them.