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

cs2764/MiniMax-M2.5-8bit-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-M2.5-8bit-abliterated-mlx"
Response includes
  • classification m1
  • files 58
  • hub_downloads_all_time 1,064
  • 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
1K
24 last 30d - cooling
Likes
0
Model age
7mo ago
created 2026-03-03
Downloads over time
Now1.1K→from399↑168%
3656238801.1K399 on Mar 41.1K on Oct 111.1K 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 8-bit region:us

Related

Total size
226 GB
Files
58
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-03-03 21:36

Files by quantization

Auxiliary files 58 files 226 GB
model-00016-of-00047.safetensors 4.88 GB eee83dc1 download
model-00028-of-00047.safetensors 4.88 GB 2c0fe195 download
model-00019-of-00047.safetensors 4.88 GB a80f2ce3 download
model-00037-of-00047.safetensors 4.88 GB 010c7eb7 download
model-00025-of-00047.safetensors 4.88 GB 4ad4796b download
model-00031-of-00047.safetensors 4.88 GB 8bdecf27 download
model-00043-of-00047.safetensors 4.88 GB 3140b7f3 download
model-00040-of-00047.safetensors 4.88 GB 603c886b download
model-00022-of-00047.safetensors 4.88 GB a835f238 download
model-00013-of-00047.safetensors 4.88 GB 0e472937 download
model-00034-of-00047.safetensors 4.88 GB bd8f5c34 download
model-00046-of-00047.safetensors 4.88 GB 8ae6f346 download
model-00007-of-00047.safetensors 4.88 GB 439261e6 download
model-00004-of-00047.safetensors 4.88 GB 2aca4be6 download
model-00010-of-00047.safetensors 4.88 GB 61b04a04 download
model-00008-of-00047.safetensors 4.83 GB 53a7472c download
model-00011-of-00047.safetensors 4.83 GB 38b4703c download
model-00042-of-00047.safetensors 4.83 GB e9afee8f download
model-00044-of-00047.safetensors 4.83 GB 0af7291a download
model-00039-of-00047.safetensors 4.83 GB c7f45a1c download
model-00020-of-00047.safetensors 4.83 GB 5f891d85 download
model-00033-of-00047.safetensors 4.83 GB 74e5b3a5 download
model-00026-of-00047.safetensors 4.83 GB e320adbe download
model-00014-of-00047.safetensors 4.83 GB a6df31f8 download
model-00030-of-00047.safetensors 4.83 GB 67d1febb download
model-00021-of-00047.safetensors 4.83 GB 1cc2fcd6 download
model-00029-of-00047.safetensors 4.83 GB 88d5a8db download
model-00035-of-00047.safetensors 4.83 GB c905505f download
model-00045-of-00047.safetensors 4.83 GB 234fc523 download
model-00015-of-00047.safetensors 4.83 GB 2b4dc6cc download
model-00017-of-00047.safetensors 4.83 GB 067192d7 download
model-00036-of-00047.safetensors 4.83 GB 886f9b21 download
model-00041-of-00047.safetensors 4.83 GB bde34ca4 download
model-00009-of-00047.safetensors 4.83 GB aa3b5455 download
model-00012-of-00047.safetensors 4.83 GB cf7665fb download
model-00023-of-00047.safetensors 4.83 GB 2b11d36c download
model-00018-of-00047.safetensors 4.83 GB 858d0c9b download
model-00027-of-00047.safetensors 4.83 GB a75a046d download
model-00032-of-00047.safetensors 4.83 GB 60b2584e download
model-00005-of-00047.safetensors 4.83 GB 58b86b62 download
model-00024-of-00047.safetensors 4.83 GB 3eba0ee3 download
model-00002-of-00047.safetensors 4.83 GB 53fd290e download
model-00038-of-00047.safetensors 4.83 GB fe912924 download
model-00003-of-00047.safetensors 4.83 GB 5f773b57 download
model-00006-of-00047.safetensors 4.83 GB 200aef0c download
model-00001-of-00047.safetensors 4.29 GB 8cf43365 download
model-00047-of-00047.safetensors 4.19 GB df37d963 download
tokenizer.json 14.8 MB 7b81e5e5 download
model.safetensors.index.json 167 KB 4480843a download
modeling_minimax_m2.py 30.2 KB 8846d38a download
config.json 15.7 KB c1ead22e download
configuration_minimax_m2.py 9.92 KB 7fcd9861 download
chat_template.jinja 6.37 KB 4623080a download
README.md 2.05 KB 7105878d download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.35 KB 07e06b8d download
abliteration_log.json 1.20 KB 6d07a720 download
generation_config.json 166 B 30b418a4 download

README current version from Hugging Face


tags:

  • mlx
  • abliteration
  • uncensored
  • experimental
    license: other

MiniMax-M2.5-8bit-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-M2.5-8bit (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-M2.5-8bit
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
Excluded modules none
Timestamp 2026-03-03T20:56:22 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. 2026-03-03Add files using upload-large-folder tool464c1da2 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