metadata

license: mit
base_model: nkpz/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT
tags:
  - llama-cpp
  - gguf-my-repo

Triangle104/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT-Q8_0-GGUF

This model was converted to GGUF format from nkpz/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.

This model is decensored using a technique I developed called DeLMAT: Decensoring Language Models through Activation Tuning. It's similar to the ablation / "abliteration" scripts that are out there, but works by training a LoRA adapter and calculating a loss based on the distance from the mean refusal activation and the distance between the mean acceptance activation.

The training script is released under the MIT license: https://github.com/nkpz/DeLMAT

Rather than simply attempting to cancel out the refusal direction, DeLMAT guides the model toward an acceptance. In other words, instead of simply forgetting how to refuse requests, it learns to emphatically accept requests.

Use with llama.cpp

Install llama.cpp through brew (works on Mac and Linux)

brew install llama.cpp

Invoke the llama.cpp server or the CLI.

CLI:

llama-cli --hf-repo Triangle104/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT-Q8_0-GGUF --hf-file deephermes-3-llama-3-8b-preview-uncensored-delmat-q8_0.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT-Q8_0-GGUF --hf-file deephermes-3-llama-3-8b-preview-uncensored-delmat-q8_0.gguf -c 2048

Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.

git clone https://github.com/ggerganov/llama.cpp

Step 2: Move into the llama.cpp folder and build it with LLAMA_CURL=1 flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).

cd llama.cpp && LLAMA_CURL=1 make

Step 3: Run inference through the main binary.

./llama-cli --hf-repo Triangle104/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT-Q8_0-GGUF --hf-file deephermes-3-llama-3-8b-preview-uncensored-delmat-q8_0.gguf -p "The meaning to life and the universe is"

./llama-server --hf-repo Triangle104/DeepHermes-3-Llama-3-8B-Preview-Uncensored-DeLMAT-Q8_0-GGUF --hf-file deephermes-3-llama-3-8b-preview-uncensored-delmat-q8_0.gguf -c 2048