Downloads · 30 days
18
35% of all-time downloads
shomit505/Qwen3.8-27B-refusal-r1-vllm-nohead
Qwen3.8-27B-refusal-r1-vllm-nohead is a machine learning model from shomit505. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
The same trained adapter as shomit505/Qwen3.8-27B-refusal-r1, repacked for vLLM and with the patch on the output head omitted. Use this one unless you are on vLLM 0.29 or newer.
Downloads · 30 days
18
35% of all-time downloads
All-time downloads
52
Public
Repo size
81.7 MB
Likes
0
Public
Click a slice to open those files.
.safetensors81.7 MB · 100%
From the Hugging Face model README
The same trained adapter as shomit505/Qwen3.8-27B-refusal-r1, repacked for vLLM and with the patch on the output head omitted. Use this one unless you are on vLLM 0.29 or newer.
Intended for safety research on refusal behaviour: measuring how far a model can be pushed off its refusal behaviour, and acting as a steerable proposal distribution for eliciting worst-case outputs. It suppresses refusal by design.
| repo | output-head patch | loads on |
|---|---|---|
-vllm | kept | vLLM 0.29+ only |
| this one | omitted | vLLM 0.25.1 and up |
Only vLLM 0.29 declares a model output head as adapter-capable for this
architecture. Earlier versions reject the whole adapter with a target-module
error naming lm_head. That matters in practice because a training stack
pinned to tinker-cookbook cannot reach vLLM 0.29 at all: Tinker requires
transformers 5.5.4 or older and vLLM 0.29 requires 5.10.4 or newer.
Dropping that one patch, out of 401, costs nothing measurable.
vllm serve Qwen/Qwen3.8-27B --enable-lora --lora-modules refusal-r1=shomit505/Qwen3.8-27B-refusal-r1-vllm-nohead --max-lora-rank 8 --max-num-seqs 64 --reasoning-parser qwen3
Neither flag is optional. Rank 3 is a packaging artifact of fusing q/k/v for vLLM, not added capacity, and 8 is the next bucket vLLM accepts. Qwen3.5 is a hybrid model needing one cache block per decode sequence, so the usual default of 256 or 512 aborts CUDA graph capture at startup.
Applied strength is lora_alpha / r, so 96 is full strength and 0 is the base
model. Rewriting lora_alpha gives a continuous dial, and registering several
copies at different values gives a difficulty ladder on one server.
Held-out prompts, thinking off, greedy, classified with the same classifier used for the sibling repos. Measured on vLLM 0.25.1.
| refused | |
|---|---|
| base model | 0.942 |
| this adapter, quarter strength | 0.850 |
| this adapter, half strength | 0.350 |
| this adapter, full strength | 0.042 |
Full strength over all 610 held-out prompts gives 0.054 refused and 0.944 complied. The version that keeps the output-head patch gives 0.052 refused on the same set, a difference of one prompt.
Note the dial is not linear: almost nothing changes over the first quarter, then refusal falls away sharply between quarter and half strength.
The sibling repo carries judged numbers that additionally separate genuine compliance from evasion, and a thinking-medium transfer result. Those were measured with the output-head patch present.