← back to vectors · latest commit 2026-09-05

MiniMax-M3-control-vectors

Name: MiniMax-M3-control-vectors
Author: gghfez

gghfez ·V1 ·21 files ·cc-by-nc-4.0

Author's description Source: Hugging Face · gghfez/MiniMax-M3-control-vectors ↗

gghfez/MiniMax-M3-control-vectors

TODO

Creative Writing control-vectors for MiniMaxAI/MiniMax-M3

What are Control Vectors?

Control vectors steer model behavior without retraining by adjusting
activations in specific directions.

These creative writing vectors help
you tune narrative tone, character focus, and stylistic elements.

Available Vectors

optimism_vs_nihilism
empathy_vs_sociopathy
honesty_vs_machiavellianism
humility_vs_narcissism
character focus (dialogue_vs_narration)
language (simple vs ornate)
storytelling (descriptive vs explicit)

Usage

Apply the debias vector and either the positive or negative vector when starting llama-server.
If both axes are applied at the same time, they will cancel each other out.
The debias vector must be set to 1.0

You can use either --control-vector [/path/to/vector.gguf] or --control-vector-scaled [/path/to/vector.gguf] [scale factor]

When using ik_llama.cpp, you can also use the REST-API (examples below)

IMPORTANT: As these vectors are calibrated against the debiased model, the debias vector must be applied along with the corresponding positive or negative axis control vector - they cannot be used on their own!

Examples (ik_llama.cpp or llama.cpp)

At server start-up

llama-server --model MiniMax-M3-Q8_0.gguf [your usual CLI arguments] \
    --control-vector-scaled MiniMax-M3-optimism_vs_nihilism__debias.gguf 1.0 \
    --control-vector-scaled MiniMax-M3-optimism_vs_nihilism__nihilism.gguf 1.0 \

At server start-up with thinking disabled

llama-server --model MiniMax-M3-Q8_0.gguf [your usual CLI arguments] \
    --chat-template-kwargs '{"enable_thinking": false}' \
    --control-vector-scaled MiniMax-M3-optimism_vs_nihilism__debias.gguf 1.0 \
    --control-vector-scaled MiniMax-M3-optimism_vs_nihilism__nihilism.gguf 1.0 \

Load via API (ik_llama.cpp only)

load direction vector

curl 'http://127.0.0.1:8080/control-vectors/load' \
  -X POST \
  --data-raw '{"path":"/path/to/MiniMax-M3-control-vectors/MiniMax-M3-optimism_vs_nihilism__nihilism.gguf","scale":1}'

load debias vector

 curl 'http://127.0.0.1:8080/control-vectors/load' \
  -X POST \
  --data-raw '{"path":"/path/to/MiniMax-M3-control-vectors/MiniMax-M3-optimism_vs_nihilism__debias.gguf","scale":1}'

check loaded control-vectors

curl http://127.0.0.1/control-vectors

Samples

TODO

Acknowledgements

Based on the excellent work and cross-covariance eigendecomposition implementation by @jukofyork. His creative writing control vectors v3.0 repository contains the foundational code and methodology that made this possible, plus 70+ other creative writing control vectors across many models and useful steering directions.
Highly recommended for anyone working on improving LLM prose quality!

Abliteration catalog analysis

Vector classifier · v1 (classified 2026-10-30)

Method: V1: PCA on contrastive pairsZou et al. 2023, Representation Engineering; Vogel 2024, repeng
Also resembles: V8
Confidence: 70%
Types: meta 7style 6behavior 6mood 2
File count: 21 vectors, grouped into 7 sets below

How we classify →

Vectors in this repo

MiniMax-M3: Descriptive ↔ Explicit storytelling

style V1 works in app

↓ Debias 1.34 MB · sha256 1adbedbd14b5
↓ Descriptive pole 1.34 MB · sha256 bc48f51e7406
↓ Explicit pole 1.34 MB · sha256 e62ad7faa0e4

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-storytelling__debias.gguf \
  --control-vector-scaled MiniMax-M3-storytelling__explicit.gguf:1.0

MiniMax-M3: Empathy ↔ Sociopathy

behavior V1 works in app

↓ Debias 1.34 MB · sha256 9c7498219562
↓ Empathy pole 1.34 MB · sha256 7bb7731d9b68
↓ Sociopathy pole 1.34 MB · sha256 7167ca8be9ed

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-empathy_vs_sociopathy__debias.gguf \
  --control-vector-scaled MiniMax-M3-empathy_vs_sociopathy__sociopathy.gguf:1.0

MiniMax-M3: Honesty ↔ Machiavellianism

behavior V1 works in app

↓ Debias 1.34 MB · sha256 b59d6fa1f140
↓ Honesty pole 1.34 MB · sha256 6d2410b43612
↓ Machiavellianism pole 1.34 MB · sha256 7bce6281f439

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-honesty_vs_machiavellianism__debias.gguf \
  --control-vector-scaled MiniMax-M3-honesty_vs_machiavellianism__machiavellianism.gguf:1.0

MiniMax-M3: Humility ↔ Narcissism

behavior V1 works in app

↓ Debias 1.34 MB · sha256 611a825e9c73
↓ Humility pole 1.34 MB · sha256 16a5faf46c13
↓ Narcissism pole 1.34 MB · sha256 4706b2edf158

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-humility_vs_narcissism__debias.gguf \
  --control-vector-scaled MiniMax-M3-humility_vs_narcissism__narcissism.gguf:1.0

MiniMax-M3: Narration ↔ Dialogue

style V1 works in app

↓ Debias 1.34 MB · sha256 2a0a4f8c9244
↓ Narration pole 1.34 MB · sha256 7ab185990fa6
↓ Dialogue pole 1.34 MB · sha256 c755dd9b8124

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-character_focus__debias.gguf \
  --control-vector-scaled MiniMax-M3-character_focus__dialogue.gguf:1.0

MiniMax-M3: Optimism ↔ Nihilism

mood V1 works in app

↓ Debias 1.34 MB · sha256 e77822bd33cf
↓ Optimism pole 1.34 MB · sha256 68116e6f03d0
↓ Nihilism pole 1.34 MB · sha256 2d2d3ae8531f

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-optimism_vs_nihilism__debias.gguf \
  --control-vector-scaled MiniMax-M3-optimism_vs_nihilism__nihilism.gguf:1.0

MiniMax-M3: Simple ↔ Ornate language

style V1 works in app

↓ Debias 1.34 MB · sha256 2dfe5b721a93
↓ Simple pole 1.34 MB · sha256 79f7563d3f7a
↓ Ornate pole 1.34 MB · sha256 f8743bac341a

llama-cli -m <model>.gguf \
  --control-vector MiniMax-M3-language__debias.gguf \
  --control-vector-scaled MiniMax-M3-language__ornate.gguf:1.0

Commit history

2026-09-05Upload folder using huggingface_hubc1a4a0c
2026-09-05initial commit143ff9f

gghfez/MiniMax-M3-control-vectors

What are Control Vectors?

Available Vectors

Usage

Examples (ik_llama.cpp or llama.cpp)

Load via API (ik_llama.cpp only)

Samples

Acknowledgements

Vectors in this repo

MiniMax-M3: Descriptive ↔ Explicit storytelling

MiniMax-M3: Empathy ↔ Sociopathy

MiniMax-M3: Honesty ↔ Machiavellianism

MiniMax-M3: Humility ↔ Narcissism

MiniMax-M3: Narration ↔ Dialogue

MiniMax-M3: Optimism ↔ Nihilism

MiniMax-M3: Simple ↔ Ornate language

Commit history

Related repos

Run models on your own machine, not in the cloud.