Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
configs:
|
| 4 |
+
- config_name: default
|
| 5 |
+
data_files:
|
| 6 |
+
- split: v0.1.4
|
| 7 |
+
path: data/v0.1.4.parquet
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# BigCodeBench (ported)
|
| 11 |
+
|
| 12 |
+
Fork of [bigcode/bigcodebench](https://huggingface.co/datasets/bigcode/bigcodebench) v0.1.4.
|
| 13 |
+
|
| 14 |
+
`canonical_solution` and `test` are ported to current numpy / pandas / scikit-learn
|
| 15 |
+
so that the reference solutions pass on a modern stack. Prompts
|
| 16 |
+
(`complete_prompt`, `instruct_prompt`, `code_prompt`) are unchanged, so the task a
|
| 17 |
+
model is asked to solve is identical to upstream.
|
| 18 |
+
|
| 19 |
+
Groundtruth pass rate: 1.000 over all 1140 tasks.
|