Qwen2.5-32B-Instruct-abliterated-v2 Benchmark Results

Qwen2.5-32B-Instruct-abliterated-v2 — benchmark results

Community abliterated (refusal-removed) build of Qwen 2.5 32B Instruct. Provider: Alibaba. Released 2024-10-11. Access: Open.

Unified ELO 1586 ± 15, rank #512 of 1776 rated models, from 183 benchmark results.

Benchmark	Score	Metric	Percentile
Open Arabic LLM - Arabic MMLU Management (University)	82.67	Accuracy (%)	100
Open LLM Leaderboard - MATH Level 5	59.52	Score	99.7
Open LLM Leaderboard - IFEval	83.34	Score	99.3
Open LLM Leaderboard - MMLU-Pro	51.35	Score	98.8
Open LLM Leaderboard - BBH	56.53	Score	98.5
Open Arabic LLM - Arabic MMLU HT Virology	53.01	Accuracy (%)	96.6
Open Japanese LLM - Jsick Exact Match	86.54	Score (%)	96.6
Open Arabic LLM - Arabic MMLU Economics (High School)	74.44	Accuracy (%)	95.7
Open Arabic LLM - Aratrust Ethics	81.67	Accuracy (%)	95.7
Open Japanese LLM - Jamp Exact Match	71.55	Score (%)	95.2
Open Arabic LLM - Arabic MMLU HT Computer Security	73	Accuracy (%)	95.1
Open Arabic LLM - Arabic MMLU Geography (Middle School)	74.26	Accuracy (%)	94.8