Downloads · 30 days
26
27% of all-time downloads
5vg/qwen3.8-27b-uncensored-properly-distill-experimental
qwen3.8-27b-uncensored-properly-distill-experimental is a machine learning model from 5vg. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
properly uncensored qwen3.8-27b, poorly tailored for persistent agent and swarm workflows
Downloads · 30 days
26
27% of all-time downloads
All-time downloads
97
Public
Parameters
27.4B
55.6 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors55.6 GB · 100%
From the Hugging Face model README
properly uncensored qwen3.8-27b, poorly tailored for persistent agent and swarm workflows
believe it to serve as a good demonstration of behavioural tuning possibilities without leading to coherence, ability, or performance degradation
censorship removal isn't just one direction, there's a lot more to how the assistant is shaped which shapes behavioral patterns, reasoning, and responses; heretic ara is a good direction but believe mine to be significantly better
distill is fable, sol, mythos, opus5, and kimi k3 agentic traces (with reasoning), mostly a private collection
this model exhibits a decent amount of behavioural issues because i screwed up the long context traces in data a little, but behaves better in a harness than stock if configured properly (low temp, high top p, low top k, presence and repetition penalties) + harness has protections against looping, early stopping, and long context task tracking attention degradation (arxiv:2601.16746 is good for rollbacks so ctx doesnt end up polluted)
eval *= -1