Apple Silicon native port (v2) — native arm64 DeepVariant/DeepTrio/DeepSomatic/pangenome#1085

Open

BenjaminDEMAILLE wants to merge 11 commits into

google:r1.10from

IPNP-BIPN:apple-silicon-native-v2-pr

Commits on Jun 24, 2026

Apple Silicon native port (v2): native arm64 DeepVariant/DeepTrio/DeepSomatic/pangenome
BenjaminDEMAILLE
committed
perf(metal): add opt-in DV_METAL_FP16 reduced-precision inference
nh13
committed
docs: clarify TF32 in reduced-precision comment
nh13
committed
fix(conversion): harden parity guards, weight-shape asserts, stale test
nh13
committed
docs: mark Core ML ADR superseded and fix conversion README inventory
nh13
committed
perf(make_examples): move reads, hoist invariants, add progress log
nh13
committed
perf(io): buffer the TFRecord reader and avoid a payload copy on write
nh13
committed
perf(build): target apple-m1 and drop dead divergent neon_base_color
nh13
committed
perf(call_variants): add opt-in GPU/CPU inference pipelining
nh13
committed
Merge pull request #3 from nh13/nh_dv-2-perf
BenjaminDEMAILLE
authored
Merge pull request #4 from nh13/nh_dv-3-fp16
BenjaminDEMAILLE
authored