← back to catalog · registered 2026-10-01 01:58

interimlabs/InterimLabs-Huihui-Qwen3.5-4B-abliterated-MLX-4bit

interimlabs 4B second-order
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/interimlabs%2FInterimLabs-Huihui-Qwen3.5-4B-abliterated-MLX-4bit"
Response includes
  • classification m1
  • files 10
  • author_summary 6 models
  • readme_text full
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)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · 30-day
730
Likes
1
Model age
7w ago
created 2026-10-01
Downloads over time
Now 1.3K → from 0 ↑ 0%
04599171.4K 0 on Oct 1 1.3K on Nov 22 OctNov
Oct 1 → Nov 22 · 53 snapshots · spans 52 days

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
apache-2.0
Tags
mlx safetensors qwen3_5_text 4-bit qwen3.5 interimlabs abliterated text-generation conversational base_model:huihui-ai/Huihui-Qwen3.5-4B-abliterated base_model:quantized:huihui-ai/Huihui-Qwen3.5-4B-abliterated license:apache-2.0

Related

Total size
2.20 GB
Files
10
Quantizations
1
Registered
2026-10-01 01:58
Last updated on HF
2026-10-08 16:59

Files by quantization

Auxiliary files 10 files 2.22 GB
model.safetensors 2.20 GB 28595def download
tokenizer.json 19.1 MB 87a7830d download
LICENSE 11.3 KB f938136e download
chat_template.jinja 7.57 KB a585dec8 download
config.json 2.06 KB 876a7ea2 download
manifest.json 1.95 KB 354d08ef download
README.md 1.55 KB 0746c225 download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.18 KB eb15ee01 download
minstrel-provenance.json 496 B b6eb27a7 download

README current version from Hugging Face


license: apache-2.0
base_model: huihui-ai/Huihui-Qwen3.5-4B-abliterated
pipeline_tag: text-generation
tags:

  • mlx
  • abliterated
  • qwen3.5

Interim Labs Huihui Qwen3.5-4B abliterated — MLX 4-bit

Text-only MLX package for Minstrel's opt-in Advanced · Unbound local model. The underlying checkpoint is Huihui Qwen3.5-4B abliterated at revision 5581467dfd52bf338c782006a6cdce05c42594be. The MLX source is TheWirelessPhoenix/Qwen3.5-4B-abliterated_4bit at revision 916d9f7c2e27ed41e2b9ed39f3a2badccb2bbc7c.

The package retains 924 language-model tensors and has no vision tensors. It uses 4-bit affine quantization with group size 64, a qwen3_5_text configuration, and an 8,192-token application context limit. minstrel-provenance.json records the pinned inputs and text-only derivation. The signed manifest.json binds this package to Minstrel's detailed-abliterated-local slot. The Apache-2.0 license file was copied from Qwen/Qwen3.5-4B at revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a. The MLX source identifies Huihui as upstream but does not establish the exact Huihui revision used for the conversion.

This is a first-pass opt-in model intended to reduce inappropriate refusals in fictional writing. It has not undergone Minstrel's writing-quality qualification. No reliability or refusal-rate improvement is claimed. Existing local-model choices remain available.

README history 4 versions

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

  1. 2026-10-08 Rebrand standalone Interim Labs release and link model series dba9324 4.8 KB
    Loading...
  2. 2026-10-01 Clarify 4B MLX license and conversion provenance cbbd3cb 1.5 KB
    Loading...
  3. 2026-10-01 Publish signed 4B MLX derivative f52987e 1.3 KB
    Loading...
  4. 2026-10-01 initial commit 0658a35 28 B
    Loading...

Discussions 1 thread

  1. 2026-10-22 Does this even work??? open 2 💬 #1
    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.

Open in Abliteration