AIbase
EN
Home
Mera Mix 4x7B
M

Mera Mix 4x7B

Developed by meraGPT
mera-mix-4x7B is a Mixture of Experts (MoE) model with half the scale of Mixtral-8x7B but comparable performance and faster inference speed.
Large Language Model
Transformers
Open Source License:Apache-2.0#Efficient MoE Model#Inference Acceleration#Multi-task Text Generation
Downloads 2,375
Release Time : 4/13/2024

Model Overview

This is a 4-expert mixture model designed to replace Mixtral-8x7B, offering similar text generation capabilities with more efficient inference performance.

Model Features

Efficient Inference
Compared to Mixtral-8x7B, the parameter scale is halved while maintaining comparable performance, with faster inference speed.
Outstanding Performance
Scored 75.91 in the OpenLLM evaluation, outperforming Mixtral-8x7B and Mixtral-8x22B.
Mixture of Experts Architecture
Adopts a 4-expert mixture design, balancing model capacity with computational efficiency.

Model Capabilities

Text Generation
Reasoning Task Processing
Multiple-choice Question Answering
Mathematical Problem Solving

Use Cases

Academic Evaluation
AI2 Reasoning Challenge
Handling complex reasoning problems
Standardized accuracy 72.95
MMLU Evaluation
Multidisciplinary knowledge understanding and application
Accuracy 64.44
Common-sense Reasoning
HellaSwag Evaluation
Common-sense reasoning for everyday situations
Standardized accuracy 89.17
Winogrande Evaluation
Coreference resolution task
Accuracy 85.64
Truthfulness Evaluation
TruthfulQA Evaluation
Truthful question answering
mc2 score 77.17
Mathematical Problem Solving
GSM8k Evaluation
Elementary school math problem solving
Accuracy 66.11
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
English简体中文繁體中文にほんご
© 2025AIbase