Minitron-4B-Base: benchmark results
NVIDIA's Minitron base transformer, obtained by pruning Nemotron-4 15B's embedding width, attention heads and MLP dimensions, followed by distillation and continued training on 94 billion tokens. It is distinct from instruction-tuned and Hindi-specialized derivatives. Provider: NVIDIA. Released 2024-07-19. Access: Open.
Aggregate Rating 1359 ± 51, rank #7436 of 10552 rated models, from 33 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 41.34 | Score | 53.4 |
| Open PL LLM - PolEmo2-IN (multiple choice, 5-shot) | 76.87 | Accuracy (%) | 50.9 |
| Open PL LLM - PolQA Open Book (generative, 5-shot) | 79.17 | Levenshtein Similarity (%) | 47.8 |
| Open PL LLM - PPC (generative, 5-shot) | 63.9 | Accuracy (%) | 44.7 |
| Open PL LLM - Belebele (generative, 5-shot) | 73.56 | Accuracy (%) | 44.6 |
| Open PL LLM - PolEmo2-OUT (generative, 5-shot) | 65.59 | Accuracy (%) | 44.6 |
| Open PL LLM - PoQuAD Open Book (generative, 5-shot) | 59.92 | Levenshtein Similarity (%) | 42.3 |
| Open PL LLM - Belebele (multiple choice, 5-shot) | 72.56 | Accuracy (%) | 42.1 |
| Open PL LLM - PSC (multiple choice, 5-shot) | 50.34 | Binary F1 (%) | 37.9 |
| Open PL LLM - Generative | 39.1 | Average Generative Score (%) | 36 |
| Open PL LLM - PolQA Closed Book (generative, 5-shot) | 39.67 | Levenshtein Similarity (%) | 35.5 |
| Open PL LLM - CBD (multiple choice, 5-shot) | 18.56 | Macro F1 (%) | 34 |
Interactive version: theaggregate.ai/model?slug=minitron-4b-base · How It Works · Data refreshed daily, snapshot 2026-11-16.