Downloads · 30 days
38
11% of all-time downloads
boweizh1204/ralph-qwen3-8b-sub2-qat5
ralph-qwen3-8b-sub2-qat5 is a machine learning model from boweizh1204. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Compressed derivative of Qwen/Qwen3-8B submitted to the sub2 bit-tier of Bittensor subnet 40 (Ralph, model compression). Architecture and parameter count are unchanged from the parent; weights are re-stored at reduced…
Downloads · 30 days
38
11% of all-time downloads
All-time downloads
337
Public
Repo size
2.9 GB
Likes
0
Public
Click a slice to open those files.
.gguf2.9 GB · 100%
From the Hugging Face model README
Compressed derivative of Qwen/Qwen3-8B submitted to the sub2 bit-tier of Bittensor subnet 40 (Ralph, model compression). Architecture and parameter count are unchanged from the parent; weights are re-stored at reduced bit-width.
Released under the Apache License 2.0 — see LICENSE. This is a derivative of Qwen3-8B, itself Apache-2.0;
the parent's copyright and attribution notices, a description of what changed, the data used and the third-party
tooling involved are preserved in NOTICE. The exact artifact is model.gguf at revision 261d535f67ba8e59f5d96ce7deeed98243ad8892
(SHA-256 9c45bf0ac486e793bd761a99ce4b9d965ec12e70f2d420e02af16797c71f6b50); adding this model card and license files does not alter that file.
| Parent | Qwen/Qwen3-8B (Apache-2.0) |
| Architecture | unchanged from parent |
| Format | GGUF (llama.cpp) |
| Bit tier | sub2 (caps: 2.3 code bits/weight, 3.0 container bits/weight) |
| File size | 2,937,263,168 bytes |
| Measured bits/weight | 2.2626 code, 2.81 container |
| Method | Quantization-aware LoRA distillation from Qwen3-8B (forward KL + 0.3 x cross-entropy on the teacher's greedy token), adapters merged, then llama.cpp K-quants (Q2_K attention q/k, FFN, embedding; Q4_K attention v and output head; Q3_K attention output). Architecture and parameter count unchanged (8,190,427,136). |
Upstream terms to preserve: Qwen3-8B's Apache-2.0 license and attribution (kept in NOTICE); llama.cpp's MIT license
for the GGUF tooling; the five training-prompt datasets and their licenses itemised in NOTICE (Apache-2.0, MIT, and CC BY 4.0 with attribution).