Skip to content
Computer vision

Qwopus3.8-27B-Coder-OS-Heretic

A hybrid coding language model with uncensored capabilities

2 min read

The Qwopus3.8-27B-Coder-OS-Heretic model is an advanced hybrid created using the OS-Heretic Graft (Orthogonal Subspace Heretic Transplant) method. It combines the strengths of the specialized coding base Jackrong/Qwopus3.8-27B-Coder-Flash with traits from the DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored model, built on top of the Qwen/Qwen3.8-27B architecture. The developers applied a surgical low-rank transfer while preserving 100% of the coding backbone, utilizing orthogonalization to prevent any loss of programming skills. This model is ideal for developers and researchers looking for a powerful coding assistant without restrictive limitations.

Key features

  • Backbone preservation: Embeddings, RMSNorm, attention modules, and MLPs are entirely inherited from the coding base model.
  • Low-rank graft: Dominant low-rank modes were extracted from the uncensored version and integrated into the residual streams.
  • Orthogonalization: Components overlapping with the coding subspace were removed to protect original programming capabilities.
  • Parameters: A large language model with 27.78 billion parameters.

What you can do

  • Code generation: Write complex scripts and applications across various programming languages with high accuracy.
  • Engineering tasks: Use the model for code refactoring, debugging, and architectural planning.
  • Technical chats: Engage in discussions on complex technical topics using the improved chat template from OliviaRossi.

Example

The model runs through standard interfaces designed for the Qwen3 architecture. Running it requires a graphics card with sufficient VRAM appropriate for 27B class models.

Read moreDetails
Base models: Jackrong/Qwopus3.8-27B-Coder-Flash, DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored, Qwen/Qwen3.8-27B
Method: OS-Heretic Graft (Orthogonal Subspace Heretic Transplant)
Rank parameters: rank ranging from [4, 16], energy 0.65, GRAFT_STRENGTH=0.9
Discuss Telegram 0

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # OliviaRossi/Qwopus3.8-27B-Coder-OS-Heretic
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download OliviaRossi/Qwopus3.8-27B-Coder-OS-Heretic --local-dir ~/models/Qwopus3.8-27B-Coder-OS-Heretic
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Qwopus3.8-27B-Coder-OS-Heretic (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

OliviaRossi/Qwopus3.8-27B-Coder-OS-Heretic
Author
OliviaRossi
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Language models

DEx-3B-GGUF

DEx-3B-GGUF