internlm2-chat-7B-ExPO: benchmark results
chujiezheng's ExPO variant of internlm2-chat-7b (2024), extrapolating between SFT and RLHF weights (alpha=0.5) to boost alignment without extra training. Provider: chujiezheng. Released 2024-05-02. Access: Open.
Aggregate Rating 1449 ± 37, rank #3806 of 10916 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Chinese LLM - HellaSwag | 62.81 | Accuracy (%) | 69.7 |
| Open Chinese LLM - ARC Challenge | 53.24 | Accuracy (%) | 54.4 |
| AlpacaEval 2.0 | 22.67 | LC Win Rate (%) | 54.1 |
| Open Chinese LLM - TruthfulQA MC | 53.96 | Accuracy (%) | 53.4 |
| Open Chinese LLM - WinoGrande | 62.98 | Accuracy (%) | 51 |
| Open Chinese LLM - CMMLU | 53.44 | Accuracy (%) | 50.3 |
| Open Chinese LLM - C-Eval Semantic | 67.38 | Accuracy (%) | 43.2 |
| Open Chinese LLM Leaderboard | 56.22 | Official Nonzero-Task Mean (%) | 39.6 |
| Open Chinese LLM - GSM8K | 39.73 | Accuracy (%) | 38.8 |
Interactive version: theaggregate.ai/model?slug=internlm2-chat-7b-expo · How It Works · Data refreshed daily, snapshot 2026-11-08.