Sina-Loki-7B-Merge: benchmark results
Azazelle's DARE-TIES merge, whose published recipe uses RatanRohith/SRBOSGPT-7B-slerp as its base and rishiraj/smol-7b, SanjiWatsuki/openchat-3.5-1210-starling-slerp, Azazelle/Dumb-Maidlet as merge inputs. Its published Mistral configuration specifies 32 decoder layers. Provider: Azazelle. Access: Open.
Aggregate Rating 1409 ± 33, rank #5660 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 64.71 | Accuracy (%) (5-shot) | 75.3 |
| Open LLM Leaderboard v1 - GSM8K | 52.39 | Accuracy (%) (5-shot) | 64.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 53.84 | MC2 (%) (0-shot) | 60.9 |
| Open LLM Leaderboard v1 - WinoGrande | 78.14 | Accuracy (%) (5-shot) | 60.9 |
| Open LLM Leaderboard v1 - HellaSwag | 81.96 | Normalized accuracy (%) (10-shot) | 48 |
| Open LLM Leaderboard v1 - ARC Challenge | 59.13 | Normalized accuracy (%) (25-shot) | 42.2 |
Interactive version: theaggregate.ai/model?slug=sina-loki-7b-merge · How It Works · Data refreshed daily, snapshot 2026-11-08.