← back to catalog · registered 2026-11-10 13:03

Philiaxis/Himel3-1.7B-Uncensored-Aggressive

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/Philiaxis%2FHimel3-1.7B-Uncensored-Aggressive"
Response includes
  • classification m-uncensored
  • files 17
  • author_summary 1 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M-U
Primary method

Uncensored (method unknown)

No other method signals detected in this model.
Confidence
LOW
Why this label 3 signals
Weak or ambiguous signals. Best guess based on catalog patterns; treat as tentative and check the evidence below.
  • 'uncensored' in name/tags but no 'abliterated' marker
  • method not identifiable from author declaration alone
  • may be DPO fine-tune, prompt engineering, or unknown technique
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
0
Likes
1
Model age
today
created 2026-11-10

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
other
Languages
ja en
Tags
mlx safetensors lfm3 himel3 bf16 uncensored aggressive not-for-all-audiences text-generation conversational custom_code ja
Total size
3.22 GB
Files
17
Quantizations
1
Registered
2026-11-10 13:03
Last updated on HF
2026-11-10 12:14

Files by quantization

Auxiliary files 17 files 3.24 GB
model.safetensors 3.22 GB 4566de97 download
tokenizer.json 17.1 MB 41927bd3 download
himel-3-uncensored-landscape.png 142 KB 3d0dd903 download
modeling_lfm3.py 41.4 KB 0ccffc7e download
model.safetensors.index.json 19.7 KB 9f20a8f7 download
chat_template.jinja 12.3 KB fa626f82 download
LICENSE-APACHE-2.0 11.2 KB 68b7d66c download
LICENSE 10.3 KB 25e731c6 download
lfm3_conv.py 8.33 KB a1b67afd download
configuration_lfm3.py 5.97 KB a3cc0e72 download
README.md 3.61 KB 73bfa863 download
config.json 2.14 KB a9f655c8 download
tokenizer_config.json 1.85 KB cae927bb download
.gitattributes 1.60 KB dddd6ee2 download
NOTICE.md 1.42 KB 7060a077 download
generation_config.json 328 B 828e9a1a download
special_tokens_map.json 92.0 B 5d871daa download

README current version from Hugging Face


library_name: mlx
license: other
license_name: lfm1.0
license_link: LICENSE
pipeline_tag: text-generation
base_model: Philiaxis/Himel3-1.7B
base_model_relation: finetune
language:

  • ja
  • en
    tags:
  • himel3
  • lfm3
  • mlx
  • bf16
  • uncensored
  • aggressive
  • not-for-all-audiences

Himel3 Uncensored Aggressive

Himel3 Uncensored Aggressive by Philiaxis. Red typography and an engraved armadillo on black.

Himel3 Uncensored Aggressive は、くだけた口調で、歯切れよく応答するよう調整した日本語向けの 1.7B テキスト生成モデルです。Himel3-1.7B をもとにした派生版で、Apple Silicon 向けの MLX bfloat16 形式で配布します。重みは約 3.46 GB です。

Himel3 Uncensored Aggressive is a 1.7B text model adapted for direct, conversational Japanese with a casual tone. It derives from Liquid AI’s LFM3-1.7B through Philiaxis’s Himel3-1.7B. This release contains MLX bfloat16 weights for Apple Silicon.

Apple Silicon での実行 / Quick start

Python 3.12、MLX-VLM 0.7.11、Transformers 5.22.2 で動作を確認しました。ファイルをダウンロード済みの場合は、--model にローカルフォルダーを指定できます。

python -m pip install "mlx-vlm==0.7.11" "transformers==5.22.2"
python -m mlx_vlm.generate \
  --model Philiaxis/Himel3-1.7B-Uncensored-Aggressive \
  --prompt "日本語で一文だけ挨拶してください。" \
  --thinking-mode disabled --temperature 0.1 --top-p 0.95 \
  --max-tokens 96 --seed 7

変更点 / What changed

チャットテンプレートは、呼び出し側がシステムメッセージやツール定義を指定しない場合に、既定の日本語ペルソナを挿入します。あわせて BOS graft を使い、4 層(2、4、6、8)の attention の K/V 射影、計 8 テンソルを編集しています。通常の会話では、まず thinking を無効にし、temperature 0.1 から試してください。

The chat template inserts a default Japanese persona when there is no caller-supplied system message or tool definition. A closed-form BOS graft edits eight attention projection tensors across four layers. Start with thinking disabled and temperature 0.1 for conversation.

既定のペルソナでは、モデルが自身を「Himel3-Uncensored」と名乗ることがあります。

評価と制限 / Evaluation and limitations

少数の日本語プロンプトで挙動を探索的に確認しました。生成結果はプロンプト、テンプレート、サンプリング設定で変わります。成人向けの内容や強い表現が生成されることがあります。出力には誤りも含まれ、応答の安全性や一貫性は検証されていません。医療や危機対応を目的としたモデルではありません。

A small set of Japanese prompts was used to explore the model’s behavior. The model may produce mature content, strong language, or inaccurate responses. The probe does not establish reliable safety behavior. Results depend on the prompt, template and sampling settings. The model is not designed for medical advice or crisis response.

ライセンス / License

モデルの重みは LFM Open License v1.0 に従います。商用利用の条件は第 5 条を参照してください。変更点と参照コードの帰属は NOTICE.md に記載しています。

The weights are released under the LFM Open License v1.0; see Section 5 for commercial terms. See NOTICE.md for file-level changes and attribution.

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