v0.1.2 - Test Coverage Expansion

See CHANGELOG.md for full details.

Highlights

• 50+ new property tests across all major crates using proptest
• 5 new fuzz targets covering safetensors, quantization, GGUF, and transformer config parsing
• BDD grid expanded to 18 scenario cells (was 13)
• 77 ignored tests (reduced from 91) — tests with resolved blockers unignored
• Server and CLI integration tests for health endpoints, CORS, security validation, template parsing
• Quantization numerical accuracy tests validating I2S, TL1, TL2, QK256 correctness
• E2E golden path tests for stop tokens, receipt schema, and max_tokens boundary

Requirements

• Rust 1.89.0+ (MSRV)
• cargo build --no-default-features --features cpu

v0.1.1 — Testing & Infrastructure Release

BitNet-rs v0.1.1

Overview

This release focuses on testing infrastructure, CI improvements, and property test coverage. All core inference functionality from v0.1.0 is preserved.

What's New

Testing

• 21 new proptest cases for bitnet-gguf (GgufValue round-trips, TensorInfo invariants, GgufMetadataKv key preservation)
• 7 new proptest cases for bitnet-sys (CompileTimeLibCapabilities summary correctness)
• InferenceReceipt::to_json_string() convenience method + snapshot test for receipt JSON output
• 10 new SRP cross-crate integration tests: logits→sampling pipeline, generation stop criteria, RoPE table correctness, device-probe determinism, engine-core session config wiring
• New fuzz target gguf_metadata_values for GGUF parser panic safety

CI / Infrastructure

• fuzz/** added to ci-core.yml path triggers — fuzz target PRs now get required CI gate checks
• BITNET_SKIP_SLOW_TESTS=1 added to CI environment — 30 model-gated tests now self-skip cleanly
• Ignored tests reduced: 110 → 78 (-32 tests unblocked)
• GitHub repository description, topics, and probot settings updated

No Breaking Changes

All public APIs from v0.1.0 are preserved. Update your dependency:

bitnet = "0.1.1"

Full Changelog

See CHANGELOG.md

v0.1.0-mvp: CPU Inference MVP with Strict Receipt Validation

BitNet.rs v0.1.0-mvp — CPU Inference MVP

This is the v0.1.0 Minimum Viable Product release, delivering production-ready CPU inference for BitNet neural networks with comprehensive quality gates.

---

🚀 What's New

Core CPU Functionality

• ✅ CPU inference with I2S/TL1/TL2 quantization (99%+ accuracy vs FP32)
• ✅ SIMD-optimized kernels (AVX2/AVX-512 on x86_64, NEON on ARM)
• ✅ Strict mode enforcement — no FP32 fallback in quantized layers
• ✅ Universal tokenizer with auto-discovery for HF/GGUF models

Quality & Validation

• ✅ Receipt verification system — proves real compute (no mock inference)
  • compute_path = "real" enforcement
  • Kernel hygiene validation (ID length ≤ 128, count ≤ 10K)
  • CPU kernel prefixes: i2s_, tl1_, tl2_
• ✅ Pinned CPU baseline — docs/baselines/20251015-cpu.json
  • Deterministic with BITNET_STRICT_MODE=1 BITNET_DETERMINISTIC=1 RAYON_NUM_THREADS=1
  • Performance SLO: ≥3.5 tokens/sec on CPU

Developer Experience

• ✅ 10-line quickstart in README (build → run → answer)
• ✅ Run → Emit → Verify workflow

  cargo run -p xtask -- benchmark --model model.gguf --tokens 128  # writes ci/inference.json
  cargo run -p xtask -- verify-receipt                            # validates receipt

• ✅ Comprehensive docs — feature flags, validation gates, troubleshooting

---

📦 Installation

Prerequisites

• Rust 1.90.0+ (MSRV: 1.90.0, 2024 edition)
• Feature flags required — always specify --no-default-features --features cpu

Build from Source

git clone https://github.com/EffortlessMetrics/BitNet-rs.git
cd BitNet-rs
git checkout v0.1.0-mvp

# Build CLI (CPU only)
cargo build --release --no-default-features --features cpu -p bitnet-cli

# Verify installation
./target/release/bitnet --version

Download Binaries

See Assets below for pre-built binaries (Linux x86_64, macOS ARM64, Windows x64).

---

🏃 Quick Start (10 lines)

# 1. Download a GGUF model
cargo run -p xtask -- download-model --id microsoft/bitnet-b1.58-2B-4T-gguf

# 2. Run inference (CPU)
cargo run -p bitnet-cli --no-default-features --features cpu -- \
  infer models/model.gguf \
  --prompt "What is 2+2?" \
  --max-tokens 50

# 3. Verify honest compute (receipt validation)
cargo run -p xtask --no-default-features --features inference -- \
  benchmark --model models/model.gguf --tokens 128  # writes ci/inference.json

cargo run -p xtask --no-default-features --features inference -- \
  verify-receipt                                    # validates receipt

---

🧪 Validation & Testing

Run Tests

cargo test --workspace --no-default-features --features cpu

Verify Receipt Against Baseline

BITNET_STRICT_MODE=1 BITNET_DETERMINISTIC=1 RAYON_NUM_THREADS=1 \
cargo run -p xtask --no-default-features --features inference -- \
  verify-receipt --path docs/baselines/20251015-cpu.json

Quality Checks

cargo fmt --all
cargo clippy --workspace --all-targets --no-default-features --features cpu -- -D warnings

---

📊 Performance Baseline

Pinned Baseline: docs/baselines/20251015-cpu.json

Metric	Value
Tokens/sec (CPU)	≥3.5 (baseline SLO)
Quantization Accuracy	99%+ vs FP32
Compute Path	"real" (no mock)
Kernels Executed	7 CPU quantized kernels

---

🔧 Configuration

Required Feature Flags

# CPU inference (default features are EMPTY)
cargo build --no-default-features --features cpu

# GPU inference (CUDA)
cargo build --no-default-features --features gpu

Environment Variables

Variable	Purpose
BITNET_STRICT_MODE=1	Enable strict validation (fail on FP32 fallback)
BITNET_DETERMINISTIC=1	Deterministic inference (with RAYON_NUM_THREADS=1)
RAYON_NUM_THREADS=1	Single-threaded for reproducibility

---

📚 Documentation

• Quickstart: docs/quickstart.md
• Feature Flags: docs/explanation/FEATURES.md
• Validation Gates: docs/reference/validation-gates.md
• Build Commands: docs/development/build-commands.md
• Troubleshooting: CLAUDE.md (search "Troubleshooting")

---

🛡️ Known Limits

• GPU support: Opt-in with --features gpu (not default)
• TL stress tests: Opt-in with BITNET_ENABLE_GPU_TESTS=1 (CI skips on CPU-only hosts)
• KV cache: Pre-allocation optimization slated for v0.2.0

---

🔗 Links

• GitHub: https://github.com/EffortlessMetrics/BitNet-rs
• Issue Tracker: https://github.com/EffortlessMetrics/BitNet-rs/issues
• Baseline Receipt: docs/baselines/20251015-cpu.json

---

🙏 Acknowledgements

Built with Rust 🦀 and inspired by Microsoft's BitNet architecture.

Full Changelog: 1f7dbd0...v0.1.0-mvp