allknowingroger/LimmyAutomerge-7B-slerp

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kLicense:apache-2.0Architecture:Transformer Open Weights Cold

LimmyAutomerge-7B-slerp is a 7 billion parameter language model created by allknowingroger, formed by a slerp merge of automerger/MeliodasNeuralsirkrishna-7B and liminerity/M7-7b. This model leverages a specific layer-wise merging strategy to combine the strengths of its constituent models. It is designed for general text generation tasks, offering a balanced performance derived from its merged architecture.

Loading preview...

Model Overview

LimmyAutomerge-7B-slerp is a 7 billion parameter language model developed by allknowingroger. It is constructed using a slerp (spherical linear interpolation) merge method via LazyMergekit, combining two distinct base models: automerger/MeliodasNeuralsirkrishna-7B and liminerity/M7-7b.

Merge Configuration

The merge process specifically targets all 32 layers of both source models. A key aspect of this merge is the differentiated t parameter application:

  • Self-attention layers: The merge ratio t varies across layers, with values like [0, 0.5, 0.3, 0.7, 1] applied.
  • MLP layers: The merge ratio t also varies, using values such as [1, 0.5, 0.7, 0.3, 0].
  • Other parameters: A default t value of 0.5 is applied.

This fine-grained control over the merge ratios for different layer types aims to optimize the combined model's performance characteristics. The model uses bfloat16 for its data type.

Usage

Developers can integrate LimmyAutomerge-7B-slerp using the Hugging Face transformers library. The provided Python example demonstrates how to load the model and tokenizer, apply a chat template, and generate text with specified parameters like max_new_tokens, temperature, top_k, and top_p.

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