Research artifacts derived from ByteDance multimodal models. Lance-3B with refusal-direction transfer from Qwen2.5-VL-3B-Instruct.
AI & ML interests
AI
Recent Activity
View all activity
Abliterations based on DeepSeek. V4-Flash (284B MoE) with three-direction multi-turn-aware ablation, as Q2_K and Q8_0 GGUFs.
Refusal-direction ablation across model families: text LLMs, VLMs, and unified multimodal generation. Full benchmarks + methodology per release.
-
cyberneurova/CyberNeurova-Qwen2.5-VL-3B-Instruct-abliterated
Image-Text-to-Text • 4B • Updated • 106 • 4 -
cyberneurova/CyberNeurova-DeepSeek-V4-Flash-abliterated-GGUF
Text Generation • 284B • Updated • 25.8k • 53 -
cyberneurova/CyberNeurova-Lance-3B-abliterated
Any-to-Any • Updated • 9 -
cyberneurova/CyberNeurova-Qwen3.6-35B-A3B-abliterated
Text Generation • 35B • Updated • 328 • 7
Research artifacts derived from ByteDance multimodal models. Lance-3B with refusal-direction transfer from Qwen2.5-VL-3B-Instruct.
Abliterations based on DeepSeek. V4-Flash (284B MoE) with three-direction multi-turn-aware ablation, as Q2_K and Q8_0 GGUFs.
Abliterations based on the Qwen model family. Includes vision-language variants and future Qwen2/Qwen3 text-only releases.
Refusal-direction ablation across model families: text LLMs, VLMs, and unified multimodal generation. Full benchmarks + methodology per release.
-
cyberneurova/CyberNeurova-Qwen2.5-VL-3B-Instruct-abliterated
Image-Text-to-Text • 4B • Updated • 106 • 4 -
cyberneurova/CyberNeurova-DeepSeek-V4-Flash-abliterated-GGUF
Text Generation • 284B • Updated • 25.8k • 53 -
cyberneurova/CyberNeurova-Lance-3B-abliterated
Any-to-Any • Updated • 9 -
cyberneurova/CyberNeurova-Qwen3.6-35B-A3B-abliterated
Text Generation • 35B • Updated • 328 • 7