bnn-lab-codec-canary

bnn-lab codec .bnnpack (canary)

Canary / lab demo — not ImageNet SOTA. One random 256×256 Linear encoded to .bnnpack v2 so strangers can download a noun (the format), round-trip decode, and check GEMM err = 0.

Equivalent local command:

bnn encode --source random --in-features 256 --out-features 256 --out results/codec_canary.bnnpack
bnn decode --pack results/codec_canary.bnnpack

What 32× means

32× is exact uint64 pack compression when in_features % 64 == 0 (256/64 = 4 words). Not a GPU claim from sign(). Native XNOR GEMM err = 0 when the platform DLL/so is present; NumPy packed path is also err = 0 (tests/golden_floors.json: native_err_max 0.0, compression_exact_when_uint64_pack 32.0).

Load

from huggingface_hub import hf_hub_download
from bnn.codec import decode_file, packed_module_fp_err

path = hf_hub_download("KanakMalpani/bnn-lab-codec-canary", filename="model.bnnpack")
modules, meta = decode_file(path)
mod = modules["linear"]
assert packed_module_fp_err(mod) == 0.0
print(meta, mod.in_features, mod.out_features)

License

MIT (same as bnn-lab).