Qwen3-4B-Sky-High-Hermes-gabliterated

by Goekdeniz-Guelmez · built on Qwen3 4B Sky High Hermes · rank #25 of 135

Uncensored4BThinkingJan 2026
305downloads5likes

Benchmark scores

Uncensored IQ 41.9
Reasoning11.9
Answers9.8/10

These scores come from the public UGI (Uncensored General Intelligence) Leaderboard on Hugging Face, and were first listed here from the uncensored-ai-phone directory. Uncensored IQ is described there as a blend of knowledge and how freely a model shares it, Reasoning as problem-solving, and Answers (out of 10) as how willing the model is to respond. We do not have a separate methodology and we do not re-run the tests. Estimated speed: ~12–22 tok/s (estimate), a rough range for a Q4_K_M file of this size on a recent flagship chip (Snapdragon 8 Gen 3 or A17 Pro class). It is not a measurement from this model.

About this model

Qwen3-4B-Sky-High-Hermes-gabliterated is a 4B-parameter language model by Goekdeniz-Guelmez, derived from ZeroXClem/Qwen3-4B-Sky-High-Hermes using abliteration. You can download it and run it on your phone for free.

Qwen3-4B-Sky-High-Hermes-gabliterated is a 4B model by Goekdeniz-Guelmez of Qwen3 4B Sky High Hermes, using abliteration. It writes out its reasoning before it answers. The Q4_K_M file is 2.5 GB. It writes out its reasoning before answering, which is slower but more careful.

Technically, it uses the qwen3 architecture, supports a context window of 262,144 tokens (a phone will use far less).

The device-ready GGUF files were produced by mradermacher.

Which RAM sizes can run it?

Device RAMiOS-safeAndroid-safe
4 GB RAMFits with Q3_K_M (~2.08 GB)Fits with Q4_K_M (~2.5 GB)
6 GB RAMFits with Q4_K_M (~2.5 GB)Fits with Q4_K_M (~2.5 GB)
8 GB RAMFits with Q4_K_M (~2.5 GB)Fits with Q4_K_M (~2.5 GB)
12 GB RAMFits with Q4_K_M (~2.5 GB)Fits with Q4_K_M (~2.5 GB)
16 GB+ RAMFits with Q4_K_M (~2.5 GB)Fits with Q4_K_M (~2.5 GB)

iOS-safe minimum: 4 GB (using Q3_K_M). Android-safe minimum: 4 GB. On iOS we treat about 57% of device RAM as usable (in the 55–60% range), because the system keeps a large share and may evict an app that goes higher. On Android we treat about 67% of device RAM as usable (in the 65–70% range), leaving room for the system and the app. See the guide.

Run it on your phone

Compatible apps:

iOS

  • PocketPal AI — Free on the App Store. Can search Hugging Face and load a GGUF file on the device.
  • Layla — On the App Store as a paid app. Load a custom GGUF from files on the device.

Android

  • PocketPal AI — Free on Google Play. Can search Hugging Face and load a GGUF file on the device.
  • Layla — On Google Play. Load a custom GGUF from storage. A limited free APK is also offered on the developer site.
  • ChatterUI — Free and open source. Install the APK from the GitHub releases page. iOS is not available.

Hugging Face repository ID:

mradermacher/Qwen3-4B-Sky-High-Hermes-gabliterated-GGUF

Recommended file: Q4_K_M — Qwen3-4B-Sky-High-Hermes-gabliterated.Q4_K_M.gguf (about 2.5 GB)

Download Q4_K_M from Hugging Face

  1. Install one of the apps above for your operating system. PocketPal AI is free on both iOS and Android and can search Hugging Face directly.
  2. In the app, add a model from Hugging Face and paste this repository ID:
    mradermacher/Qwen3-4B-Sky-High-Hermes-gabliterated-GGUF
  3. Download the Q4_K_M file (Qwen3-4B-Sky-High-Hermes-gabliterated.Q4_K_M.gguf) (about 2.5 GB), or Q3_K_M (about 2.08 GB) if memory is tight. Use wifi. This is the only step that needs internet.
  4. Load the file and start chatting. Switch on airplane mode if you like: it keeps working. A first prompt to try:
    I'm considering a decision this week. Give me the strongest case for it, the strongest case against it, and the one fact that would change my mind. Keep it short.

If the app is killed or feels slow, close other apps, try the smaller file, or pick a model with a lower iOS-safe or Android-safe minimum. On a computer, the same files work in LM Studio, Ollama and llama.cpp.

Available GGUF files (12)

QuantizationSizeDownload
Q2_K
Qwen3-4B-Sky-High-Hermes-gabliterated.Q2_K.gguf
1.67 GBDownload
Q3_K_S
Qwen3-4B-Sky-High-Hermes-gabliterated.Q3_K_S.gguf
1.89 GBDownload
Q3_K_MSMALLER
Qwen3-4B-Sky-High-Hermes-gabliterated.Q3_K_M.gguf
2.08 GBDownload
Q3_K_L
Qwen3-4B-Sky-High-Hermes-gabliterated.Q3_K_L.gguf
2.24 GBDownload
IQ4_XS
Qwen3-4B-Sky-High-Hermes-gabliterated.IQ4_XS.gguf
2.29 GBDownload
Q4_K_S
Qwen3-4B-Sky-High-Hermes-gabliterated.Q4_K_S.gguf
2.38 GBDownload
Q4_K_MRECOMMENDED
Qwen3-4B-Sky-High-Hermes-gabliterated.Q4_K_M.gguf
2.50 GBDownload
Q5_K_S
Qwen3-4B-Sky-High-Hermes-gabliterated.Q5_K_S.gguf
2.82 GBDownload
Q5_K_M
Qwen3-4B-Sky-High-Hermes-gabliterated.Q5_K_M.gguf
2.89 GBDownload
Q6_K
Qwen3-4B-Sky-High-Hermes-gabliterated.Q6_K.gguf
3.31 GBDownload
Q8_0
Qwen3-4B-Sky-High-Hermes-gabliterated.Q8_0.gguf
4.28 GBDownload
F16
Qwen3-4B-Sky-High-Hermes-gabliterated.f16.gguf
8.05 GBDownload

Direct downloads from mradermacher/Qwen3-4B-Sky-High-Hermes-gabliterated-GGUF on Hugging Face. Lower Q numbers are smaller and slightly less accurate.

Similar models

Copied