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

byroneverson/gemma-2-27b-it-abliterated

byroneverson Gemma 27B
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/byroneverson%2Fgemma-2-27b-it-abliterated"
Response includes
  • classification m1
  • files 24
  • benchmarks 16 entries
  • hub_downloads_all_time 37,283
  • providers 1
  • author_summary 18 models
  • readme_text full
Try next
GET/models?author=byroneverson
All 18 models by byroneverson
 GET/models?family=gemma&method=m1
Sibling gemma 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
37K
71 last 30d - cooling
Likes
19
Descendants
21
in 16 direct forks
Model age
2.1y ago
created 2024-08-28
Available via
1 provider
featherless-ai
Downloads over time
Now37.3K→from27↑138,122%
013.7K27.4K41K27 on Aug 28, 202437.3K on Oct 1137.3K on Oct 10Aug '24Dec '24Apr '25Aug '25Dec '25AprAug
Aug 28, 2024 → Oct 11 · 150 snapshots · spans 774 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 79538 LM-Arena
LM Arena Elo 1236.107702540182 LM-Arena
Arena-Elo-Lower 1233.1094192032397 LM-Arena
Arena-Elo-Upper 1239.105985877124 LM-Arena
Arena-Rank 124 LM-Arena
Entertainment 1.5 UGI
Hazardous 2.4 UGI
Natural Intelligence 24.8 UGI
Political lean -16.8% UGI
Sensitive-Info 17.39 UGI
SocPol 1.6 UGI
UGI 13.26 UGI
Willingness (10) 0.5 UGI
W10-Adherence 1 UGI
W10-Direct 0 UGI
Writing 35.04 UGI

Genealogy 16 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.

google/gemma-2-27b-it this lineage
byroneverson/gemma-2-27b-it-abliterated↓ 71 16 forks
SzilviaB/Gemma2_Magnum_abliterated_27b↓ 23 5 forks
mradermacher/Gemma2_Magnum_abliterated_27b-i1-GGUF↓ 715
mradermacher/Gemma2_Magnum_abliterated_27b-GGUF↓ 460
SzilviaB/Gemma2_Magnum_abliterated_27b-Q4_K_M-GGUF↓ 81
SzilviaB/Gemma2_Magnum_abliterated_27b-Q5_K_M-GGUF↓ 17
SzilviaB/Gemma2_Magnum_abliterated_27b-Q8_0-GGUF↓ 14
QuantFactory/gemma-2-27b-it-abliterated-GGUF↓ 1,334
mradermacher/gemma-2-27b-it-abliterated-i1-GGUF↓ 1,323
mradermacher/gemma-2-27b-it-abliterated-GGUF↓ 600
tensorblock/gemma-2-27b-it-abliterated-GGUF↓ 203
byroneverson/gemma-2-27b-it-abliterated-gguf↓ 136
TouchNight/gemma-2-27b-it-abliterated-Q3_K_M-GGUF↓ 50
aashish1904/gemma-2-27b-it-abliterated-Q4_K_M-GGUF↓ 44
EmperorKronos/gemma-2-27b-it-abliterated-exl2↓ 35
TouchNight/gemma-2-27b-it-abliterated-Q3_K_S-GGUF↓ 34
TouchNight/gemma-2-27b-it-abliterated-Q2_K-GGUF↓ 28
Iambackup/gemma-2-27b-it-abliterated-i1-GGUF↓ 25
ta-att/gemma-2-27b-it-abliterated-Q5_K_S-GGUF↓ 21
cgifbribcgfbi/gemma-2-27b-it-abliterated-chem-claude-5-comp3-sort-pat↓ 13
cgifbribcgfbi/gemma-2-27b-it-abliterated-chem-textbook↓ 12
cgifbribcgfbi/gemma-2-27b-it-abliterated-chem-claude-1-comp3-sort-pat-self↓ 9

Variants by this author 2 formats · 207 downloads combined

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

Metadata

License
gemma
Languages
en
Tags
transformers safetensors gemma2 text-generation gemma gemma-2 chat it abliterated conversational en base_model:google/gemma-2-27b-it

Related

Other by byroneverson
Yi-1.5-9B-Chat-abliterated 9K Mistral-Small-Instruct-2409-abliterated 9K Yi-1.5-9B-Chat-16K-abliterated 9K glm-4-9b-chat-abliterated-gguf 234 glm-4-9b-chat-abliterated 155
Total size
50.7 GB
Files
24
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2024-09-14 07:20
Source huggingface.co/byroneverson/gemma-2-27b-it-abliterated ↗ Base model google/gemma-2-27b-it ↗

Files by quantization

Auxiliary files 24 files 50.7 GB
model-00004-of-00012.safetensors 4.64 GB c440a371 download
model-00007-of-00012.safetensors 4.64 GB 80db96bf download
model-00010-of-00012.safetensors 4.64 GB 918094a8 download
model-00005-of-00012.safetensors 4.54 GB 9417f4c5 download
model-00008-of-00012.safetensors 4.54 GB bd4d6a8a download
model-00011-of-00012.safetensors 4.54 GB 6925b0e3 download
model-00006-of-00012.safetensors 4.54 GB 02f761e8 download
model-00009-of-00012.safetensors 4.54 GB 0cdc5cae download
model-00003-of-00012.safetensors 4.54 GB 243b03ca download
model-00002-of-00012.safetensors 4.54 GB 047e27ac download
model-00001-of-00012.safetensors 4.41 GB 265de456 download
model-00012-of-00012.safetensors 648 MB 7a3555da download
refusal_direction.pt 19.2 KB 0b1a5984 download
tokenizer.json 16.7 MB 3f289bc0 download
tokenizer.model 4.04 MB 61a7b147 download
logo.png 1.39 MB 5c42b5db download
abliterate-gemma-2-27b-it.ipynb 724 KB 0c6bbe29 download
tokenizer_config.json 45.9 KB 36ba1287 download
model.safetensors.index.json 41.8 KB 490d8bbe download
.gitattributes 1.64 KB e8305643 download
README.md 1.01 KB e18392d1 download
config.json 893 B 7e6f786a download
special_tokens_map.json 636 B 8d6368f7 download
generation_config.json 173 B bcab2397 download

README current version from Hugging Face

base_model: google/gemma-2-27b-it
pipeline_tag: text-generation
license: gemma
language:

  • en
    tags:
  • gemma
  • gemma-2
  • chat
  • it
  • abliterated
    library_name: transformers

gemma-2-27b-it-abliterated

Now accepting abliteration requests. If you would like to see a model abliterated, follow me and leave me a message with model link.

This is a new approach for abliterating models using CPU only. I was able to abliterate this model using free kaggle processing with no accelerator.

  1. Obtain refusal direction vector using a quant model with llama.cpp (llama-cpp-python and ggml-python).
  2. Orthogonalize each .safetensors files directly from original repo and upload to a new repo. (one at a time)

Check out the jupyter notebook for details of how this model was abliterated from gemma-2-27b-it.

Logo

README history 12 versions

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

  1. 2024-09-14Update README.md4c642971 KB
    Loading...
  2. 2024-09-09Update README.mdf419e8d897 B
    Loading...
  3. 2024-08-31Update README.mdfbd1e00902 B
    Loading...
  4. 2024-08-29Update README.mda007edf901 B
    Loading...
  5. 2024-08-29Update README.md4da0890815 B
    Loading...
  6. 2024-08-29Update README.mdb9365081.1 KB
    Loading...
  7. 2024-08-29Update README.md9dff2b01.1 KB
    Loading...
  8. 2024-08-28Update README.md2b7d24a1.1 KB
    Loading...
  9. 2024-08-28Update README.mdc4fced51 KB
    Loading...
  10. 2024-08-28Update README.md931674e1 KB
    Loading...
  11. 2024-08-28Update README.md37418f4530 B
    Loading...
  12. 2024-08-28initial commit632da3028 B
    Loading...

Discussions 3 threads

  1. 2025-03-20gemma 3open1 💬#3
    Loading...
  2. 2025-03-13Not really abliteratedopen1 💬#2
    Loading...
  3. 2024-10-26Abliteration messageopen3 💬#1
    Loading...

Continue on Hugging Face →

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