djstrong commited on
Commit
d2d2329
1 Parent(s): 59f4209

n-shot filter

Browse files
src/about.py CHANGED
@@ -37,7 +37,7 @@ NUM_FEWSHOT = 0 # Change with your few shot
37
 
38
 
39
  # Your leaderboard name
40
- TITLE = """<h1 align="center" id="space-title">Open PL LLM Leaderboard (0-shot)</h1>"""
41
 
42
  # What does your leaderboard evaluate?
43
  INTRODUCTION_TEXT = """
 
37
 
38
 
39
  # Your leaderboard name
40
+ TITLE = """<h1 align="center" id="space-title">Open PL LLM Leaderboard (0-shot and 5-shot)</h1>"""
41
 
42
  # What does your leaderboard evaluate?
43
  INTRODUCTION_TEXT = """
src/display/utils.py CHANGED
@@ -26,7 +26,7 @@ auto_eval_column_dict = []
26
  # Init
27
  auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
28
  auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
29
- auto_eval_column_dict.append(["n_shot", ColumnContent, ColumnContent("n_shot", "number", True)])
30
  #Scores
31
  auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
32
  for task in Tasks:
@@ -96,6 +96,14 @@ class NShotType(Enum):
96
  n0 = ModelDetails("0")
97
  n5 = ModelDetails("5")
98
 
 
 
 
 
 
 
 
 
99
  class Precision(Enum):
100
  float16 = ModelDetails("float16")
101
  bfloat16 = ModelDetails("bfloat16")
 
26
  # Init
27
  auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
28
  auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
29
+ auto_eval_column_dict.append(["n_shot", ColumnContent, ColumnContent("n_shot", "str", True)])
30
  #Scores
31
  auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
32
  for task in Tasks:
 
96
  n0 = ModelDetails("0")
97
  n5 = ModelDetails("5")
98
 
99
+ @staticmethod
100
+ def from_str(n):
101
+ if n in ["0", 0]:
102
+ return NShotType.n0
103
+ if n in ["5", 5]:
104
+ return NShotType.n5
105
+ return NShotType.Unknown
106
+
107
  class Precision(Enum):
108
  float16 = ModelDetails("float16")
109
  bfloat16 = ModelDetails("bfloat16")
src/leaderboard/read_evals.py CHANGED
@@ -9,7 +9,7 @@ import dateutil
9
  import numpy as np
10
 
11
  from src.display.formatting import make_clickable_model
12
- from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
13
  from src.submission.check_validity import is_model_on_hub
14
 
15
  NUM_FEWSHOT = 0
@@ -31,7 +31,7 @@ class EvalResult:
31
  num_params: int = 0
32
  date: str = "" # submission date of request file
33
  still_on_hub: bool = False
34
- n_shot: int = 0
35
 
36
  @classmethod
37
  def init_from_json_file(self, json_filepath, n_shot_num):
@@ -92,7 +92,7 @@ class EvalResult:
92
  revision= config.get("model_sha", ""),
93
  still_on_hub=still_on_hub,
94
  architecture=architecture,
95
- n_shot=n_shot_num
96
  )
97
 
98
  def update_with_request_file(self, requests_path):
@@ -203,7 +203,7 @@ class EvalResult:
203
  print(f"Could not find still on hub")
204
 
205
  try:
206
- data_dict[AutoEvalColumn.n_shot.name] = self.n_shot
207
  except KeyError:
208
  print(f"Could not find still on hub")
209
 
 
9
  import numpy as np
10
 
11
  from src.display.formatting import make_clickable_model
12
+ from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType, NShotType
13
  from src.submission.check_validity import is_model_on_hub
14
 
15
  NUM_FEWSHOT = 0
 
31
  num_params: int = 0
32
  date: str = "" # submission date of request file
33
  still_on_hub: bool = False
34
+ n_shot: NShotType = NShotType.n0
35
 
36
  @classmethod
37
  def init_from_json_file(self, json_filepath, n_shot_num):
 
92
  revision= config.get("model_sha", ""),
93
  still_on_hub=still_on_hub,
94
  architecture=architecture,
95
+ n_shot=NShotType.from_str(n_shot_num)
96
  )
97
 
98
  def update_with_request_file(self, requests_path):
 
203
  print(f"Could not find still on hub")
204
 
205
  try:
206
+ data_dict[AutoEvalColumn.n_shot.name] = self.n_shot.value.name
207
  except KeyError:
208
  print(f"Could not find still on hub")
209