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

cs2764/MiniMax-M2.5_dq4-abliterated-mlx

cs2764 Minimax 229B
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/cs2764%2FMiniMax-M2.5_dq4-abliterated-mlx"
Response includes
  • classification m1
  • files 38
  • hub_downloads_all_time 736
  • author_summary 28 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 · lifetime
736
14 last 30d - cooling
Likes
0
Model age
7mo ago
created 2026-03-02
Downloads over time
Now741→from188↑294%
160372584796188 on Mar 4741 on Oct 11741 on Oct 8MarAprMayJunJulAugSepOct
Mar 4 → Oct 11 · 71 snapshots · spans 221 days

Metadata

License
other
Tags
mlx safetensors minimax_m2 abliteration uncensored experimental custom_code license:other 4-bit region:us

Related

Total size
125 GB
Files
38
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-03-02 20:48

Files by quantization

Auxiliary files 38 files 125 GB
model-00001-of-00027.safetensors 4.97 GB 8fe9d455 download
model-00003-of-00027.safetensors 4.95 GB ba430930 download
model-00005-of-00027.safetensors 4.71 GB c32cd766 download
model-00014-of-00027.safetensors 4.71 GB 1b3a03f4 download
model-00023-of-00027.safetensors 4.71 GB 52f4e42d download
model-00008-of-00027.safetensors 4.71 GB 9a30e523 download
model-00020-of-00027.safetensors 4.71 GB 5a8c04d2 download
model-00010-of-00027.safetensors 4.66 GB d8a96dad download
model-00012-of-00027.safetensors 4.66 GB 680e53e3 download
model-00018-of-00027.safetensors 4.66 GB e58a85de download
model-00025-of-00027.safetensors 4.66 GB ca49cd04 download
model-00016-of-00027.safetensors 4.66 GB 496b8b4b download
model-00027-of-00027.safetensors 4.59 GB 5ac8d524 download
model-00026-of-00027.safetensors 4.57 GB f4160733 download
model-00017-of-00027.safetensors 4.57 GB 9db91fd1 download
model-00011-of-00027.safetensors 4.57 GB 11bb0414 download
model-00022-of-00027.safetensors 4.52 GB 7e3d60a6 download
model-00015-of-00027.safetensors 4.52 GB fa452923 download
model-00019-of-00027.safetensors 4.52 GB 6abd7171 download
model-00024-of-00027.safetensors 4.52 GB 669053f7 download
model-00013-of-00027.safetensors 4.52 GB 8eaf4d28 download
model-00021-of-00027.safetensors 4.52 GB c1f479c3 download
model-00006-of-00027.safetensors 4.52 GB e2d9c45b download
model-00007-of-00027.safetensors 4.52 GB 0d4a5120 download
model-00009-of-00027.safetensors 4.52 GB 0d6a1c40 download
model-00004-of-00027.safetensors 4.52 GB fec71c74 download
model-00002-of-00027.safetensors 4.45 GB 44d3585f download
tokenizer.json 14.8 MB 7b81e5e5 download
model.safetensors.index.json 167 KB 4be5e855 download
config.json 145 KB 6c6ba639 download
modeling_minimax_m2.py 30.2 KB 8846d38a download
configuration_minimax_m2.py 9.92 KB 7fcd9861 download
chat_template.jinja 6.37 KB 4623080a download
README.md 2.02 KB 0c588ecb download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.35 KB 07e06b8d download
abliteration_log.json 1.14 KB 9c3158ab download
generation_config.json 166 B 30b418a4 download

README current version from Hugging Face


tags:

  • mlx
  • abliteration
  • uncensored
  • experimental
    license: other

MiniMax-M2.5_dq4-abliterated

This model was created using the mlx-abliteration toolkit,
which is based on the FiditeNemini/mlx-abliteration project.

Base Model

Original model: ./models/MiniMax-M2.5_dq4 (local path)

What is Abliteration?

Abliteration is a mechanistic interpretability technique that identifies and orthogonalizes the
"refusal direction" in a model's activation space, surgically removing refusal behavior without
full fine-tuning.

Abliteration Parameters

Parameter Value
Base model ./models/MiniMax-M2.5_dq4
Ablation method projection
Refusal vector policy per-layer
Refusal direction method projected
Ablation strength 2.0
Probed layers all
PCA components (ablate-k) 1
Attention only True
Timestamp 2026-03-02T20:40:28 UTC

This model uses a Mixture-of-Experts (MoE) architecture. Abliteration targets only the attention projection weights (q/k/v/o_proj) to preserve expert routing quality.

⚠️ Disclaimer

This model is intended for research, experimentation, and testing purposes only.

  • This model may produce harmful, offensive, inappropriate, or otherwise objectionable content.
  • The abliteration process removes safety guardrails that were intentionally built into the original model.
  • Do not use this model in production systems, consumer-facing applications, or any context
    where harmful outputs could cause real-world harm.
  • The authors and contributors of this toolkit bear no responsibility for any misuse of this model
    or any harm caused by outputs generated by this model.
  • By using this model, you agree that you are solely responsible for ensuring its use complies
    with all applicable laws and ethical guidelines.

This model is shared purely for academic and technical exploration of model internals.

README history 1 version

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

  1. 2026-03-02Add files using upload-large-folder tool219733c2 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.

Open in Abliteration