ichigoberry/MonarchPipe-7B-slerp

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Apr 1, 2024License:cc-by-nc-2.0Architecture:Transformer0.0K Open Weights Warm

MonarchPipe-7B-slerp is a 7 billion parameter language model created by ichigoberry, formed by merging OpenPipe/mistral-ft-optimized-1227 and mlabonne/AlphaMonarch-7B using a slerp method. This model leverages the Mistral architecture and offers an 8192-token context length. It is designed to combine the strengths of its base models, showing competitive performance in general language understanding and reasoning tasks.

Loading preview...

MonarchPipe-7B-slerp Overview

MonarchPipe-7B-slerp is a 7 billion parameter language model developed by ichigoberry. It is a merged model, combining the capabilities of OpenPipe/mistral-ft-optimized-1227 and mlabonne/AlphaMonarch-7B using a slerp (spherical linear interpolation) merge method via LazyMergekit. This merging technique aims to blend the strengths of its constituent models, which are both based on the Mistral architecture.

Key Capabilities & Performance

The model demonstrates competitive performance across various benchmarks, as evaluated by the Nous benchmark suite. While its overall average score is slightly below AlphaMonarch-7B, it shows strong results in specific areas:

  • AGIEval: Achieves 46.12, outperforming AlphaMonarch-7B in this specific metric.
  • General Reasoning: Designed to handle a range of general language understanding and reasoning tasks, benefiting from the fine-tuning of its base models.

Configuration & Usage

The merge configuration utilizes a slerp method with specific t parameters applied to different layers (self_attn and mlp) to fine-tune the blending process. The model supports a context length of 8192 tokens and is intended for use with bfloat16 data types. Developers can easily integrate MonarchPipe-7B-slerp into their projects using the Hugging Face transformers library, with provided Python code examples for text generation.

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