Commit
•
496b27b
1
Parent(s):
acae8d5
Update README.md
Browse files
README.md
CHANGED
|
@@ -11,7 +11,8 @@ size_categories:
|
|
| 11 |
- 10K<n<100K
|
| 12 |
viewer: false
|
| 13 |
---
|
| 14 |
-
|
|
|
|
| 15 |
FinEval is a collection of high-quality multiple-choice questions covering various domains such as finance, economics, accounting, and certifications. It consists of 4,661 questions spanning across 34 distinct academic subjects. To ensure a comprehensive assessment of model performance, FinEval employs various methods including zero-shot, few-shot, answer-only, and chain-of-thought prompts. Evaluating state-of-the-art large language models in both Chinese and English on FinEval reveals that only GPT-4 achieves an accuracy of 60% across different prompt settings, highlighting substantial growth potential of large language models in financial domain knowledge. Our work provides a more comprehensive benchmark for evaluating financial knowledge, utilizing simulated exam data and encompassing a wide range of large language model assessments.
|
| 16 |
|
| 17 |
Each subject consists of three splits: dev, val, and test. The dev set per subject consists of five exemplars with explanations for few-shot evaluation. The val set is intended to be used for hyperparameter tuning. And the test set is for model evaluation. Labels on the test split are not released, users are required to submit their results to automatically obtain test accuracy.
|
|
|
|
| 11 |
- 10K<n<100K
|
| 12 |
viewer: false
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
|
| 16 |
FinEval is a collection of high-quality multiple-choice questions covering various domains such as finance, economics, accounting, and certifications. It consists of 4,661 questions spanning across 34 distinct academic subjects. To ensure a comprehensive assessment of model performance, FinEval employs various methods including zero-shot, few-shot, answer-only, and chain-of-thought prompts. Evaluating state-of-the-art large language models in both Chinese and English on FinEval reveals that only GPT-4 achieves an accuracy of 60% across different prompt settings, highlighting substantial growth potential of large language models in financial domain knowledge. Our work provides a more comprehensive benchmark for evaluating financial knowledge, utilizing simulated exam data and encompassing a wide range of large language model assessments.
|
| 17 |
|
| 18 |
Each subject consists of three splits: dev, val, and test. The dev set per subject consists of five exemplars with explanations for few-shot evaluation. The val set is intended to be used for hyperparameter tuning. And the test set is for model evaluation. Labels on the test split are not released, users are required to submit their results to automatically obtain test accuracy.
|