Zyphra/ZAYA1-8B (Apache-2.0) converted to 1BP — the single-file model format used by 1bit.systems's inference engine.

1BP packs everything a loader needs into one memory-mappable file: a 256-byte header with model config, a variable-length tensor index, and Q4NX-tiled (32×256, 4-bit groups of 32 with bf16 scale/zero-point) weight data. No external config.json, no separate tokenizer file for the tensors themselves.

All 1283 tensors from the source model, losslessly categorized by rank:

Total: 8.84B parameters, matching the source GGUF exactly (ZAYA1-8B-Q4_K_M.gguf, 5.57 GB) — file is 6.58 GB.

An earlier conversion of this model only captured ~773M of the 8.84B parameters (the converter silently dropped every 1D and 3D tensor — all norms and all MoE expert weights). This upload is the fixed, complete version — see the feat/onebp-moe-experts history in the 1bit-systems repo for the format/converter/loader fix.

Base model: Zyphra/ZAYA1-8B, Apache-2.0. This repo contains a re-quantized (Q4NX, 4-bit) derivative in a different on-disk format; the original license terms apply to the underlying weights.