eduagarcia commited on
Commit
f49e1e5
2 Parent(s): a3b0a0f 77a5f61

Merge branch 'main' of https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard

Browse files
Files changed (3) hide show
  1. app.py +2 -1
  2. proprietary_models_results.json +21 -0
  3. src/display/about.py +16 -1
app.py CHANGED
@@ -62,7 +62,8 @@ from src.tools.plots import (
62
  def restart_space():
63
  print("Running Restart")
64
  try:
65
- API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
 
66
  except:
67
  print("Restart failed")
68
 
 
62
  def restart_space():
63
  print("Running Restart")
64
  try:
65
+ #API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
66
+ pass
67
  except:
68
  print("Restart failed")
69
 
proprietary_models_results.json CHANGED
@@ -103,5 +103,26 @@
103
  },
104
  "result_metrics_average": 0.6993914868794414,
105
  "result_metrics_npm": 0.551208000273598
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106
  }
107
  ]
 
103
  },
104
  "result_metrics_average": 0.6993914868794414,
105
  "result_metrics_npm": 0.551208000273598
106
+ },
107
+ {
108
+ "model": "gemini-1.5-pro-preview-0409",
109
+ "name": "Gemini 1.5 Pro Preview (0409)",
110
+ "link": "https://cloud.google.com/vertex-ai",
111
+ "date": "2024-04-15",
112
+ "status": "full",
113
+ "main_language": "English",
114
+ "result_metrics": {
115
+ "enem_challenge": 0.8509447165850245,
116
+ "bluex": 0.7719054242002782,
117
+ "oab_exams": 0.6888382687927107,
118
+ "assin2_sts": 0.8159702278408203,
119
+ "assin2_rte": 0.9328989988467518,
120
+ "faquad_nli": 0.7290756302521009,
121
+ "hatebr_offensive": 0.8697698647467024,
122
+ "portuguese_hate_speech": 0.7539414414414414,
123
+ "tweetsentbr": 0.772785080895884
124
+ },
125
+ "result_metrics_average": 0.7984588504001905,
126
+ "result_metrics_npm": 0.6908188311933006
127
  }
128
  ]
src/display/about.py CHANGED
@@ -24,6 +24,11 @@ The leaderboard's backend runs on a [fork](https://github.com/eduagarcia/lm-eval
24
 
25
  {SUPPORT_DESCRIPTION}
26
  """
 
 
 
 
 
27
  task_count = 0
28
  TASKS_LIST= ""
29
  for task in Tasks:
@@ -104,7 +109,17 @@ To get more information about quantization, see:
104
 
105
  ## Useful links
106
  - [Community resources](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/174)
107
- - [Collection of best models](https://huggingface.co/collections/{PATH_TO_COLLECTION})
 
 
 
 
 
 
 
 
 
 
108
  """
109
 
110
  REMOTE_CODE_EXAPLANATION = f"- *Yes.*"
 
24
 
25
  {SUPPORT_DESCRIPTION}
26
  """
27
+
28
+ CITATION_TEXT = ""
29
+ if 'citation' in TASK_CONFIG['readme']:
30
+ CITATION_TEXT += TASK_CONFIG['readme']['citation']
31
+
32
  task_count = 0
33
  TASKS_LIST= ""
34
  for task in Tasks:
 
109
 
110
  ## Useful links
111
  - [Community resources](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/174)
112
+ - [Collection of best models](https://huggingface.co/collections/{PATH_TO_COLLECTION})
113
+
114
+ """
115
+
116
+ if CITATION_TEXT:
117
+ ABOUT_DESCRIPTION_CONTEXT += f"""
118
+ ## Citation
119
+
120
+ ```bibtex
121
+ {CITATION_TEXT}
122
+ ```
123
  """
124
 
125
  REMOTE_CODE_EXAPLANATION = f"- *Yes.*"