Silicon-Medley: benchmark results
Azazelle's SLERP merge, whose published recipe uses mistralai/Mistral-7B-v0.1 as its base and SanjiWatsuki/Silicon-Maid-7B, chargoddard/piano-medley-7b as merge inputs. Its published Mistral configuration specifies 32 decoder layers. Provider: Azazelle. Access: Open.
Aggregate Rating 1433 ± 34, rank #4543 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - HellaSwag | 86.21 | Normalized accuracy (%) (10-shot) | 79.6 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 61.34 | MC2 (%) (0-shot) | 77.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 67.24 | Normalized accuracy (%) (25-shot) | 76.2 |
| Open LLM Leaderboard v1 - MMLU | 64.51 | Accuracy (%) (5-shot) | 71.8 |
| Open LLM Leaderboard v1 - GSM8K | 58.38 | Accuracy (%) (5-shot) | 70.1 |
| Open LLM Leaderboard v1 - WinoGrande | 79.24 | Accuracy (%) (5-shot) | 68.1 |
Interactive version: theaggregate.ai/model?slug=silicon-medley · How It Works · Data refreshed daily, snapshot 2026-11-08.