FlagTensor Release Notes

Contents

FlagTensor Release Notes#

v0.2.0#

Note

This is a preview release. The version number shown is a pre-release identifier and may change upon final release. Content in this preview is for reference only and does not constitute a commitment or warranty for the final product.

  • Added Features

    • Unary Operators — abs, acos, acosh, asin, asinh, atan, atanh, ceil, conj, cos, cosh, exp, floor, identity, log, mish, neg, rcp, relu, sigmoid, sin, sinh, soft_plus, soft_sign, sqrt, swish, tan, tanh (28 operators).

    • Binary Operators — add, max, min, mul.

    • Contraction Operators — gett, tgett, ttgt, tensor_contraction_trinary, trinary_generic.

    • Sparse Operators — block_sparse_tensor_contraction.

    • Operator Registry — Added conf/operators.yaml with full operator metadata.

    • Multi-GPU Test Runner — tools/run_tests.py with live progress display and YAML-driven operator selection.

    • CI/CD Pipeline — Quality gates (lint/format), correctness and performance pipelines.

  • Enhanced Features

    • Hand-optimized Triton kernels with per-architecture autotune (Ampere, Hopper).

    • Vendor-agnostic backend abstraction with 15 registered vendors.

    • Architecture-specific kernel specialization (_nvidia/hopper/, _nvidia/ampere/).

    • Per-operator test infrastructure with pytest marks and JSON result recording.

v0.1.0#

Initial release of FlagTensor.

  • Added Features

    • Tensor-primitive library with multi-backend support.

    • Unary operations (ReLU and others).

    • Binary operations for element-wise tensor arithmetic.

    • Contraction operations for multi-dimensional reductions.

    • Correctness and performance comparison against cuTensor baselines.

Contents