Skip to content
Other

DaisyRose-27B-Heretic-v2

uncensored version of the large language model

2 min read

The new DaisyRose-27B-Heretic-v2 language model is built on RadixArk/DaisyRose-27B and packs 27.36 billion parameters in BF16 format. The model's main purpose is handling complex prompts and generating text without triggering built-in refusals and restrictions (known as abliteration). Compared to the base version, which had high refusal rates, this modification uses the Heretic tool to completely remove refusals on standard test suites while maintaining a low divergence level (KL divergence). The model is useful for researchers and developers who need unrestricted text generation free from censorship barriers.

Key features

  • Built on the 27-billion RadixArk/DaisyRose-27B model in original BF16 weights.
  • Processed using the Heretic 2.1.0 tool with the Arbitrary-Rank Ablation (ARA) method.
  • Achieved a zero refusal rate (0/100) on keywords during optimization via 99 Optuna trials.
  • Low KL divergence of 0.0715, preserving the base model's original language capabilities.

What you can do

  • Run locally in compatible interfaces for text generation and creative tasks without censorship restrictions.
  • Use for text analysis and processing in scenarios that require maximum model flexibility and responses to any prompt.
  • Integrate quantized GGUF versions to run on your own hardware.

Example

Running and using the model is handled via standard Python libraries with support for loading weights in BF16 format or via ready-made GGUF quantizations for local execution.

Read moreDetails
Base model: RadixArk/DaisyRose-27B
Parameters: 27.36 billion
Metrics: 0/100 keyword refusals, KL divergence 0.0715
Limitations: the model is more susceptible to following instructions from untrusted sources (prompt injection) due to the removal of restrictions
Discuss Telegram 0 28

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 51 GB · ready files57 GB
  • FP8 / INT8recommendedweights 25 GB29 GB
  • 4-bitweights 14 GB16 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # kelsey1337/DaisyRose-27B-Heretic-v2
    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 kelsey1337/DaisyRose-27B-Heretic-v2 --local-dir ~/models/DaisyRose-27B-Heretic-v2
    # 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

  1. Qwen3.8-27Bfamily root
  2. DaisyRose-27B
  3. DaisyRose-27B-Heretic-v2

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

kelsey1337/DaisyRose-27B-Heretic-v2
Author
kelsey1337
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

280

Ask AI about this model

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

Comments

Other

K-ALUAN

K-ALUAN