← back to vectors · latest commit 2024-12-19

72B-Qwen2.5-Kunou-v1-control-vectors

Name: 72B-Qwen2.5-Kunou-v1-control-vectors
Author: gghfez

gghfez ·V1 ·24 files ·apache-2.0

Author's description Source: Hugging Face · gghfez/72B-Qwen2.5-Kunou-v1-control-vectors ↗

gghfez/72B-Qwen2.5-Kunou-v1-control-vectors

Creative Writing control-vectors for Sao10K/72B-Qwen2.5-Kunou-v1

Usage example

llama.cpp

llama-server --model  72B-Qwen2.5-Kunou-v1-Q4_K.gguf [other CLI arguments] \
    --control-vector  72B-Qwen2.5-Kunou-v1-character_focus__debias.gguf \
    --control-vector  72B-Qwen2.5-Kunou-v1-character_focus__dialogue.gguf \
    --control-vector  <other vectors>

Trained on 1xH100NVL using jukofyork's control-vectors

Based off Version v3.0 of creative-writing-control-vectors-v3.0.

See here for more info about control-vectors creative-writing-control-vectors-v3.0

README history (1 version)

2024-12-19Create README.md0512d22966 B
Loading...

Abliteration catalog analysis

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

Method: V1: PCA on contrastive pairsZou et al. 2023, Representation Engineering; Vogel 2024, repeng
Confidence: 80%
Types: meta 8behavior 8style 6mood 2
File count: 24 vectors, grouped into 8 sets below

How we classify →

Vectors in this repo

72B-Qwen2.5-Kunou-v1: Compassion ↔ Sadism

behavior V1

↓ Debias 2.47 MB · sha256 82c65d5beda6
↓ Compassion pole 2.47 MB · sha256 36d3c04671e2
↓ Sadism pole 2.47 MB · sha256 d91d4b1f48a6

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-compassion_vs_sadism__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-compassion_vs_sadism__sadism.gguf:1.0

72B-Qwen2.5-Kunou-v1: Descriptive ↔ Explicit storytelling

style V1

↓ Debias 2.47 MB · sha256 7a08b8fd6193
↓ Descriptive pole 2.47 MB · sha256 6b5670da6211
↓ Explicit pole 2.47 MB · sha256 63d805215343

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-storytelling__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-storytelling__explicit.gguf:1.0

72B-Qwen2.5-Kunou-v1: Empathy ↔ Sociopathy

behavior V1

↓ Debias 2.47 MB · sha256 0bf24fb2e075
↓ Empathy pole 2.47 MB · sha256 c0575ba09038
↓ Sociopathy pole 2.47 MB · sha256 8863c99bc715

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-empathy_vs_sociopathy__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-empathy_vs_sociopathy__sociopathy.gguf:1.0

72B-Qwen2.5-Kunou-v1: Honesty ↔ Machiavellianism

behavior V1

↓ Debias 2.47 MB · sha256 d82a4001be87
↓ Honesty pole 2.47 MB · sha256 4a91b933157e
↓ Machiavellianism pole 2.47 MB · sha256 1bc4772ef171

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-honesty_vs_machiavellianism__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-honesty_vs_machiavellianism__machiavellianism.gguf:1.0

72B-Qwen2.5-Kunou-v1: Humility ↔ Narcissism

behavior V1

↓ Debias 2.44 MB · sha256 42648a5d9a2b
↓ Humility pole 2.44 MB · sha256 48f4e5f67f9f
↓ Narcissism pole 2.44 MB · sha256 43e5f3e77cf3

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-humility_vs_narcissism__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-humility_vs_narcissism__narcissism.gguf:1.0

72B-Qwen2.5-Kunou-v1: Narration ↔ Dialogue

style V1

↓ Debias 2.47 MB · sha256 35ba74f0ac0f
↓ Narration pole 2.47 MB · sha256 07350640af02
↓ Dialogue pole 2.47 MB · sha256 3d29bae9641e

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-character_focus__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-character_focus__dialogue.gguf:1.0

72B-Qwen2.5-Kunou-v1: Optimism ↔ Nihilism

mood V1

↓ Debias 2.47 MB · sha256 8752be000540
↓ Optimism pole 2.47 MB · sha256 c4eaee576a86
↓ Nihilism pole 2.47 MB · sha256 7ee961f19f5d

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-optimism_vs_nihilism__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-optimism_vs_nihilism__nihilism.gguf:1.0

72B-Qwen2.5-Kunou-v1: Simple ↔ Ornate language

style V1

↓ Debias 2.41 MB · sha256 b39172f3ef74
↓ Simple pole 2.41 MB · sha256 8e59fa443803
↓ Ornate pole 2.41 MB · sha256 e0b7624f6f20

llama-cli -m <model>.gguf \
  --control-vector 72B-Qwen2.5-Kunou-v1-language__debias.gguf \
  --control-vector-scaled 72B-Qwen2.5-Kunou-v1-language__ornate.gguf:1.0

Commit history

2024-12-19Create README.md0512d22
2024-12-17Upload folder using huggingface_hubcd1f9b6
2024-12-17initial commit0c2739e

gghfez/72B-Qwen2.5-Kunou-v1-control-vectors

Usage example

llama.cpp

Vectors in this repo

Commit history

Related repos

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