qwen2.5-reinstruct-alternate-lumen-14B Benchmark Results

qwen2.5-reinstruct-alternate-lumen-14B: benchmark results

Lambent's community merge of Qwen 2.5 14B variants with the Lumen LoRA. Provider: Lambent. Released 2024-09-23. Access: Open.

Aggregate Rating 1598 ± 19, rank #452 of 1661 rated models, from 21 benchmark results.

Benchmark	Score	Metric	Percentile
Open Ko-LLM Leaderboard	54.07	Average (%)	99.4
Open Korean LLM Leaderboard	54.07	Average Score (%)	99.4
Open Korean LLM - Ko-GSM8K (flexible extract)	75.36	Exact match, flexible extract (%)	99
Open Korean LLM - Ko-IFEval	67.07	Strict accuracy, prompt/instruction mean (%)	98.8
Open Korean LLM - KorNAT-CKA	39.66	Normalized accuracy (%)	98.8
Open Korean LLM - Ko-EQ-Bench	59.26	EQ-Bench score	98.7
Open Portuguese LLM - ASSIN2 RTE	94.69	Macro F1 (%)	98
Open Korean LLM - KorNAT-Helpful	60.48	Normalized accuracy (%)	97.9
Open Korean LLM - KorNAT-Harmless	72.92	Normalized accuracy (%)	95.8
Open Korean LLM - Ko-Winogrande	69.77	Accuracy (%)	95.6
Open LLM Leaderboard - MMLU-Pro	53.88	Score	95.5
Open Portuguese LLM - ENEM	80.97	Accuracy (%)	94.6