Welcome, brave one; you've come a long mile.

NOTE for newer users: "R1" here means "Revision 1". This model predates DeepSeek's R1; DeepSeek inadvertently made using this versioning scheme very annoying!

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

Mag Mell R1 was tested with Temp 1.25 and MinP 0.2. This was fairly stable up to 10K, but this might be too "hot". If issues with coherency occur, try increasing MinP or decreasing Temperature.

Other samplers shouldn't be necessary. XTC was shown to break outputs. DRY should be okay if used sparingly. Other penalty-type samplers should probably be avoided.

The base model for Mag Mell is Mistral-Nemo-Base-2407-chatml, and as such ChatML formatting is recommended.

Early testing versions had a tendency to leak tokens, but this should be more or less hammered out. It recently (12-18-2024) came to attention that Cache Quantization may either cause or exacerbate this issue.

Mag Mell is a multi-stage merge, Inspired by hyper-merges like Tiefighter and Umbral Mind. Intended to be a general purpose "Best of Nemo" model for any fictional, creative use case.

6 models were chosen based on 3 categories; they were then paired up and merged via layer-weighted SLERP to create intermediate "specialists" which are then evaluated in their domain. The specialists were then merged into the base via DARE-TIES, with hyperparameters chosen to reduce interference caused by the overlap of the three domains. The idea with this approach is to extract the best qualities of each component part, and produce models whose task vectors represent more than the sum of their parts.

The three specialists are as follows:

I've been dreaming about this merge since Nemo tunes started coming out in earnest. From our testing, Mag Mell demonstrates worldbuilding capabilities unlike any model in its class, comparable to old adventuring models like Tiefighter, and prose that exhibits minimal "slop" (not bad for no finetuning,) frequently devising electrifying metaphors that left us consistently astonished.

I don't want to toot my own bugle though; I'm really proud of how this came out, but please leave your feedback, good or bad.

Special thanks as usual to Toaster for his feedback and Fizz for helping fund compute, as well as the KoboldAI Discord for their resources.

This model was merged using the DARE TIES merge method using IntervitensInc/Mistral-Nemo-Base-2407-chatml as a base.

The following models were included in the merge:

The following YAML configurations were used to produce this model: