Patrick Haller commited on
Commit
08cfea7
1 Parent(s): 983ff7e
Files changed (3) hide show
  1. app.py +29 -4
  2. current_results.json +10 -2
  3. text.py +9 -8
app.py CHANGED
@@ -41,12 +41,37 @@ with demo:
41
  gr.HTML(f"<h3>{INTRODUCTION_TEXT}<h3>")
42
  gr.HTML('<hr style="border-top: 3px dotted #bbb" class="dotted">')
43
 
 
 
 
44
  gr.HTML("<h3>📊 Results</h3>")
45
- gr.components.Dataframe(
46
- value=df,
47
- datatype=["html"]
 
 
 
48
 
49
- )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50
 
51
  gr.HTML("<h3>Legend</h3>")
52
  gr.HTML("<p>🔶: Evaluated only on the first part of each AoC day</p>")
 
41
  gr.HTML(f"<h3>{INTRODUCTION_TEXT}<h3>")
42
  gr.HTML('<hr style="border-top: 3px dotted #bbb" class="dotted">')
43
 
44
+ gr.HTML("<h2 style='text-align: center'><a href='https://github.com/HallerPatrick/pecc'>Github Repository</a></h2>")
45
+ gr.HTML("<p style='text-align: center'>Our paper got accepted at LREC-COLING 2024! The paper will be available soon. 🤗</p>")
46
+
47
  gr.HTML("<h3>📊 Results</h3>")
48
+ gr.HTML("<p>Results are reported as <b>Pass@3 Accuracy</b>.</p>")
49
+ with gr.Tabs() as tabs:
50
+ with gr.TabItem("All"):
51
+ gr.components.Dataframe(
52
+ value=df,
53
+ datatype=["html"]
54
 
55
+ )
56
+ with gr.TabItem("AoC"):
57
+ aoc_df = df[["Model", "Evaluation", "AOC Original", "AOC Leet"]]
58
+ average_scores = aoc_df.iloc[:, 2:].mean(axis=1).round(2)
59
+ aoc_df.insert(loc=2, column="⬆️ Average", value=average_scores)
60
+ aoc_df = aoc_df.sort_values(by=["Evaluation", "⬆️ Average"], ascending=[False, False])
61
+ gr.components.Dataframe(
62
+ value=aoc_df,
63
+ datatype=["html"]
64
+ )
65
+
66
+ with gr.TabItem("Euler"):
67
+ euler_df = df[["Model", "Euler Original", "Euler Story"]]
68
+ average_scores = euler_df.iloc[:, 1:].mean(axis=1).round(2)
69
+ euler_df.insert(loc=1, column="⬆️ Average", value=average_scores)
70
+ euler_df = euler_df.sort_values(by="⬆️ Average", ascending=False)
71
+ gr.components.Dataframe(
72
+ value=euler_df,
73
+ datatype=["html"]
74
+ )
75
 
76
  gr.HTML("<h3>Legend</h3>")
77
  gr.HTML("<p>🔶: Evaluated only on the first part of each AoC day</p>")
current_results.json CHANGED
@@ -12,8 +12,8 @@
12
  "Model": "Mixtral-8x7B-Instruct-v0.1",
13
  "euler_original": 2.86,
14
  "euler_story": 2.23,
15
- "aoc_original": 8.67,
16
- "aoc_leet": 8.42,
17
  "instruction_only": false,
18
  "link": "https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1"
19
  },
@@ -49,5 +49,13 @@
49
  "aoc_leet": 29.85,
50
  "aoc_original": 50.0,
51
  "instruction_only": false
 
 
 
 
 
 
 
 
52
  }
53
  ]
 
12
  "Model": "Mixtral-8x7B-Instruct-v0.1",
13
  "euler_original": 2.86,
14
  "euler_story": 2.23,
15
+ "aoc_original": 15.31,
16
+ "aoc_leet": 13.01,
17
  "instruction_only": false,
18
  "link": "https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1"
19
  },
 
49
  "aoc_leet": 29.85,
50
  "aoc_original": 50.0,
51
  "instruction_only": false
52
+ },
53
+ {
54
+ "Model": "claude-haiku",
55
+ "euler_original": 7.07,
56
+ "euler_story": 6.08,
57
+ "aoc_original": 51.28,
58
+ "aoc_leet": 46.26,
59
+ "instruction_only": false
60
  }
61
  ]
text.py CHANGED
@@ -15,12 +15,13 @@ The model receives the task itself.
15
  CITATION_BUTTON_LABEL = "Copy the following snippet to cite our paper"
16
 
17
  CITATION_BUTTON_TEXT = r"""
18
- @misc{pecc,
19
- author = {Patrick Haller and Jonas Golde and Alan Akbik},
20
- title = {PECC - Problem Extraction and Coding Challenges},
21
- year = {2024},
22
- publisher = {GitHub},
23
- journal = {GitHub repository},
24
- howpublished = {}
25
- }
26
  """
 
 
 
 
 
 
 
 
 
15
  CITATION_BUTTON_LABEL = "Copy the following snippet to cite our paper"
16
 
17
  CITATION_BUTTON_TEXT = r"""
18
+ Coming soon!
 
 
 
 
 
 
 
19
  """
20
+ # @misc{pecc,
21
+ # author = {Patrick Haller and Jonas Golde and Alan Akbik},
22
+ # title = {PECC - Problem Extraction and Coding Challenges},
23
+ # year = {2024},
24
+ # publisher = {GitHub},
25
+ # journal = {GitHub repository},
26
+ # howpublished = {}
27
+ # }