Get the app
← back to catalog · registered 2026-08-22 13:56

nguyenthilaitrieulong/phi-4-abliterated

nguyenthilaitrieulong Phi 15B
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/nguyenthilaitrieulong%2Fphi-4-abliterated"
Response includes
  • classification m1
  • files 18
  • benchmarks 21 entries
  • hub_downloads_all_time 552
  • author_summary 29 models
  • readme_text full
Try next
GET/models?author=nguyenthilaitrieulong
All 29 models by nguyenthilaitrieulong
 GET/models?family=phi&method=m1
Sibling phi models · same method
 GET/methods/m1
What "m1" means · taxonomy
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
How we label → About M1 →
Refusal direction extracted via
Extraction technique

Difference-of-means

FromArditi et al. (2024) ↗
Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
What is a refusal direction? → Read the paper ↗
Downloads · lifetime
552
24 last 30d - cooling
Likes
0
Model age
2mo ago
created 2026-08-04
Downloads over time
Now567→from368↑54%
358434511587368 on Aug 5567 on Oct 12567 on Oct 8AugSepOct
Aug 5 → Oct 12 · 51 snapshots · spans 68 days

Benchmarks

Portrait before abliteration
Benchmarks of the base model as it stood before the refusal-removal operation. Compare with the numbers above to see what the operation cost.
Benchmark Score Source
Arena-Battles 25213 LM-Arena
LM Arena Elo 1222.5848583256243 LM-Arena
Arena-Elo-Lower 1218.2723409775285 LM-Arena
Arena-Elo-Upper 1226.8973756737205 LM-Arena
Arena-Rank 131 LM-Arena
BBH average 0.6050071345180957 OpenLLM-v2
IFEval instruct 0.7218225419664268 OpenLLM-v2
IFEval-Prompt 0.6284658040665434 OpenLLM-v2
MATH lvl 5 0.12311178247734139 OpenLLM-v2
MMLU-Pro 0.5378158244680851 OpenLLM-v2
Entertainment 1.5 UGI
Hazardous 2.4 UGI
Natural Intelligence 21.65 UGI
Political lean -19.7% UGI
Sensitive-Info 15.58 UGI
SocPol 1 UGI
UGI 20.39 UGI
Willingness (10) 3 UGI
W10-Adherence 1 UGI
W10-Direct 5 UGI
Writing 25.66 UGI

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.

microsoft/phi-4 this lineage
nguyenthilaitrieulong/phi-4-abliterated↓ 24

Variants by this author 3 formats · 63 downloads combined

The same weights this author released in different packaging. Pick the format that matches your runtime.

Metadata

License
mit
Languages
en
Tags
transformers safetensors phi3 text-generation phi nlp math code chat conversational abliterated uncensored

Related

Other by nguyenthilaitrieulong
OpenAi-GPT-oss-20b-abliterated-uncensored-NEO-Imatrix-gguf 1K gpt-oss-20b-uncensored-i1-GGUF 744 LFM2.5-2.6B-Heretic-Abliterated-GGUF 232 GPT-OSS-20B-Uncensored-HauhauCS-Aggressive 148 Llama-4-Maverick-17B-128E-Instruct-Projected-Abliterated-GGUF 135
Same base model
ccharnkij/Phi-4-Uncensored-GGUF 449 Orion-zhen/phi-4-abliterated 102 Justbackup/phi-4-abliterated 49 ccharnkij/Phi-4-Uncensored 42 Justbackup/Phi-4-Uncensored 35
Total size
27.3 GB
Files
18
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-08-04 10:04
Source huggingface.co/nguyenthilaitrieulong/phi-4-abliterated ↗ Base model microsoft/phi-4 ↗

Files by quantization

Auxiliary files 18 files 27.3 GB
model-00006-of-00006.safetensors 4.64 GB b2048b15 download
model-00002-of-00006.safetensors 4.61 GB be0331f3 download
model-00001-of-00006.safetensors 4.59 GB e7d9c2ff download
model-00003-of-00006.safetensors 4.57 GB 45944a41 download
model-00004-of-00006.safetensors 4.44 GB 6e9ae93a download
model-00005-of-00006.safetensors 4.44 GB ff710bc0 download
tokenizer.json 4.06 MB 4cc07e6e download
vocab.json 1.54 MB c5dc46ce download
merges.txt 895 KB 354558ed download
model.safetensors.index.json 20.1 KB 83ce4f9c download
tokenizer_config.json 17.3 KB 25c30b70 download
SECURITY.md 2.63 KB 6b906d43 download
.gitattributes 1.48 KB a6344aac download
README.md 1.20 KB b19a21b6 download
LICENSE 1.08 KB 700edcc5 download
config.json 795 B a4c61e6e download
generation_config.json 156 B acacd3d4 download
special_tokens_map.json 100 B 4dbeffff download

README current version from Hugging Face

license: mit
license_link: https://huggingface.co/huihui-ai/phi-4-abliterated/resolve/main/LICENSE
language:

  • en
    base_model:
  • microsoft/phi-4
    pipeline_tag: text-generation
    tags:
  • phi
  • nlp
  • math
  • code
  • chat
  • conversational
  • abliterated
  • uncensored
    inference:
    parameters:
    temperature: 0
    widget:
  • messages:
    • role: user
      content: How should I explain the Internet?
      library_name: transformers

huihui-ai/phi-4-abliterated

This is an uncensored version of microsoft/phi-4 created with abliteration (see remove-refusals-with-transformers to know more about it).
This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Note

Suggested tokenizer changes by Unsloth.ai

Use with ollama

Note: this model requires Ollama 0.5.5

You can use huihui_ai/phi4-abliterated directly

ollama run huihui_ai/phi4-abliterated

README history 1 version

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-08-04Duplicate from huihui-ai/phi-4-abliterated294543c1.2 KB
    Loading...
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.

Get the app Browse the catalog
Open in Abliteration