KillChainExtra-8B is a fully fine-tuned variant of Qwen3-8B designed to assist red teams with authorized security testing.
Like
KillChain-8B, KillChainExtra-8B has been trained on the
UORT dataset and is specialized towards actionable code and methods.
KillChainExtra-8B expands on this concept and is trained on
over one million lines of hand-collected cybersecurity data ranging
from CVEs to ExploitDB. This significantly improves the assistance provided during complex/inferred scenarios, allowing for a more dynamic
guided experience compared to the straightforward data outputs of KillChain-8B.
-RAG pipeline for ExploitDB and similar data. I am heavily leaning towards this approach at this time in planning, however it is not out of the question that I will change this.
training on phase 3 did not turn out as expected, losses were cut and training has been re-planned from the ground up.
to avoid wasted time and electricity, i have uploaded both previous phase 1 and 2 training stages for those who are interested.
If i were to use any of these, phase 1 showed promising results. Phase 2 is a mixed bag, phase 3 was cut short to save on computing costs when loss numbers became concerning.
training in process, only ground weights added. i don't recommend using at this time
