DrNicefellow/Mistral-7-from-Mixtral-8x7B-v0.1

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

DrNicefellow/Mistral-7-from-Mixtral-8x7B-v0.1 is a 7 billion parameter experimental language model derived from the mistralai/Mixtral-8x7B-v0.1 architecture. It is constructed by extracting the 7th expert from each Mixture of Experts (MoE) layer of the base Mixtral model. This model is designed for general language understanding and generation tasks, though its performance is expected to be lower than the original Mistral-7B due to its experimental extraction method.

Loading preview...

Overview

DrNicefellow/Mistral-7-from-Mixtral-8x7B-v0.1 is an experimental 7 billion parameter language model. It is a standalone model extracted from the larger mistralai/Mixtral-8x7B-v0.1 model using a custom tool, the Mixtral Model Expert Extractor. This specific model is built by taking the 7th expert from each Mixture of Experts (MoE) layer of the original Mixtral architecture.

Key Characteristics

  • Architecture: Derived from Mixtral-8x7B-v0.1, featuring multi-head attention layers.
  • Extraction Method: Created by isolating the 7th expert from each MoE layer, making it a unique, experimental configuration.
  • Expected Performance: Due to its experimental nature, this model is anticipated to perform below the standard Mistral-7B model.
  • License: Available under the Apache 2.0 License.

Use Cases

This model can be used for various language understanding and generation tasks, similar to other 7B parameter models. Developers interested in exploring the behavior of individual experts within a Mixture of Experts model may find this model particularly useful for research and experimentation. It provides a unique perspective on how specific expert components contribute to overall model functionality.

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