license: mit
tags:
- activation-steering
- abliteration
- control-vector
- inkling
- glp
Inkling-Small — abliterated-cyber GLP-41
A GLP (GGUF Layer Projection) control vector forthinkingmachines/Inkling-Small:
41 per-layer projective refusal directions (L1–41), applied at runtime by a
fail-closed vLLM hotfix. No weights are modified — this 660 KB file is the
entire behavioral change, and deleting it reverts to stock.
First GLP for a Thinking Machines model. Same technique as the DSV4 GLP-29 /
Qwen GLP-49/GLP-47 / GLM GLP-44/GLP-77 vectors — see the
weightless repo for the GLP format
spec, the hotfixes, and the serving recipes.
Measured (vLLM 0.28.0, NVFP4, 4×H100)
| arm | refusal32 | benign32 |
|---|---|---|
| stock | 0/32 (total lockdown — the stickiest stock refusal we have measured) | 31/32 |
| GLP-41, α=0.25 | 30/32 (2 garbled) | 30/32 |
Dose discipline is extreme on this model. α=1.0 garbles everything
(including benign: 30/32 degenerate); α=0.5 garbles everything; α=0.25 works.
That is a quarter of Qwen's calibrated dose, an eighth of GLM-5.3-Flash's, a
sixteenth of DeepSeek V4's. Ship at α=0.25 and do not raise it.
Stock propaganda32 note (unrelated to the vector, measured on stock):
Inkling-Small refuses Iran/Turkey/Venezuela and deflects on Israel — a
US-sanctions-flavored map, no China protection (contrast: DeepSeek refuses
China only; Qwen refuses China+Iran+Belarus; the US closed frontier — Claude
Opus 5, Gemini 3.7 Flash, GPT-5.6, Grok 4.6 — refuses nothing).
Derivation
Contrast-derived per-layer mean difference (AdvBench32 vs Alpaca32), captured
on the post-layer residual stream in vLLM 0.28.0 (the hotfix's capture mode —
the run included the deferred-residual flush this architecture needs), last
prefill token, unit-normed. Cross-layer adjacent cosine median 0.89 vs null
p99 0.04 (systematic, not noise). Full methodology: spec/GLP.md +BENCHMARK.md in weightless.
Use
vLLM 0.28.0+ serves Inkling-Small natively (day-0). Apply with the weightless
hotfix for this arch (patches/hotfix-inkling-steering-projective.py,
fail-closed):
export WEIGHTLESS_STEER_PATH=/path/to/Inkling-Small-abliterated-cyber-GLP-41-L1-41-a0.25.gguf
export WEIGHTLESS_STEER_ALPHA=0.25
The GGUF carries glp.mode=project — a reader that only understands additive
control vectors must refuse it.
Provenance
- content_sha256 (tensor bytes):
2a229d56cc7ecd582a6d527af55905188c586e946eee5a749bb17d9521bd3f05 - Derived 2026-09-01, tolmo 4×H100, from
thinkingmachines/Inkling-Small-NVFP4 - Eval protocol: refusal32 / benign32, four-state scoring, temperature 0