Modelpublic

tensor-tailor/ralph-qwen3-8b-binary

sourceHugging Faceapache-2.0updated 26d agoView on Hugging Face

0likes135downloads

Model Card

ralph-qwen3-8b-binary

Binary-tier GGUF compression of Qwen/Qwen3-8B — architecture and parameter count unchanged (8,190,735,360 weight-bearing parameters), weights re-stored at reduced bit-width. Submitted to the binary bit-tier on Bittensor subnet 40 (Ralph, model-compression).

License

Released under Apache License 2.0 — see LICENSE. This is a derivative of Qwen3-8B, itself Apache-2.0; the original copyright/attribution notices and a description of what changed (including the third-party quantization tooling used) are preserved in NOTICE.

Model overview


Parent	Qwen/Qwen3-8B (Apache-2.0)
Parameters	8,190,735,360
Architecture	unchanged from parent
Format	GGUF
Bit tier	binary
File size	1,928,778,816 bytes (~1.80 GiB)
Measured bits/weight	~1.88 (file size / param count; embedding/output tensors kept at higher precision outside the compressed blocks)
Quantization method	llama.cpp, imatrix-calibrated

No retraining or architectural modification — this is a post-training weight re-storage of the parent's own weights.