Bianca-7B [float16] [rev: c962dd3095ff]: benchmark results
Azazelle's rescaled_sample merge, whose published recipe uses mistralai/Mistral-7B-v0.1 as its base and Nexusflow/Starling-LM-7B-beta, jan-hq/supermario-slerp-v3, Endevor/InfinityRP-v1-7B, NeverSleep/Noromaid-7B-0.4-DPO as merge inputs. Its published Mistral configuration specifies 32 decoder layers. The card heading retains another label; the exact repository and its recipe define this publication, without a global name alias. This catalog row preserves its explicit evaluation dtype label; shared checkpoint revision does not establish identical evaluator precision or protocol. Provider: Azazelle. Access: Open.
Aggregate Rating 1448 ± 36, rank #3839 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 71.27 | Accuracy (%) (5-shot) | 95.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 69.71 | Normalized accuracy (%) (25-shot) | 84 |
| Open LLM Leaderboard v1 - MMLU | 65.25 | Accuracy (%) (5-shot) | 83.4 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 63.95 | MC2 (%) (0-shot) | 81.5 |
| Open LLM Leaderboard v1 - HellaSwag | 86.11 | Normalized accuracy (%) (10-shot) | 78.8 |
| Open LLM Leaderboard v1 - WinoGrande | 80.9 | Accuracy (%) (5-shot) | 76.2 |
Interactive version: theaggregate.ai/model?slug=bianca-7b-float16-rev-c962dd3095ff · How It Works · Data refreshed daily, snapshot 2026-11-08.