royallab/MN-LooseCannon-12B-v2

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 20, 2024Architecture:Transformer0.0K Featherless Exclusive Warm

royallab/MN-LooseCannon-12B-v2 is a 12 billion parameter language model based on the Mistral Nemo architecture. Developed by royallab, this model is a multi-step merge of nothingiisreal/MN-12B-Celeste-V1.9, anthracite-org/magnum-v2.5-12b-kto, and Sao10K/MN-12B-Lyra-v1. It is specifically designed for conversational applications, recommending ChatML instruct format for optimal performance. The model's primary characteristic is its merged nature, inheriting biases from niche roleplaying forums.

Loading preview...

MN-LooseCannon-12B-v2 Overview

MN-LooseCannon-12B-v2 is a 12 billion parameter language model built upon the Mistral Nemo architecture. It is the result of a two-step merging process, combining three distinct models: nothingiisreal/MN-12B-Celeste-V1.9, anthracite-org/magnum-v2.5-12b-kto, and Sao10K/MN-12B-Lyra-v1. This merge was conducted with the explicit permission of the v1 creator, NGalrion.

Key Characteristics

  • Architecture: Based on the Mistral Nemo framework.
  • Merging Strategy: Utilizes a ties merge method, performed in two stages to integrate the constituent models.
  • Recommended Usage: Optimized for the ChatML instruct format, though it can also follow the base Mistral Nemo instruct format.

Limitations and Considerations

  • Bias: The model exhibits biases similar to those found in niche online roleplaying communities, in addition to biases inherited from its base models.
  • Factual Accuracy: It is not intended for providing factual information or advice.

Users should refer to the individual repositories of the merged models for detailed training information, as MN-LooseCannon-12B-v2 itself is a merge.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p