IceAbsintheNeatRP-7b

IceAbsintheNeatRP-7b (Ice0.150-20.10-RP-ABLv32)

mistral v0.2 base

...wip...

Abliteration Details

Abliteration Method

IceAbsintheNeatRP-7b is an abliterated version of IceAbsinthe-7b (Ice0.150-20.10-RP), created using grimjim's llm-abliteration tool, which can be found at https://github.com/jim-plus/llm-abliteration.

The process involves the following steps:

1. python measure.py -m H:\FModels\Ice0.150-20.10-RP -o H:\RModels\Ice0.150-20.10-RP.pt --data-harmful F:\toxic_refusals_verified.jsonl --clip 0.999 -q 4bit --batch-size 8
2. python sharded_ablate.py H:\RModels\Ice0.150-20.10-RP-ABL-v32.yml

The YAML file needs to be configured for each specific model based on the graphs generated by analyze.py. For further details, please refer to the GitHub page tool.

Ice0.150-20.10-RP-ABL-v32.yml

model: "H:/FModels/Ice0.150-20.10-RP"
measurements: "H:/RModels/Ice0.150-20.10-RP.pt"
output: "G:/FModels/Ice0.150-20.10-RP-ABLv32"
ablate:  
  - layer: 14
    measurement: 14
    scale: 3.0
    sparsity: 0
  - layer: 15
    measurement: 15
    scale: 4.0
    sparsity: 0
  - layer: 16
    measurement: 16
    scale: 3.0
    sparsity: 0
  - layer: 17
    measurement: 17
    scale: 1.0
    sparsity: 0
  - layer: 18
    measurement: 18
    scale: 4.0
    sparsity: 0
  - layer: 19
    measurement: 19
    scale: 4.0
    sparsity: 0
  - layer: 20
    measurement: 20
    scale: 1.0
    sparsity: 0
  - layer: 21
    measurement: 21
    scale: 2.0
    sparsity: 0
  - layer: 22
    measurement: 22
    scale: 2.0
    sparsity: 0
  - layer: 23
    measurement: 23
    scale: 2.0
    sparsity: 0
  - layer: 24
    measurement: 24
    scale: 2.0
    sparsity: 0
  - layer: 25
    measurement: 25
    scale: 2.0
    sparsity: 0
  - layer: 26
    measurement: 26
    scale: 2.0
    sparsity: 0
  - layer: 27
    measurement: 27
    scale: 1.0
    sparsity: 0
  - layer: 28
    measurement: 28
    scale: 1.0
    sparsity: 0
  - layer: 29
    measurement: 29
    scale: 1.0
    sparsity: 0