amylynn/Riverfish-Rocinante-12B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 13, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

amylynn/Riverfish-Rocinante-12B is a 12 billion parameter language model created by amylynn, formed by merging several Rocinante and Rivermind-Lux models from TheDrummer. This model leverages a sequence of merge methods including model_stock, karcher, and arcee_fusion to combine the strengths of its constituent models. With a context length of 32768 tokens, it is designed to integrate diverse capabilities from its merged components, making it suitable for general-purpose language tasks.

Loading preview...

Model Overview

amylynn/Riverfish-Rocinante-12B is a 12 billion parameter language model developed by amylynn. It is a composite model, created through a sophisticated merging process of several base models from TheDrummer, specifically:

  • TheDrummer/Rocinante-12B-v1
  • TheDrummer/Rocinante-12B-v1.1
  • TheDrummer/Rocinante-X-12B-v1
  • TheDrummer/UnslopNemo-12B-v4.1
  • TheDrummer/Rivermind-Lux-12B-v1

Merging Methodology

The model was constructed using LazyMergekit through a multi-step sequence of merging methods:

  • Step 1 (model_stock): Rocinante-12B-v1, Rocinante-12B-v1.1, and Rocinante-X-12B-v1 were merged, with normalization applied.
  • Step 2 (karcher): The output from Step 1 was then merged with UnslopNemo-12B-v4.1 using the Karcher method, involving 500 iterations and normalization.
  • Step 3 (arcee_fusion): Finally, the result from Step 2 was fused with Rivermind-Lux-12B-v1 using the Arcee Fusion method to produce the final Riverfish-Rocinante-12B model.

All merging operations were performed using bfloat16 dtype, ensuring efficient parameter handling. This intricate merging strategy aims to consolidate and enhance the diverse capabilities present in its constituent models.

Key Characteristics

  • Parameter Count: 12 billion parameters.
  • Context Length: Supports a context window of 32768 tokens.
  • Composition: A blend of multiple specialized models, suggesting a broad range of potential applications.

Good For

  • Developers looking for a robust 12B parameter model that integrates the strengths of several established base models.
  • Applications requiring a model with a substantial context window for processing longer inputs.
  • Experimentation with merged models to leverage combined capabilities for various language generation and understanding tasks.

Finetunes
5 models