FrankenDPO-4x7B-bf16: benchmark results
Kquant03's mixture-of-experts release whose published Mixtral configuration specifies 4 local experts, 2 selected per token and 32 decoder layers. Its recipe names SanjiWatsuki/Kunoichi-DPO-v2-7B as the base; nominal expert sizes are not measured total or active parameter counts. The card and recipe disagree on the Neuronovo expert release, so that expert identity remains unresolved. Provider: Kquant03. Access: Open.
Aggregate Rating 1449 ± 36, rank #3827 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 83.5 | Accuracy (%) (5-shot) | 89.2 |
| Open LLM Leaderboard v1 - GSM8K | 67.7 | Accuracy (%) (5-shot) | 84.2 |
| Open LLM Leaderboard v1 - ARC Challenge | 68.69 | Normalized accuracy (%) (25-shot) | 81 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 63.14 | MC2 (%) (0-shot) | 80.2 |
| Open LLM Leaderboard v1 - MMLU | 64.93 | Accuracy (%) (5-shot) | 78.8 |
| Open LLM Leaderboard v1 - HellaSwag | 86.07 | Normalized accuracy (%) (10-shot) | 78.5 |
Interactive version: theaggregate.ai/model?slug=frankendpo-4x7b-bf16 · How It Works · Data refreshed daily, snapshot 2026-11-08.