Bielik-7B-Instruct-v0.1: benchmark results
SpeakLeash and ACK Cyfronet AGH's Bielik-7B-Instruct-v0.1, an instruction-tuned fine-tune of Bielik-7B-v0.1 trained on curated Polish text corpora using Poland's PLGrid supercomputing infrastructure. Provider: SpeakLeash. Released 2024-03-30. Access: Open.
Aggregate Rating 1382 ± 14, rank #1266 of 1635 rated models, from 102 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open PL LLM - PolQA Closed Book (generative, 5-shot) | 69.78 | Levenshtein Similarity (%) | 77 |
| Open PL LLM - PolQA Open Book (generative, 5-shot) | 86 | Levenshtein Similarity (%) | 75.1 |
| Open PL LLM - CBD (multiple choice, 5-shot) | 29.4 | Macro F1 (%) | 68.7 |
| Open PL LLM - CBD (generative, 5-shot) | 30.69 | Macro F1 (%) | 68.6 |
| MT-Bench PL - Reasoning | 6.15 | Judge Score (0-10) | 62.2 |
| Open PL LLM - PSC (multiple choice, 5-shot) | 70.3 | Binary F1 (%) | 61.1 |
| Open PL LLM - 8TAGS (generative, 5-shot) | 73.22 | Accuracy (%) | 56.6 |
| Open PL LLM - Multiple Choice | 42.52 | Average Multiple-Choice Score (%) | 55.2 |
| Open PL LLM - DYK (multiple choice, 5-shot) | 51.55 | Binary F1 (%) | 55.1 |
| Open PL LLM - DYK (generative, 5-shot) | 52.7 | Binary F1 (%) | 54.1 |
| Open PL LLM - Generative | 50.19 | Average Generative Score (%) | 53.5 |
| Open PL LLM Leaderboard | 44.7 | Average Score (%) | 51.7 |
Interactive version: theaggregate.ai/model?slug=bielik-7b-instruct-v0-1 · How It Works · Data refreshed daily, snapshot 2026-10-15.