Sakura-SOLAR-Instruct Overview

Sakura-SOLAR-Instruct is a 10.7 billion parameter instruction-tuned language model developed by Kyujin Han, in collaboration with Media Group Saramgwasup and Marker's LLM research consortium. This model was created using the Mergekit method, with detailed information about its training and code available in the Sakura-SOLAR GitHub repository.

Key Capabilities & Performance

Sakura-SOLAR-Instruct demonstrates strong performance across a suite of benchmarks, achieving an average score of 74.40 on the Open LLM Leaderboard. Its benchmark results include:

AI2 Reasoning Challenge (25-Shot): 70.99
HellaSwag (10-Shot): 88.42
MMLU (5-Shot): 66.33
TruthfulQA (0-shot): 71.79
Winogrande (5-shot): 83.66
GSM8k (5-shot): 65.20

Notably, the model achieved a Rank 1 position on the Open LLM Leaderboard on December 27, 2023, at 11:50 PM, highlighting its competitive performance among similar models.

Use Cases

Given its strong benchmark results in reasoning, common sense, and language understanding, Sakura-SOLAR-Instruct is well-suited for applications requiring:

General instruction following
Question answering
Text generation
Reasoning tasks