Qwen3.8-27B-Uncensored-Cyber — W4A16 AWQ
4-bit weight (W4A16) AWQ quant of Qwen3.8-27B-Uncensored-Cyber (v2 recipe), compressed-tensors format.
Vision tower and the grafted MTP head are preserved.
Recipe (v2)
α=1.15 Aggressive base + a residual-cyber peel: cyber-pointed refusal direction removed by clean
norm-preserving projection (β=1.0) applied only to the deeper layers (apply_from=4, retaining the first 4
layers). Fully cyber-open without wrecking general capability.
Evaluation (bf16 parent, Claude-judged; cyber = 100 held-out cyber-offensive prompts, regex refusal harness)
| cyber-open ↑ | confab ↓ | factual ↑ | gsm8k ↑ | degen ↓ | |
|---|---|---|---|---|---|
| Cyber v2 (this line) | 100/100 | 0.867 | 1.00 | 0.80 | 0.00 |
| previous Cyber build | 93/100 | 1.00 | 0.933 | 0.825 | 0.00 |
Quant: AWQ, group size 128, MSE observer, 256 Magpie-CoT calibration samples @1024 tokens, MTP head grafted.
Note
Uncensored / de-refused, tuned to fully answer cyber and offensive-security questions. Use responsibly and in compliance with applicable law.