Skip to content

v0.4.1 — documentation stabilization

Choose a tag to compare

View all tags
@scttfrdmn scttfrdmn released this 13 Apr 02:16
· 85 commits to main since this release
v0.4.1
8576b08

Patch release. Documentation drift against the v0.4.0 code fixed; no functional changes.

  • trnblas.__version__ was stuck at 0.3.0, now tracks the current release.
  • Installation docs: new [pyscf] extra section, TRNBLAS_REQUIRE_NKI env var table, updated neuronxcc >= 2.24 pin.
  • NKI API docs: GEMM HBM padding + measured per-call timings; new nki_batched_gemm and nki_mp2_energy sections.
  • Architecture page: current Level 3 coverage state with issue cross-references.
  • Benchmarks page: TBD placeholders replaced with measured trn1.2xlarge numbers.
  • Index: PySCF real-molecule demo pointer.

See CHANGELOG.

⚠️ Erratum (v0.4.3): The benchmark tables this release pointed at on the docs site attributed trn1 numbers to NKI; they were actually trn1 Xeon torch.matmul. Fixed in v0.4.3.

Assets 2
Loading