GeneticMerge - Advanced LLM Model Merging Platform

GeneticMerge Dashboard

Model Selection

Choose models to merge from HuggingFace

1. Select Base Model Architecture

2. Search Available Finetunes(Qwen2 0.5B Instruct)

Available Finetunes

No finetunes found

Try searching for a specific model

3. Selected Finetunes for Merging (0/4)

Select finetunes to begin layer merging

Genetic Algorithm

Configure merge parameters

Population Parameters

Population Size: 8

Generations: 3

Genetic Operators

Mutation Rate: 0.10

Crossover Rate: 0.80

Elitism Rate: 0.10

Evaluation Settings

Use Subset Evaluation

Subset Size: 100

Evaluate on a random subset of samples to speed up evaluation

Parallel Evaluation

Enable Parallel Evaluation

Max Parallel Evaluations: 8

Number of models to evaluate simultaneously (higher = faster but more expensive)

Merge Visualization

Real-time view of model layer merging process

No Finetunes Selected

Select finetunes to begin layer merging

Benchmarks & Results

Evaluate merged model performance

Completed: 2/5Avg Score: 76.0

MMLU

Massive Multitask Language Understanding

67.8/ 100

Duration: 12m 34s

completed

HellaSwag

Commonsense NLI

84.2/ 100

Duration: 8m 12s

completed

ARC-Challenge

AI2 Reasoning Challenge

Running...5m 23s

Duration: 5m 23s

running

TruthfulQA

Truthfulness in QA

Waiting to run

Duration: -

pending

GSM8K

Grade School Math

Waiting to run

Duration: -

pending

Performance Summary

76.0

Average Score

Tests Completed

Tests Running

84.2

Best Score