pythia-1B-deduped: benchmark results
EleutherAI's Pythia-1B-deduped, the deduplicated-data variant in the eight-size Pythia Scaling Suite built for interpretability research, trained on a globally deduplicated version of the Pile. Provider: EleutherAI. Released 2023-04-03. Access: Open.
Aggregate Rating 1102 ± 26, rank #1628 of 1640 rated models, from 22 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Portuguese LLM - FaQuAD NLI | 43.97 | Macro F1 (%) | 17.1 |
| Open LLM Leaderboard v1 - GSM8K | 1.14 | Accuracy (%) (5-shot) | 15.7 |
| Open Portuguese LLM - ENEM | 19.94 | Accuracy (%) | 11.9 |
| Open Portuguese LLM - OAB Exams | 23.78 | Accuracy (%) | 11 |
| Open Medical LLM - MMLU Medical Genetics | 32 | Accuracy (%) | 10.9 |
| Open Portuguese LLM - BLUEX | 20.58 | Accuracy (%) | 10.4 |
| Open LLM Leaderboard v1 - HellaSwag | 49.65 | Normalized accuracy (%) (10-shot) | 10.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 38.94 | MC2 (%) (0-shot) | 9.6 |
| Open Portuguese LLM - ASSIN2 RTE | 34.09 | Macro F1 (%) | 8.6 |
| Open Medical LLM - MedMCQA | 30.6 | Accuracy (%) | 8.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 29.1 | Normalized accuracy (%) (25-shot) | 8.1 |
| Open LLM Leaderboard v1 - WinoGrande | 53.59 | Accuracy (%) (5-shot) | 7.9 |
Interactive version: theaggregate.ai/model?slug=pythia-1b-deduped · How It Works · Data refreshed daily, snapshot 2026-11-07.