SnowflakeCore vs GPT-2

Comprehensive Benchmark Comparison

SnowflakeCore Demonstrates Strong Performance in Text Quality Metrics

N-gram Diversity Comparison

Text Generation Speed

N-gram Repetition Rate

Vocabulary Diversity

Key Results Summary

Improved N-gram Diversity

SnowflakeCore models show higher distinct bigram and trigram ratios.

Reduced Repetition

Significantly lower n-gram repetition rates, especially for trigrams.

Enhanced Vocabulary Usage

G1-Tiny2 achieves the highest vocabulary diversity ratio.

Speed Comparison

GPT-2 is faster at ~47 tok/sec compared to SnowflakeCore's ~22-23 tok/sec.

Model Specifications

SnowflakeCore-G1-Tiny

Parameters: 355,871,744

Size: 1357.54 MB

Speed: 22.12 tok/sec

SnowflakeCore-G1-Tiny2

Parameters: 355,871,744

Size: 1357.54 MB

Speed: 22.13 tok/sec

GPT-2

Parameters: 124,439,808

Size: 474.70 MB

Speed: 47.73 tok/sec

Conclusion: SnowflakeCore demonstrates superior text quality metrics with comparable parameter efficiency, indicating architectural advancements over GPT-2 in these specific areas.