Binary Neural Networks

Packed 1-bit inference for CPU and edge — with metrics that refuse to lie

Binary Neural Networks (bnn 1.0.0) is the honest optimiser toolkit for packed binary / ternary inference on CPU and edge. It bit-packs weights into uint64, runs real XNOR–popcount SIMD kernels, and prints dual metrics — pack math and wall-clock — as separate numbers.

It is not a claim that sign() is 32× faster on GPU. Compression 32× is exact uint64 pack size. Kernel speed is XNOR–popcount. Those are different physics.

32.00×

weight pack
_{exact, uint64-aligned}

~23.9×

SIMD vs NumPy FP32
_{64×4096×4096, one CPU}

err = 0

every ISA path
_{AVX-512 · AVX2 · NEON · scalar}

5.1×

faster kernel
_{aggregate, 12 shapes}

29.68×

measured RAM
_{not the 32× brochure}

REFUSE

when cosine is junk
_{drop-in is a gate, not a vibe}

Six numbers, six kinds of claim — that distinction is the product. Pack ratio is exact math. err = 0 is exact integer arithmetic. SIMD and kernel speedups are wall-clock on one machine. Resident RAM is measured from real buffers, which is why it sits below theoretical 32×. REFUSE is the report telling you not to ship a wrap whose cosine collapsed.

Jump: 60 seconds · When to use / when not · Kernel · Wrap · SIMD · Benchmarks · Bridges · Docs

60 seconds to a dual-metric report

PyPI bnn-lab is not live yet (human Trusted Publisher residual). Until that upload, install from Git. The short name bnn on PyPI is an unrelated package — import and CLI here stay bnn.

A non-editable pip install "bnn-lab @ git+…" is a library wheel. It does not ship repo scripts/. bnn repro, bnn optimise, and bnn recommend call those scripts — use them only after a clone. Do not put bnn repro on the next line after a wheel install.

Git pip — Python API only

pip install "bnn-lab @ git+https://github.com/KanakMalpani/Binary-Neural-Networks.git@v1.0.0"

import torch
import torch.nn as nn
from bnn.optimise import OptimiseConfig, optimise_model

class Tiny(nn.Module):
    def __init__(self):
        super().__init__()
        self.ffn_fc1 = nn.Linear(64, 256)
        self.ffn_fc2 = nn.Linear(256, 64)

    def forward(self, x):
        return self.ffn_fc2(torch.relu(self.ffn_fc1(x)))

x = torch.randn(8, 64)
result = optimise_model(
    Tiny(), x, OptimiseConfig(policy="hybrid_ffn", min_in_features=64)
)
print(result.payload["compression_replaced_weights"])  # 32× pack — size, not latency
print(result.payload["status"])                        # OK or REFUSE_DROP_IN_CLAIM

Clone — CLI (`bnn repro` / `bnn optimise` / `bnn recommend`)

git clone https://github.com/KanakMalpani/Binary-Neural-Networks.git
cd Binary-Neural-Networks
pip install -e ".[dev]" -c constraints.txt
python -m bnn.kernels.compile_native
bnn repro
bnn optimise --policy auto --report results\optimise_report.json
bnn recommend --goal edge-vision

Expect REPRO: PASS (exit 0). The report prints compression, cosine, wall-clock, and REFUSE/OK. Prefer bnn optimise over legacy bnn wrap --ultra.

No compiler? Install still succeeds — the NumPy packed path stays correct. Windows native needs MSVC x64 (MinGW 32-bit → WinError 193). Full path: docs/GUIDE_E2E.md.

The thesis (locked)

Most “binary” demos train with STE, then infer with sign() + nn.Linear. That is a simulation. On commodity GPUs it is often slower than FP32. The 32× in papers is usually bit-pack compression, not end-to-end latency.

Locked claim	Meaning
Speedups come from packed kernels	XNOR + popcount on CPU/edge — not `sign()` theatre
Training STE ≠ inference throughput	STE trains latents; inference uses pack + GEMM
Compression 32×	Exact for aligned uint64 binary pack — size, not e2e latency
Commodity GPU quality	INT4 / FP8 / AWQ / vLLM — documented bridges, not fake BNN wins
Repro culture	`bnn repro` + `tests/golden_floors.json` + committed `results/*.json`

When to use / when not

Honesty is the product. If another stack wins, this lab says so and routes you there (bnn recommend). Compact callout under the thesis — not a table buried at the bottom.

Use this lab when

You want smaller weights on CPU / edge and will retrain or wrap wide GEMMs
You need real XNOR–popcount, not sign() + nn.Linear
You want a report that refuses drop-in when cosine is junk
You are researching packed binary / ternary kernels, Bi-Real-style vision, or .bnnpack
You want dual metrics: 32× pack and measured wall-clock, separately

Do not use this lab when

GPU server quality → FP8 / AWQ-INT4 + vLLM — not sign(), not “GPU 32×”
Local CPU LLM chat → bitnet.cpp (BitNet) or GGUF Q4_K_M
Phone / NPU stock SDK → INT8 / INT4 — vendors do not ship native 1-bit
Production ASR / diffusion fidelity → INT8 Whisper / ORT / FP8 PTQ (audio here is synthetic)
You need to claim 32× e2e latency from pack math alone — forbidden forever

flowchart TD
  Q{"Primary goal?"}
  Q -->|"GPU server quality"| GPU["FP8 / AWQ-INT4 + vLLM<br/>NOT classic BNN"]
  Q -->|"CPU local LLM"| LLM{"BitNet checkpoint?"}
  LLM -->|yes| BN["bitnet.cpp"]
  LLM -->|no| GGUF["GGUF Q4_K_M"]
  Q -->|"Edge vision, retrain OK"| EV["this repo — Bi-Real + packed GEMM"]
  Q -->|"Phone NPU stock SDK"| NPU["INT8 / INT4<br/>no stock 1-bit"]
  Q -->|"Packed XNOR research"| LAB["this repo — bnn optimise"]
  Q -->|"Diffusion / production ASR"| AVOID["INT8 / FP8 PTQ<br/>not this audio lane"]
  style LAB fill:#ddf4ff,stroke:#0969da
  style GPU fill:#fff8c5,stroke:#9a6700
  style BN fill:#fff8c5,stroke:#9a6700
  style NPU fill:#fff8c5,stroke:#9a6700
  style AVOID fill:#fff8c5,stroke:#9a6700

bnn recommend --goal edge-vision

(bnn recommend needs the clone / editable install above — not a git-pip wheel.)

Also skip (or hybrid-skip) small GEMMs / attention projections (packing overhead wins; auto leaves them FP) and drop-in HF LLMs without QAT (cold binary PTQ cosine often collapses — the report REFUSEs unless --force).

Full tree: docs/GUIDE_E2E.md · docs/18_DECISION_TREE_AND_COMPLETE_ROADMAP.md · limits: MODEL_CARD.md.

Core kernel pipeline

FP32 tensors are not “made binary” by sign() in nn.Linear. Inference quantizes to ±1, packs 32 or 64 values into a uint32/uint64 word, then a SIMD kernel does XNOR + popcount and scales back. That pack is the 32×. The popcount is the speed.

flowchart LR
  FP["FP32 weights<br/>and activations"] --> Q["Quantize to ±1"]
  Q --> P["Bit-pack<br/>32/64 values into<br/>uint32 / uint64"]
  P --> X["SIMD XNOR + popcount<br/>AVX-512 / AVX2 / NEON"]
  X --> S["Scale / dequant<br/>alpha · y + bias"]
  S --> Y["Output activations"]
  style P fill:#ddf4ff,stroke:#0969da
  style X fill:#ddf4ff,stroke:#0969da

flowchart TB
  subgraph Train["Training — STE simulation"]
    W["Full-precision latents"]
    STE["Straight-through estimator"]
    W --> STE
    STE --> W
  end
  subgraph Infer["Inference — packed path"]
    Pack["uint64 pack"]
    GEMM["XNOR-popcount GEMM"]
    Out["alpha · y + bias"]
    Pack --> GEMM --> Out
  end
  STE -.->|"checkpoint / policy"| Pack
  style Train fill:#f6f8fa,stroke:#57606a
  style Infer fill:#ddf4ff,stroke:#0969da

Wrap & optimisation flow

bnn optimise does not blindly binarize every Linear. It measures, assigns a per-layer policy, optionally QAT/distills, then refuses drop-in when cosine is below the gate.

flowchart LR
  M["FP / HF model"] --> Sens["Sensitivity<br/>per-layer"]
  Sens --> Pol{"Layer policy"}
  Pol -->|"binary"| Bin["Packed XNOR<br/>32× size"]
  Pol -->|"ternary"| Ter["Bitplane 1.58-bit"]
  Pol -->|"skip"| Skip["Keep FP32"]
  Bin --> Rec["QAT / distill<br/>optional"]
  Ter --> Rec
  Skip --> Rec
  Rec --> Gate{"Drop-in honesty<br/>cosine vs threshold"}
  Gate -->|"pass"| OK["status: OK<br/>drop_in_ok true"]
  Gate -->|"fail"| RF["status: REFUSE<br/>unless --force"]
  style OK fill:#dafbe1,stroke:#1a7f37
  style RF fill:#ffebe9,stroke:#cf222e

Default --policy auto on the documented hybrid demo lands cosine ~0.70 and REFUSE_DROP_IN — that is working as designed, not a silent 32× quality claim. Ternary+QAT can reach cosine 0.991 and still lose wall-clock (e2e 0.73×). The product gap is hybrid/binary that is both drop-in and faster — not paperwork.

Layer search is monotonic and tested: 32× is available at cosine 0.27. That is why the search exists.

`quality_floor`	final cosine	theoretical compression	assignment
0.00	0.271	32.0×	3 binary
0.90	0.950	1.71×	1 ternary, 2 skip
0.999	1.000	1.00×	3 skip

Details: docs/42_QAT_AND_LAYER_SEARCH.md · tutorial 07.

SIMD execution ladder

One C source. ISA is chosen at run time — never -march=native baking the builder’s CPU into a wheel. AVX-512 is used when present, never required. WASM SIMD128 is a pedagogy path (wasm/), not a substitute for the native kernel.

flowchart TB
  Entry["binary_gemm"] --> Det{"cpuid / xgetbv<br/>or ARM features"}
  Det -->|"x86_64 + VPOPCNTDQ"| AVX512["AVX-512"]
  Det -->|"x86_64 else"| AVX2["AVX2 nibble LUT"]
  Det -->|"ARM64 / Apple Silicon"| NEON["NEON vcnt"]
  Det -->|"none of the above"| Scalar["Scalar popcount"]
  AVX512 --> Done["OpenMP + 4-row blocking<br/>optional alpha/bias epilogue"]
  AVX2 --> Done
  NEON --> Done
  Scalar --> Done
  Done --> Native{"native library loaded?"}
  Native -->|yes| Fast["Packed SIMD GEMM"]
  Native -->|no| NP["Portable NumPy fallback<br/>correct, not always fast"]
  WASM["WASM SIMD128<br/>pedagogy only"] -.->|"not the production dispatch"| Entry
  style AVX512 fill:#ddf4ff,stroke:#0969da
  style AVX2 fill:#ddf4ff,stroke:#0969da
  style NEON fill:#ddf4ff,stroke:#0969da
  style WASM fill:#f6f8fa,stroke:#57606a
  style NP fill:#fff8c5,stroke:#9a6700

Platform	Native	Production ladder
Linux x86-64 (GCC/Clang)	yes	AVX-512 → AVX2 → scalar
Windows x64 (MSVC)	yes	AVX-512 → AVX2 → scalar
macOS / Linux arm64	yes	NEON
macOS x86-64	yes	AVX2 → scalar
Browser / teaching	WASM SIMD128	pedagogy — `wasm/`
Anything else	NumPy packed GEMM	correctness first

Deep dive: docs/41_PORTABLE_SIMD_KERNEL.md.

bnn validate-native          # selected ISA path, err = 0
BNN_KERNEL=scalar bnn bench  # force scalar / avx2 / avx512 / neon

Dual-metric benchmarks

Never equate pack math with latency.

flowchart LR
  subgraph Theory["Theory / size"]
    T1["32× weight pack<br/>uint64 aligned"]
    T2["~64× word ops<br/>XNOR-popcount"]
  end
  subgraph Wall["Wall-clock / energy-proxy"]
    W1["Kernel microbench<br/>prepacked GEMM"]
    W2["E2E forward<br/>incl. act-pack"]
  end
  Theory -.->|"do not equate"| Wall

Quantity	Kind	Do not claim as
Weight pack 32.00×	Exact	End-to-end latency
Native GEMM err = 0	Exact (when native loaded)	Accuracy of a wrapped LLM
ISA paths agree	Exact	Cross-machine float identity
~23.9× vs NumPy FP32 at 64×4096×4096	Wall-clock (machine-dependent)	Full-model FPS / GPU 32×
Wrap e2e speedup	Wall-clock (machine-dependent)	Drop-in quality

Committed snapshot (CPU; results/SUMMARY.md):

Check	Result
Pack compression	32.0× (exact)
Native GEMM vs ±1 FP32	err = 0
Every ISA path agrees bit-for-bit	err = 0 (binary and ternary)
64×4096×4096 compute vs NumPy FP32	~23.9× (machine-dependent)
Wrapped Linear, measured RAM	29.68× (theoretical 32.00×)
MNIST binary / ternary	96.36% / 97.16% (FP 97.67%)
CIFAR Bi-Real vs FP CNN	61.14% vs 71.14% (~10 pp)
Audio binary vs FP (synthetic tones)	96.0% vs 94.5% — not production ASR

Where the kernel speed came from (before → after, same process, min-of-5)

The old kernel opened a new OpenMP parallel region per batch row and re-streamed all of W B times. One team per call plus 4-row register blocking, then runtime SIMD dispatch:

Shape (B×N×M)	before	after
8 × 4096 × 4096	0.212 ms	0.062 ms	3.4×
64 × 4096 × 4096	1.999 ms	0.437 ms	4.6×
256 × 1024 × 1024	0.739 ms	0.119 ms	6.2×
512 × 512 × 512	2.038 ms	0.103 ms	19.9×
aggregate, 12 shapes	10.74 ms	2.12 ms	5.1×

Tiny shapes are call-overhead bound and unchanged — as expected. Wall-clock moves with core count, memory bandwidth and thermals; err = 0 does not. Details: docs/41_PORTABLE_SIMD_KERNEL.md.

Floors live in tests/golden_floors.json. Wall-clock ratios move with CPU, threads, and OpenMP — gates check conclusions, not bit-identical floats.

Ecosystem & bridges

This lab occupies packed PyTorch BNN optimisation now that Larq (TF/Keras) is archived. It does not compete with bitnet.cpp on LLM tok/s, or with torchao/vLLM on GPU INT4/FP8. When those win, bnn bridge / bnn recommend say so.

flowchart TB
  subgraph Lab["This lab — bnn-lab"]
    OPT["bnn.optimise"]
    KER["Packed XNOR / ternary GEMM"]
    COD[".bnnpack codec"]
    OPT --> KER
    OPT --> COD
    COD --> KER
  end
  PT["PyTorch STE train"] --> OPT
  HF["Hugging Face models"] --> OPT
  KER -.->|"ONNX = bridge only"| ONNX["ONNX Runtime"]
  OPT -.->|"GPU quality"| TAO["torchao / AWQ / FP8<br/>vLLM"]
  OPT -.->|"BitNet-style CPU LLM"| BCPP["bitnet.cpp"]
  OPT -.->|"generic local LLM"| GGUF2["llama.cpp GGUF"]
  style Lab fill:#ddf4ff,stroke:#0969da
  style TAO fill:#fff8c5,stroke:#9a6700
  style BCPP fill:#fff8c5,stroke:#9a6700

flowchart TB
  CLI["bnn CLI"]
  OPT2["bnn.optimise<br/>policy · calib · QAT"]
  WRAP["bnn.wrap<br/>PackedLinear / hybrid FFN"]
  CODEC["bnn.codec"]
  KER2["bnn.kernels"]
  STEZ["STE zoo · vision · audio · seq"]
  CLI --> OPT2
  CLI --> CODEC
  CLI --> STEZ
  OPT2 --> WRAP
  WRAP --> KER2
  CODEC --> KER2
  WRAP --> CODEC

Installing does not require a compiler. setup.py builds the kernel when a toolchain is present and falls back to NumPy otherwise. Prebuilt wheels from wheels.yml are five cibuildwheel jobs — linux-x86_64, linux-aarch64, windows-amd64, macos-arm64, macos-x86_64 — and land as Actions / v* tag artifacts until Trusted Publisher. There is no Windows ARM64 wheel. Live pip install bnn-lab from PyPI still needs Trusted Publishing.

What you can run next

Path	Command / entry	Docs
Optimiser	`bnn optimise --policy auto`	GUIDE §4 · tutorial 07 · HF 08
Per-layer search	`bnn.wrap.search_layer_modes(...)`	docs/42
QAT recovery	`bnn optimise --qat-steps 200`	docs/42 — search before QAT
Memory footprint	`bnn memory --dim 1024 --ff 4096`	docs/43
Codec	`bnn encode` / `bnn decode`	GUIDE §5
MNIST STE	`bnn train --epochs 3 --seed 42`	pedagogy — not a throughput win
Vision	`bnn train-image --epochs 8 --subset 30000`	tutorial 04
Audio	`bnn train-audio --epochs 5`	tutorial 05 — synthetic only
Seq2seq / profile	`bnn train-seq2seq` · `bnn profile`	tutorial 06

Start here
Human path	`docs/GUIDE_E2E.md` — install → repro → optimise
Browsable docs	GitHub Pages
Reproduce	`REPRODUCIBILITY.md` · `bnn repro`
AI agents	`AGENTS.md`
Knowledge graph	`knowledge_graph/` · `docs/44_KNOWLEDGE_GRAPH.md`
Roadmap	`ROADMAP.md`
Compatibility	`docs/COMPATIBILITY_MATRIX.md`
Limits	`MODEL_CARD.md`

bnn/           STE, layers, models, optimise, export, determinism
bnn/wrap/      hybrid policy, calib, QAT, PackedLinear
bnn/kernels/   portable XNOR GEMM (+ optional native)
bnn/codec/     .bnnpack encode / decode
bnn/vision/    CIFAR Bi-Real, tiny binary ViT
bnn/audio/     STFT + synthetic tones
bnn/seq/       binary Transformer encoder / decoder
results/       committed golden JSON + SUMMARY.md
tests/         pytest + golden_floors.json

Public API: import bnn — docs/api/README.md. CLI: bnn --help · bnn --version.

Is / is not

Is	Is not
Honest CPU / edge proof of packed binary speedups	A promise of 32× e2e everywhere
Trainable BNN + BitLinear pedagogy + optimiser	Full BitNet LLM pretrain
Dual-metric culture and repro gates	Bit-identical floats across OS/CPU
Bridges toward INT4 / FP8 / bitnet.cpp	A cuDNN / TensorRT replacement
Tagged v1.0.0 lab (PyPI upload still human)	A fake-binary GPU 32× story

Contributing & quality

CONTRIBUTING.md · CHANGELOG.md · SECURITY.md · docs/LAUNCH_CHECKLIST.md
Product direction: ROADMAP.md (Phases A→F; workstreams W1–W14)
API reference is generated from docstrings (mkdocs build --strict in CI) — see docs/api/; a renamed symbol breaks the build rather than silently emptying a page
Site: kanakmalpani.github.io/Binary-Neural-Networks
Supply chain: wheels + sdist carry signed build provenance (gh attestation verify), and pip-audit is a hard gate on the shipped dependency set with every ignore triaged in ci.yml
Agents: AGENTS.md — do not invent alternate golden shapes
CI: ci.yml — quality (ruff/mypy/coverage ≥80%), Windows + Linux native (export-check, repro), portability (linux-arm64 NEON, macos-arm64 NEON, macos-x86_64) per COMPATIBILITY_MATRIX.md, Python 3.11–3.13; plus CodeQL, Scorecard, wheels

bnn export-check
bnn validate-native
bnn bench
bnn eval-suite

License: MIT · Citation: CITATION.cff · Repo: KanakMalpani/Binary-Neural-Networks

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

bnn_lab-1.0.0.tar.gz (198.0 kB view details)

Uploaded Aug 14, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl (292.0 kB view details)

Uploaded Aug 14, 2026 CPython 3.13manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl (276.8 kB view details)

Uploaded Aug 14, 2026 CPython 3.13manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl (153.5 kB view details)

Uploaded Aug 14, 2026 CPython 3.13macOS 11.0+ ARM64

bnn_lab-1.0.0-cp312-cp312-win_amd64.whl (154.2 kB view details)

Uploaded Aug 14, 2026 CPython 3.12Windows x86-64

bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl (292.0 kB view details)

Uploaded Aug 14, 2026 CPython 3.12manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl (276.8 kB view details)

Uploaded Aug 14, 2026 CPython 3.12manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl (154.2 kB view details)

Uploaded Aug 14, 2026 CPython 3.12macOS 11.0+ x86-64

bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl (153.5 kB view details)

Uploaded Aug 14, 2026 CPython 3.12macOS 11.0+ ARM64

bnn_lab-1.0.0-cp311-cp311-win_amd64.whl (154.2 kB view details)

Uploaded Aug 14, 2026 CPython 3.11Windows x86-64

bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl (292.0 kB view details)

Uploaded Aug 14, 2026 CPython 3.11manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl (276.8 kB view details)

Uploaded Aug 14, 2026 CPython 3.11manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl (154.2 kB view details)

Uploaded Aug 14, 2026 CPython 3.11macOS 11.0+ x86-64

bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl (153.5 kB view details)

Uploaded Aug 14, 2026 CPython 3.11macOS 11.0+ ARM64

File details

Details for the file bnn_lab-1.0.0.tar.gz.

File metadata

Download URL: bnn_lab-1.0.0.tar.gz
Upload date: Aug 14, 2026
Size: 198.0 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0.tar.gz
Algorithm	Hash digest
SHA256	`f50c686ebc00c584df529afbcb55fc4ff0b521616b7d7201aaac1835360d4238`
MD5	`4acc987cfe094a0248f64f0579293e50`
BLAKE2b-256	`072f5dd1ae505e17556aced3fd9aa1e657a83b8017c606e83f74eb9c0abbbe64`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0.tar.gz:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0.tar.gz
- Subject digest: f50c686ebc00c584df529afbcb55fc4ff0b521616b7d7201aaac1835360d4238
- Sigstore transparency entry: 2468030148
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Upload date: Aug 14, 2026
Size: 292.0 kB
Tags: CPython 3.13, manylinux: glibc 2.17+ x86-64, manylinux: glibc 2.28+ x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Algorithm	Hash digest
SHA256	`e7a87aac56090c32af3347ce21ecd32ad12ae05a4603eed8603bb3ab1a695025`
MD5	`1d85d2d72d727f6f49bbee76ac4b9989`
BLAKE2b-256	`a1da80f20397840c0f0133ced27a5cab8ff0e9d8d264e965ba641a3a77978d89`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
- Subject digest: e7a87aac56090c32af3347ce21ecd32ad12ae05a4603eed8603bb3ab1a695025
- Sigstore transparency entry: 2468030684
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Upload date: Aug 14, 2026
Size: 276.8 kB
Tags: CPython 3.13, manylinux: glibc 2.17+ ARM64, manylinux: glibc 2.28+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Algorithm	Hash digest
SHA256	`5632c50cab25df3520a0d1ccb4bd875591e5a5989c3341d41d81c913b3de1f35`
MD5	`a712e240796f2c31a46a10aab9ee7b00`
BLAKE2b-256	`2eed21f10c8bdd72b2df0328cf0de0273906cbce7bb6b44af1e214a2ab164d2f`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
- Subject digest: 5632c50cab25df3520a0d1ccb4bd875591e5a5989c3341d41d81c913b3de1f35
- Sigstore transparency entry: 2468030307
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl
Upload date: Aug 14, 2026
Size: 153.5 kB
Tags: CPython 3.13, macOS 11.0+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl
Algorithm	Hash digest
SHA256	`f4b87781cac7055318555edb7f702fb1a5205acdedaec554c649e7b1db6c2543`
MD5	`ecfec213f0562b47e84f457ee96098dd`
BLAKE2b-256	`38ee856c961568d0562c49c980031d66047393488892ca587cf54a99fc9fb628`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp313-cp313-macosx_11_0_arm64.whl
- Subject digest: f4b87781cac7055318555edb7f702fb1a5205acdedaec554c649e7b1db6c2543
- Sigstore transparency entry: 2468030468
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp312-cp312-win_amd64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp312-cp312-win_amd64.whl
Upload date: Aug 14, 2026
Size: 154.2 kB
Tags: CPython 3.12, Windows x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp312-cp312-win_amd64.whl
Algorithm	Hash digest
SHA256	`af63b1d5ca70889a9f85b50de3f647f8b08f3c2f373437966c4744bd13df187b`
MD5	`31f66493180b6e5b77f90562d18fa809`
BLAKE2b-256	`7e74f145331e42fb7bad407470f3f22152b382c71206a06f89a1043d45b1c2ec`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp312-cp312-win_amd64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp312-cp312-win_amd64.whl
- Subject digest: af63b1d5ca70889a9f85b50de3f647f8b08f3c2f373437966c4744bd13df187b
- Sigstore transparency entry: 2468030629
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Upload date: Aug 14, 2026
Size: 292.0 kB
Tags: CPython 3.12, manylinux: glibc 2.17+ x86-64, manylinux: glibc 2.28+ x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Algorithm	Hash digest
SHA256	`068d48116bab01b45495a102f73289bf418d04d3a3054d7390ab88c3c15e5fe2`
MD5	`f28e21c70deabd19b59f7c17c66a8c43`
BLAKE2b-256	`8c406cd8a7fbba43f4b311c0173d1669546192928f0d73d9ec13a1118ba02bbd`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
- Subject digest: 068d48116bab01b45495a102f73289bf418d04d3a3054d7390ab88c3c15e5fe2
- Sigstore transparency entry: 2468030417
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Upload date: Aug 14, 2026
Size: 276.8 kB
Tags: CPython 3.12, manylinux: glibc 2.17+ ARM64, manylinux: glibc 2.28+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Algorithm	Hash digest
SHA256	`6bb86d3355d7a9ba199c1efd741b9dbd0c91e531a2e06425d6595224b30ee169`
MD5	`74532905d5b9183585629af3ea722919`
BLAKE2b-256	`74cac3af6fea6249e4ad8f6b485f47a790b0334127a9914bdd63e3e0d22b44ea`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
- Subject digest: 6bb86d3355d7a9ba199c1efd741b9dbd0c91e531a2e06425d6595224b30ee169
- Sigstore transparency entry: 2468030745
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl
Upload date: Aug 14, 2026
Size: 154.2 kB
Tags: CPython 3.12, macOS 11.0+ x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl
Algorithm	Hash digest
SHA256	`3bef052d33a8a526fdfbafbcdab285101810da72d66b53b6b0ce80eb851328ca`
MD5	`33a84a80251d0d839417927374a868a6`
BLAKE2b-256	`86d15ccb320a9fa8873c2813fd0a0d6743115136bfb1d0a3fac3ae235d5b730a`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp312-cp312-macosx_11_0_x86_64.whl
- Subject digest: 3bef052d33a8a526fdfbafbcdab285101810da72d66b53b6b0ce80eb851328ca
- Sigstore transparency entry: 2468030786
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl
Upload date: Aug 14, 2026
Size: 153.5 kB
Tags: CPython 3.12, macOS 11.0+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl
Algorithm	Hash digest
SHA256	`a672d749c5c55b78d2295e3549fa1f09b3284c65ca18431118d53c9886cbb159`
MD5	`b863b6e7ba2bfa52db75508e0488e76c`
BLAKE2b-256	`e945df03891a5ef4dd5bedf8ec7b1dd289746c5abbddc79bf7d97532de8d2aef`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp312-cp312-macosx_11_0_arm64.whl
- Subject digest: a672d749c5c55b78d2295e3549fa1f09b3284c65ca18431118d53c9886cbb159
- Sigstore transparency entry: 2468030256
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp311-cp311-win_amd64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp311-cp311-win_amd64.whl
Upload date: Aug 14, 2026
Size: 154.2 kB
Tags: CPython 3.11, Windows x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp311-cp311-win_amd64.whl
Algorithm	Hash digest
SHA256	`79af000afc92cc8191dc1fd566d191dfd1afb513b00ef8c0d217ec0c64299cb0`
MD5	`aa6a3d073c95a394b776a36d112db21b`
BLAKE2b-256	`9f9c43545064393c22e8b53043bf80b2f8635a95734b7eff55940daefa294dee`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp311-cp311-win_amd64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp311-cp311-win_amd64.whl
- Subject digest: 79af000afc92cc8191dc1fd566d191dfd1afb513b00ef8c0d217ec0c64299cb0
- Sigstore transparency entry: 2468030200
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Upload date: Aug 14, 2026
Size: 292.0 kB
Tags: CPython 3.11, manylinux: glibc 2.17+ x86-64, manylinux: glibc 2.28+ x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
Algorithm	Hash digest
SHA256	`398d6f5de17a599ded3b47435186edb9d5b98b9c8d28d44d304149876ef429b3`
MD5	`419e445582fb0dd28e5c072a223cef48`
BLAKE2b-256	`b13de13ce5ba61316ab6d589c2be36cb4b7eb9d6686b7784fd28cdcdbbc8638a`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl
- Subject digest: 398d6f5de17a599ded3b47435186edb9d5b98b9c8d28d44d304149876ef429b3
- Sigstore transparency entry: 2468030531
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Upload date: Aug 14, 2026
Size: 276.8 kB
Tags: CPython 3.11, manylinux: glibc 2.17+ ARM64, manylinux: glibc 2.28+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
Algorithm	Hash digest
SHA256	`91b040142376dcefddada2fb7b0dccfe3198e2f1fe994991f76994c8d49d5e2a`
MD5	`a906a1b508bcd34db1d8a02b564a3ee7`
BLAKE2b-256	`2f509848b5d74962bdde3e4048bc13a610b67e12873693a6dfcd468f31816436`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl
- Subject digest: 91b040142376dcefddada2fb7b0dccfe3198e2f1fe994991f76994c8d49d5e2a
- Sigstore transparency entry: 2468030344
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl
Upload date: Aug 14, 2026
Size: 154.2 kB
Tags: CPython 3.11, macOS 11.0+ x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl
Algorithm	Hash digest
SHA256	`cfadfcd4bc11e6cd49c20ec71cb51994d9a7403e17826081e2616e80e8ca88b6`
MD5	`91ae40c14ca8da8d10076c55a16ade86`
BLAKE2b-256	`beffe66eb903e557b0ef73152f639e3f3d069b0b2dd41120048f186d87988010`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp311-cp311-macosx_11_0_x86_64.whl
- Subject digest: cfadfcd4bc11e6cd49c20ec71cb51994d9a7403e17826081e2616e80e8ca88b6
- Sigstore transparency entry: 2468030580
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

File details

Details for the file bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl.

File metadata

Download URL: bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl
Upload date: Aug 14, 2026
Size: 153.5 kB
Tags: CPython 3.11, macOS 11.0+ ARM64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl
Algorithm	Hash digest
SHA256	`fbb107679761ca6e195d6393305db2eacbf810b47d9c343d6ef86cd902e4b0e4`
MD5	`5da77455844fc55b4dd0a198e1ee7b40`
BLAKE2b-256	`7336a93ff749ab409b6feedadda4f48e383ddec7b3373706f83a28588f5704c2`

See more details on using hashes here.

Provenance

The following attestation bundles were made for bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl:

Publisher: wheels.yml on KanakMalpani/Binary-Neural-Networks

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: bnn_lab-1.0.0-cp311-cp311-macosx_11_0_arm64.whl
- Subject digest: fbb107679761ca6e195d6393305db2eacbf810b47d9c343d6ef86cd902e4b0e4
- Sigstore transparency entry: 2468030371
- Sigstore integration time: Aug 14, 2026
Source repository:
- Permalink: KanakMalpani/Binary-Neural-Networks@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Branch / Tag: refs/heads/main
- Owner: https://github.com/KanakMalpani
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: wheels.yml@a7b046fe0ad4a22d46038cc4b53cc4e00965c6a6
- Trigger Event: workflow_dispatch

Release history Release notifications | RSS feed

This release

1.0.0

Aug 14, 2026

bnn-lab 1.0.0

Binary Neural Networks

Packed 1-bit inference for CPU and edge — with metrics that refuse to lie

32.00×

~23.9×

err = 0

5.1×

29.68×

REFUSE

60 seconds to a dual-metric report

Git pip — Python API only

Clone — CLI (bnn repro / bnn optimise / bnn recommend)

The thesis (locked)

When to use / when not

Core kernel pipeline

Wrap & optimisation flow

SIMD execution ladder

Dual-metric benchmarks

Ecosystem & bridges

What you can run next

Is / is not

Contributing & quality

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License

Requires

Provides Extra

Tags

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

Clone — CLI (`bnn repro` / `bnn optimise` / `bnn recommend`)