Skip to content

Releases: gHashTag/zig-golden-float

v2.1.0 — Production Release

29 Apr 19:38
@gHashTag gHashTag
v2.1.0
0b4717f

Choose a tag to compare

View all tags
v2.1.0 — Production Release Latest
Latest

GoldenFloat v2.1.0 — Production Release

Highlights

README & Documentation

  • Complete README rewrite with format table, architecture map, benchmarks, binding instructions, φ-FMA reference, IGLA overview
  • CI badge, release badge, proper installation instructions

Version Alignment

  • All package versions aligned: build.zig.zon 2.1.0, Cargo.toml 2.1.0, Rust crate 2.0.0, Python 2.0.0, C header v2.0.0

CI — All 5 Bindings Green

  • Zig, Rust, Python, Go, C++ — all passing on every push
  • Fixed shared library install step (addInstallArtifact with options struct)
  • Added LD_LIBRARY_PATH and GOLDENFLOAT_LIB_DIR to CI workflow

Bug Fixes

  • trinity_constants.zig: fixed type error in phiLrSchedule (f64 * u32)
  • phi_attention.zig: explicit if-block style for Zig 0.15 compatibility
  • CMakeLists.txt: replaced // comment with #
  • Go binding: proper cgo type casts (C.float, C.uint16_t, C.bool)
  • Go binding: added #include "gf16.h" and extern declarations to cgo preamble
  • C++ binding: added cpp/include to include dirs, fixed JSON type handling
  • Python binding: honor GOLDENFLOAT_LIB_DIR env var, added phi/trinity/version restype
  • All bindings: one() constant now uses from_f32(1.0) (GF16_ONE=0x3C00 encodes 0.25)
  • Root Cargo.toml: fixed cfg(unix) target predicate

Stats

  • 12 commits since v2.0.0
  • All 5 language bindings CI-green
  • 18K+ lines of Zig source

Installation

zig fetch --save https://github.com/gHashTag/zig-golden-float/archive/refs/tags/v2.1.0.tar.gz
Assets 2
Loading

v2.0.0 — Production Release

29 Apr 19:03
@gHashTag gHashTag
v2.0.0
12a9bc9

Choose a tag to compare

View all tags
v2.0.0 — Production Release

GoldenFloat v2.0.0 — Production Release

Highlights

Formats & Codecs

  • GF16/fp16/bf16/GF8/GFTernary with canonical IEEE-754 round-to-nearest-even
  • φ-optimized FMA, FMS, FNMA, phiFma, phiDot
  • C-ABI shared library (libgoldenfloat.{so,dylib,dll}) v2.0.0

IGLA-GF16 Architecture

  • φ-Sparse Attention with Fibonacci distance mask (sparsity 2.15%)
  • Trinity Weight Initialization (4 physics sectors)
  • φ-LR Schedule (warmup + φ-decay)
  • JEPA-T Predictor (6+3 layer φ-split)
  • 5 whitepaper proofs verified

Benchmarks

  • BENCH-007b: GF8 saturates at [-10,10], GF16 stable
  • BENCH-008: Fashion-MNIST MLP quantization pipeline validated
  • BENCH-010: bf16/gf16 diverge on Uniform [-100,+100]; H1 confirmed
  • GF16 outperforms bf16 by 16× on all distributions

Bug Fixes

  • bf16 encoder: canonical (bits +| 0x7FFF) >> 16 (was frexp ±7 clamp)
  • fp16 subnormal decoder: fixed 131072× magnitude error
  • GF16 NaN preservation
  • CI, Rust/Go/C++ bindings

Stats

  • 18 issues resolved
  • 11 PRs merged
  • 80 commits since v1.0.0
  • 52 Zig source files, 18K+ lines
  • 4 language bindings (C/C++, Rust, Python, Go)

Installation

zig fetch --save https://github.com/gHashTag/zig-golden-float/archive/refs/tags/v2.0.0.tar.gz

Full Changelog

See CHANGELOG.md

Loading

GoldenFloat v1.0.0

31 Mar 12:00
@gHashTag gHashTag
v1.0.0
e68250e

Choose a tag to compare

View all tags
GoldenFloat v1.0.0 
Release v1.0.0 — GoldenFloat φ-Optimized ML Kernel

Features:
- GF16: 6-bit exponent, 9-bit mantissa (φ-optimized)
- TF3: Ternary float format
- VSA: Vector Symbolic Architecture operations
- HybridBigInt: Ternary big integer arithmetic
- 422 tests passing

Bypasses 62 Zig compiler bugs:
- 21 Urgent issues
- 11 filed by Zig core team
- 13 platforms supported

Platforms: x86_64, macOS, Windows, WASM, WASI, AVR, MIPS,
ARM, FreeBSD, RISC-V, PowerPC, Android, SPIR-V
Loading