Triangle104/MN-Violet-Lotus-12B-Q4_K_M-GGUF

This model was converted to GGUF format from FallenMerick/MN-Violet-Lotus-12B using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.

---

Model details:

This is the model I was trying to create when Chunky-Lotus emerged. Not only does this model score higher on my local EQ benchmarks (80.00 w/ 100% parsed @ 8-bit), but it does an even better job at roleplaying and producing creative outputs while still adhering to wide ranges of character personalities. The high levels of emotional intelligence are really quite noticeable as well.

Once again, models tend to score higher on my local tests when compared to their posted scores, but this has become the new high score for models I've personally tested.

I really like the way this model writes, and I hope you'll enjoy using it as well!

Merge Details

This is a merge of pre-trained language models created using mergekit.

Merge Method

This model was merged using the Model Stock merge method. Models Merged

The following models were included in the merge:

Epiculous/Violet_Twilight-v0.2
NeverSleep/Lumimaid-v0.2-12B
flammenai/Mahou-1.5-mistral-nemo-12B
Sao10K/MN-12B-Lyra-v4

Configuration

The following YAML configuration was used to produce this model:

models:

• model: FallenMerick/MN-Twilight-Maid-SLERP-12B #(unreleased)
• model: Sao10K/MN-12B-Lyra-v4
• model: flammenai/Mahou-1.5-mistral-nemo-12B merge_method: model_stock base_model: mistralai/Mistral-Nemo-Instruct-2407 parameters: normalize: true dtype: bfloat16

In this recipe, Violet Twilight and Lumimaid were first blended using the SLERP method to create a strong roleplaying foundation. Lyra v4 is then added to the mix for its great creativity and roleplaying performance, along with Mahou to once again curtail the outputs and prevent the resulting model from becoming too wordy. Model Stock was used for the final merge in order to really push the resulting weights in the proper direction while using Nemo Instruct as a strong anchor point.

---

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/MN-Violet-Lotus-12B-Q4_K_M-GGUF --hf-file mn-violet-lotus-12b-q4_k_m.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/MN-Violet-Lotus-12B-Q4_K_M-GGUF --hf-file mn-violet-lotus-12b-q4_k_m.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/MN-Violet-Lotus-12B-Q4_K_M-GGUF --hf-file mn-violet-lotus-12b-q4_k_m.gguf -p "The meaning to life and the universe is"

or

./llama-server --hf-repo Triangle104/MN-Violet-Lotus-12B-Q4_K_M-GGUF --hf-file mn-violet-lotus-12b-q4_k_m.gguf -c 2048