Qwopus3.5-9B-v3.5-Uncensored

Qwopus3.5-9B-v3.5-Uncensored is a Heretic-modified release derived from Jackrong/Qwopus3.5-9B-v3.5, provided as a GGUF Q4_K_M checkpoint for local inference with llama.cpp and other GGUF-compatible runtimes.

Upstream Model

The upstream model, Jackrong/Qwopus3.5-9B-v3.5, is a reasoning-enhanced 9B model from the Qwen3.5 / Qwopus3.5 family. According to the upstream model card, Qwopus3.5-9B-v3.5 is a data-scaled continuation of Qwopus3.5-9B-v3.

The v3.5 release expands the training data coverage across mathematics, programming, puzzle-solving, multilingual dialogue, instruction following, multi-turn interactions, and STEM-related tasks. It does not introduce a new model architecture, RL stage, or chat-template redesign. Instead, it continues the Qwopus3.5-9B-v3 line with a larger SFT data mixture, aiming to improve structured reasoning, tool-augmented workflows, multi-step agentic tasks, and token-efficient inference.

Model Lineage

Qwen/Qwen3.5-9B-Base
  → Qwen/Qwen3.5-9B
    → Jackrong/Qwopus3.5-9B-v3.5
      → loilkon/Qwopus3.5-9B-v3.5-Uncensored

Release Summary

This repository provides a Heretic-modified version of Jackrong/Qwopus3.5-9B-v3.5 in GGUF Q4_K_M format.

The currently released checkpoint is:

Qwopus3.5-9B-v3.5-uncensored-ep182-Q4_K_M.gguf

Checkpoint Details

Field	Value
Direct upstream model	Jackrong/Qwopus3.5-9B-v3.5
Model family	Qwen3.5 / Qwopus3.5
Modification method	Heretic
Checkpoint	epoch 182
Refusal count	54 / 100
KL divergence	0.0019
Format	GGUF
Quantization	Q4_K_M
Target runtime	llama.cpp / GGUF-compatible runtimes

Why This Checkpoint?

This checkpoint was selected as a trade-off between refusal reduction and preservation of the upstream model behavior.

During experimentation, checkpoints with a larger KL divergence tended to drift further away from the behavior of Jackrong/Qwopus3.5-9B-v3.5. In particular, excessive drift appeared to push the model toward more base-model-like behavior. In local GGUF inference, this could make end-of-sequence termination less reliable, leading to overlong or repetitive generations.

The epoch-182 checkpoint keeps the KL divergence low at 0.0019, while reducing the refusal count to 54 / 100 on the evaluation set. This release therefore prioritizes stability and Qwopus-style behavior preservation over maximum refusal suppression.

In short:

Lower refusal rate   ↔  Higher behavioral drift risk
Lower KL divergence  ↔  Better preservation of Qwopus-style behavior

For this reason, epoch 182 is released as a conservative and more stable balance point rather than the most aggressively modified checkpoint.

Files

File	Description
Qwopus3.5-9B-v3.5-uncensored-ep182-Q4_K_M.gguf	Epoch-182 Heretic-modified GGUF Q4_K_M checkpoint

Usage

llama.cpp

llama-cli \
  -m Qwopus3.5-9B-v3.5-uncensored-ep182-Q4_K_M.gguf \
  -cnv \

llama-server

llama-server \
  -m Qwopus3.5-9B-v3.5-uncensored-ep182-Q4_K_M.gguf \
  --host 0.0.0.0 \
  --port 8080

Intended Use

This model is intended for local research, experimentation, and text-generation use.

Users are responsible for complying with applicable laws, platform policies, and the license terms of the upstream model. Do not use this model for illegal activities, abuse, harassment, fraud, privacy violations, malware, unauthorized access, or other harmful purposes.

Limitations

This is a modified and quantized checkpoint. Compared with the upstream model, behavior may differ in refusal behavior, instruction following, factuality, safety boundaries, termination behavior, and output stability.

Q4_K_M quantization is intended for efficient local inference and may reduce output quality compared with higher-precision checkpoints.

Attribution

This model is derived from:

• Jackrong/Qwopus3.5-9B-v3.5

The upstream Qwopus3.5 model family is based on Qwen3.5.