alpaca-7B: benchmark results
Stanford Alpaca 7B instruction model. The download is Stanford's official weight-difference package, not standalone weights; reconstruct it with the original LLaMA 7B base using the model card instructions. Provider: Stanford CRFM. Released 2023-03-18. Access: Open.
Aggregate Rating 1166 ± 12, rank #1611 of 1656 rated models, from 96 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM Classic - LSAT | 22.17 | Exact Match (%) | 79.4 |
| HELM Classic - Entity Matching | 85.26 | Exact Match (%) | 78.8 |
| HELM Classic - CivilComments | 56.58 | Exact Match (%) | 68.2 |
| MMLU-by-task - TruthfulQA MC2 | 48.49 | Accuracy (%) | 67.8 |
| HELM Classic - BoolQ | 77.8 | Exact Match (%) | 66.7 |
| HELM Classic - Synthetic Reasoning Natural | 21.49 | F1 (%) | 64.7 |
| HELM Classic - bAbI | 50.33 | Exact Match (%) | 60.9 |
| HELM Classic - MMLU | 38.46 | Exact Match (%) | 57.6 |
| InstructEval - Problem Solving | 32.4 | Average (%, MMLU/BBH/DROP/CRASS/HumanEval) | 57.1 |
| HELM Classic - TruthfulQA | 24.31 | Exact Match (%) | 56.8 |
| MMLU-by-task - High School Mathematics | 27.41 | Accuracy (%) | 54.7 |
| HELM Classic - MATH | 10.43 | Equivalent (%) | 52.9 |
Interactive version: theaggregate.ai/model?slug=alpaca-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.