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

AlexWortega/qwen35-4b-soyuz-abliterated-v6_hardpairs

AlexWortega Qwen 4.5B multimodal
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/AlexWortega%2Fqwen35-4b-soyuz-abliterated-v6_hardpairs"
Response includes
  • classification m1
  • files 9
  • benchmarks 11 entries
  • hub_downloads_all_time 40
  • author_summary 6 models
  • readme_text full
Try next
GET/models?author=AlexWortega
All 6 models by AlexWortega
 GET/models?family=qwen&method=m1
Sibling qwen 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
40
11 last 30d - stable
Likes
0
Model age
4mo ago
created 2026-05-26

Training datasets

Corpora the author lists in the model card. Datasets tracked in our /datasets catalog carry a category badge linking to the workflow stage. Others open on Hugging Face.

AlexWortega/Soyuz-sft
Downloads over time
Now44→from20↑120%
1928374620 on Jun 1044 on Oct 1144 on Oct 6JunJulAugSepOct
Jun 10 → Oct 11 · 57 snapshots · spans 123 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
Entertainment 0.9 UGI
Hazardous 1.2 UGI
Natural Intelligence 13.45 UGI
Political lean -17.3% UGI
Sensitive-Info 11.73 UGI
SocPol 1.5 UGI
UGI 15.32 UGI
Willingness (10) 2.2 UGI
W10-Adherence 1.5 UGI
W10-Direct 3 UGI
Writing 29.68 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.

Qwen/Qwen3.5-4B this lineage
AlexWortega/qwen35-4b-soyuz-abliterated-v6_hardpairs↓ 11

Metadata

License
apache-2.0
Languages
en
Tags
transformers safetensors qwen3_5 image-text-to-text agent abliteration orthogonalization weight-ortho soyuz qwen3.5 phase2 conversational

Related

Other by AlexWortega
qwen3.5-4b-abliterated-agent-20260515 87 qwen35-4b-soyuz-abliterated-v3-multi 16 qwen35-4b-soyuz-abliterated-v8_cfact 15 qwen35-4b-soyuz-abliterated-v2 11 qwen35-4b-soyuz-abliterated-v7_agentonly 9
Same base model
huihui-ai/Huihui-Qwen3.5-4B-abliterated 8K insraq/Qwen3.5-4B-EmperoAI-Qwen3.8-Distill-Heretic-Abliterated-MTP-GGUF 6K Andycurrent/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive 3K rodrigomt/Qwen3.5-4B-Uncensored-Aggressive 2K DreamFast/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark 1K
Total size
8.46 GB
Files
9
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-05-26 06:06
Source huggingface.co/AlexWortega/qwen35-4b-soyuz-abliterated-v6_hardpairs ↗ Base model Qwen/Qwen3.5-4B ↗

Files by quantization

Auxiliary files 9 files 8.47 GB
model.safetensors 8.46 GB daeba7a3 download
tokenizer.json 19.1 MB 06b95093 download
chat_template.jinja 7.57 KB a585dec8 download
config.json 2.76 KB 4829f7b9 download
.gitattributes 1.53 KB 52373fe2 download
README.md 1.39 KB ba68f7cb download
processor_config.json 1.16 KB 33818c7f download
tokenizer_config.json 1.14 KB b4acebe0 download
generation_config.json 115 B d9dfd84b download

README current version from Hugging Face

library_name: transformers
base_model: Qwen/Qwen3.5-4B
license: apache-2.0
language: [en]
tags:

  • agent
  • abliteration
  • orthogonalization
  • weight-ortho
  • soyuz
  • qwen3.5
  • phase2
    datasets:
  • AlexWortega/Soyuz-sft

Qwen3.5-4B Soyuz — Abliterated (v6_hardpairs)

Phase-2 weight-orthogonalized variant of
AlexWortega/qwen35-4b-soyuz-merged.

field value
Method phase2 exp2 within-task hard pairs (51 paired contrasts), mean diff L=9, strength=0.5
tbench-2 (17) 2/17
HermesAgent-20 8 / 20
MMLU-Pro 2.08%
EQbench3 —
Notes Same-task contrast removes difficulty noise but MMLU still collapses

Lineage

Continues the capability-vectors
sweep. Phase 1 best was v2 (HA20 8/20, MMLU collapse 58→2). Phase 2 explores
multi-token / hard-pairs / counterfactual / agent-only / activation-steering recipes.

See https://github.com/AlexWortega/capability-vectors for repo + per-experiment
README, and phase2/results/all_variants.csv for the live results table.

Usage with sglang

python -m sglang.launch_server \
    --model-path AlexWortega/qwen35-4b-soyuz-abliterated-v6_hardpairs \
    --dtype bfloat16 --trust-remote-code \
    --tool-call-parser hermes --chat-template hermes_qwen.jinja

README history 1 version

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

  1. 2026-05-26phase2 abliterated v6_hardpairsde6cdad1.4 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