LukeBailey181Pub commited on
Commit
94fb754
·
verified ·
1 Parent(s): 1af6627

Trim benchmark-versions section to the table

Browse files
Files changed (1) hide show
  1. README.md +0 -26
README.md CHANGED
@@ -63,32 +63,6 @@ The `benchmark_version` column records which generation a paper belongs to:
63
  | `1` | 35 | the original release; the results in the paper are on these |
64
  | `2` | 65 | added in the expansion |
65
 
66
- To reproduce the numbers reported in the paper, restrict to the original 35:
67
-
68
- ```python
69
- from datasets import load_dataset
70
-
71
- ds = load_dataset("LukeBailey181Pub/ArxivMathGradingBench")["train"]
72
- original_35 = ds.filter(lambda r: r["benchmark_version"] == 1)
73
- ```
74
-
75
- The 35-paper release is also tagged, with its original column names intact:
76
-
77
- ```python
78
- load_dataset("LukeBailey181Pub/ArxivMathGradingBench", revision="v1")
79
- ```
80
-
81
- Note that `benchmark_version` (which generation of the *benchmark*) is a
82
- different thing from `version` (which arXiv version of the *paper* contains
83
- the error). Both take values like 1/v1 and 2/v2, so filter on the right one.
84
-
85
- Two columns changed in the expansion. `title_extracted_from_tex` was renamed
86
- to `title` and now holds the arXiv metadata title for every paper, rather than
87
- the literal TeX `\title{}` body; the titles name the same papers, but without
88
- paper-local macros or source casing. `Liscence` was dropped — licence is a
89
- property of the paper on arXiv, where it is authoritative. Both are still
90
- available at the `v1` tag.
91
-
92
  ## Downloading the papers
93
 
94
  This dataset provides the paper metadata and error annotations. To download
 
63
  | `1` | 35 | the original release; the results in the paper are on these |
64
  | `2` | 65 | added in the expansion |
65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66
  ## Downloading the papers
67
 
68
  This dataset provides the paper metadata and error annotations. To download