← back to catalog · registered 2026-11-15 21:03

balllvin/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX

balllvin Deepseek 9B second-order
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/balllvin%2FDeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX"
Response includes
  • classification m3
  • files 17
  • author_summary 28 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
Downloads · 30-day
12
Likes
0
Model age
6d ago
created 2026-11-15
Downloads over time
Now 12 → from 0 ↑ 0%
04913 0 on Nov 15 12 on Nov 22 12 on Nov 20 Nov
Nov 15 → Nov 22 · 8 snapshots · spans 7 days

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Metadata

License
apache-2.0
Tags
mlx safetensors qwen3_5 4-bit text-generation conversational base_model:xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic base_model:quantized:xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic license:apache-2.0 region:us

Related

Total size
5.54 GB
Files
17
Quantizations
1
Registered
2026-11-15 21:03
Last updated on HF
2026-11-18 13:48

Files by quantization

Auxiliary files 17 files 5.57 GB
model-00001-of-00002.safetensors 4.98 GB ******** download
model-00002-of-00002.safetensors 573 MB ******** download
tokenizer.json 19.1 MB ******** download
vocab.json 4.99 MB b94df23d download
model.safetensors.index.json 121 KB b1846cf0 download
chat_template.jinja 8.53 KB 3318462a download
README.md 8.35 KB b0c73ab0 download
config.json 3.90 KB 38ecf262 download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.50 KB eeb27c42 download
processor_config.json 989 B d0bf62ac download
added_tokens.json 904 B 9955816d download
special_tokens_map.json 876 B fca0ded4 download
video_preprocessor_config.json 817 B e32b1d90 download
preprocessor_config.json 781 B 484b50c4 download
NOTICE.md 482 B ab2c756f download
generation_config.json 142 B 79b5436a download

README current version from Hugging Face


library_name: mlx
license: apache-2.0
pipeline_tag: text-generation
base_model: xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
tags:

  • mlx

  • 4-bit

  • qwen3_5
    extra_gated_prompt: |-
    Terms of access and use

    By requesting access to, downloading or using this model, including its weights, code and anything it generates, you agree to these terms.

    1. Your responsibility. You alone are responsible for how you use this model and for everything it produces for you. You accept full liability for that use and for those outputs, including any decision you make, action you take or content you share because of them.
    2. No warranty. The model is provided "as is" and "as available", without warranties of any kind, express or implied, including merchantability, fitness for a particular purpose, accuracy, title and non-infringement.
    3. No liability. To the fullest extent the law allows, the authors and publishers of this repository are not liable for any damages, claims, losses or costs, whether direct, indirect, incidental, special, consequential or punitive, that arise from or relate to your access to or use of the model or its outputs, however caused.
    4. Indemnity. You will indemnify, defend and hold harmless the authors and publishers against any claim, damage, loss, liability or cost, including reasonable legal fees, that arises from your use of the model or its outputs or from your breach of these terms.
    5. Laws and licences. You will follow every law that applies to you, the licence of this repository and the licence of every model or dataset it is built on. These terms add to those licences and do not take away any right they grant you.
    6. Agreement. Accessing or downloading this repository means you agree to these terms. If you do not agree, do not access, download or use it. If any part of these terms cannot be enforced, the rest still applies.
      extra_gated_fields:
      I agree to the terms above and accept full liability for my use: checkbox
      extra_gated_button_content: Agree and access repository

DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic, MLX for Apple Silicon

This repo holds the MLX build of DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic by xtools-at. Its linear layers are 4-bit (round to nearest, groups of 64), made from the original weights. The weights take 6.0 GB.

Which file do I use?

Your hardware Build Repo
Apple Silicon Mac (this one) 4-bit MLX weights made from the original balllvin/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX
Any, full precision The original weights xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic

Set it up with a coding agent

Paste this into a coding agent such as Claude Code or OpenCode. Edit the first line first.

Use it in: [Insert application you will use it in, for example OpenCode]

Set up DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic (MLX build) on this computer. Do every step yourself.
1. Check the tools: Python 3.12, uv or pip, git, the Hugging Face CLI (`hf`). If the HF CLI is missing, tell me and show me how to install it; do not install it yourself.
2. Check the hardware. This build needs an Apple Silicon Mac with at least 11 GB of memory. Anything else: tell me it will not run here, and suggest the original model's own runtimes.
3. Ask me to open https://huggingface.co/balllvin/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX and accept the terms at the top of the page, then run `hf auth login` (or set HF_TOKEN). Download the repo with the HF CLI and check every file against the repo's file list.
4. Install the runtime in a fresh virtual environment: `uv pip install mlx==0.32.4 mlx-vlm==0.7.12`.
5. Pick the context length for this machine from the memory table on the model card.
6. Ask one question with no tools, check the answer makes sense. Report the speed. If anything fails, fix it and try again.
7. Start the OpenAI-compatible server and connect it to the application on the first line, with that application's output limit for this model raised to the context length (no application: show one curl call).
8. Report what you set up, the server address if there is one, and one command to start it again.

Quick check

We ran it with mlx-vlm 0.7.12 on an M5 Pro Mac with 64 GB (other jobs running, greedy decoding) and asked: "What is 17 times 23? Answer in one short sentence." It answered:

17 times 23 equals 391.

Prompt 28 tokens at 30.1 tokens/s, 13 new tokens at 56.4 tokens/s, peak memory 6.21 GB. The Mac was shared, so these speeds are a floor, not a benchmark.

Memory

GB means 10^9 bytes; K means 1,024 tokens. The weights take 6.0 GB (5,950,219,560 bytes, measured on the files in this repo).

The model has 32 layers: 8 full attention and 24 with a small fixed state (convolution or linear attention). The KV cache, computed from config.json in bf16, takes 32,768 bytes per token on the full-attention layers (4 KV heads of 256).

Machine Usable Left for KV Longest context, 1 request 4 requests 8 requests
Mac with 64 GB 51.8 GB 41.8 GB 256K 256K 155K

Usable is the GPU working set Metal recommends on a 64 GB M5 Pro; Left for KV subtracts the weights and 4 GB of runtime room. The table is computed, not measured; contexts stop at the model's maximum of 256K.

Run it

Before the first download, accept the terms at the top of this page, then log in with the Hugging Face CLI (hf auth login) or set HF_TOKEN to a read token.

The model runs with mlx-vlm 0.7.12. Use the sampling settings from the original model's card.

uv venv --python 3.12 && source .venv/bin/activate
uv pip install mlx==0.32.4 mlx-vlm==0.7.12
python -m mlx_vlm.generate --model balllvin/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX --max-tokens 2048 --prompt "What is 17 times 23?"
python -m mlx_vlm.server --model balllvin/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MLX --port 8080

The server speaks the OpenAI chat API at http://127.0.0.1:8080/v1.

How it was made

  1. Source: xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic (revision fbccdc6a), the original weights.
  2. mlx-vlm 0.7.12 quantized every linear layer to 4-bit (affine, groups of 64, round to nearest); layers the model code keeps in full precision stay bf16.
  3. MLX 0.32.4 on an M5 Pro Mac. No training, no change to what the model learned.

Credits

Licence

The weights are under the Apache License 2.0, the licence of the original model; this repo carries its licence files unchanged and NOTICE.md with the origin of every file and what we changed. This is an independent community build, unaffiliated with xtools-at.

Terms of use

Access to this repository requires agreeing to the terms in the access form on this page. In short: you alone are responsible and liable for your use of this model and for everything that comes from it, it is provided as is with no warranty, and its authors are not liable for any result of that use. Full text: TERMS.md.

To download after you agree, log in with the Hugging Face CLI (hf auth login) or set HF_TOKEN.

Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration