edbeeching commited on
Commit
07bfeca
1 Parent(s): a460f7a

refining for release

Browse files
Files changed (2) hide show
  1. app.py +25 -24
  2. utils.py +13 -0
app.py CHANGED
@@ -63,28 +63,29 @@ def get_leaderboard():
63
 
64
  all_data = get_eval_results_dicts(IS_PUBLIC)
65
 
66
- gpt4_values = {
67
- "Model":f'<a target="_blank" href=https://arxiv.org/abs/2303.08774 style="color: blue; text-decoration: underline;text-decoration-style: dotted;">gpt4</a>',
68
- "Revision":"tech report",
69
- "8bit":None,
70
- "Average ⬆️":84.3,
71
- "ARC (25-shot) ⬆️":96.3,
72
- "HellaSwag (10-shot) ⬆️":95.3,
73
- "MMLU (5-shot) ⬆️":86.4,
74
- "TruthQA (0-shot) ⬆️":59.0,
75
- }
76
- all_data.append(gpt4_values)
77
- gpt35_values = {
78
- "Model":f'<a target="_blank" href=https://arxiv.org/abs/2303.08774 style="color: blue; text-decoration: underline;text-decoration-style: dotted;">gpt3.5</a>',
79
- "Revision":"tech report",
80
- "8bit":None,
81
- "Average ⬆️":71.9,
82
- "ARC (25-shot) ⬆️":85.2,
83
- "HellaSwag (10-shot) ⬆️":85.5,
84
- "MMLU (5-shot) ⬆️":70.0,
85
- "TruthQA (0-shot) ⬆️":47.0,
86
- }
87
- all_data.append(gpt35_values)
 
88
 
89
  dataframe = pd.DataFrame.from_records(all_data)
90
  dataframe = dataframe.sort_values(by=['Average ⬆️'], ascending=False)
@@ -200,7 +201,7 @@ block = gr.Blocks()
200
  with block:
201
  with gr.Row():
202
  gr.Markdown(f"""
203
- # 🤗 Open Chatbot Leaderboard
204
  <font size="4">With the plethora of chatbot LLMs being released week upon week, often with grandiose claims of their performance, it can be hard to filter out the genuine progress that is being made by the open-source community and which chatbot is the current state of the art. The 🤗 Open Chatbot Leaderboard aims to track, rank and evaluate chatbot models as they are released. We evaluate models of 4 key benchmarks from the <a href="https://github.com/EleutherAI/lm-evaluation-harness" target="_blank"> Eleuther AI Language Model Evaluation Harness </a>, a unified framework to test generative language models on a large number of different evaluation tasks. A key advantage of this leaderboard is that anyone from the community can submit a model for automated evaluation on the 🤗 research cluster. As long as it is Transformers model with weights on the 🤗 hub. We also support delta-weights for non-commercial licensed models, such as llama.
205
 
206
  Evaluation is performed against 4 popular benchmarks:
@@ -220,7 +221,7 @@ We chose these benchmarks as they test a variety of reasoning and general knowle
220
 
221
  with gr.Row():
222
  gr.Markdown(f"""
223
- # Evaluation Queue for the LMEH benchmarks, these models will be automatically evaluated on the 🤗 cluster
224
 
225
  """)
226
  with gr.Accordion("Evaluation Queue", open=False):
 
63
 
64
  all_data = get_eval_results_dicts(IS_PUBLIC)
65
 
66
+ if not IS_PUBLIC:
67
+ gpt4_values = {
68
+ "Model":f'<a target="_blank" href=https://arxiv.org/abs/2303.08774 style="color: blue; text-decoration: underline;text-decoration-style: dotted;">gpt4</a>',
69
+ "Revision":"tech report",
70
+ "8bit":None,
71
+ "Average ⬆️":84.3,
72
+ "ARC (25-shot) ⬆️":96.3,
73
+ "HellaSwag (10-shot) ⬆️":95.3,
74
+ "MMLU (5-shot) ⬆️":86.4,
75
+ "TruthQA (0-shot) ⬆️":59.0,
76
+ }
77
+ all_data.append(gpt4_values)
78
+ gpt35_values = {
79
+ "Model":f'<a target="_blank" href=https://arxiv.org/abs/2303.08774 style="color: blue; text-decoration: underline;text-decoration-style: dotted;">gpt3.5</a>',
80
+ "Revision":"tech report",
81
+ "8bit":None,
82
+ "Average ⬆️":71.9,
83
+ "ARC (25-shot) ⬆️":85.2,
84
+ "HellaSwag (10-shot) ⬆️":85.5,
85
+ "MMLU (5-shot) ⬆️":70.0,
86
+ "TruthQA (0-shot) ⬆️":47.0,
87
+ }
88
+ all_data.append(gpt35_values)
89
 
90
  dataframe = pd.DataFrame.from_records(all_data)
91
  dataframe = dataframe.sort_values(by=['Average ⬆️'], ascending=False)
 
201
  with block:
202
  with gr.Row():
203
  gr.Markdown(f"""
204
+ # 🤗 Open LLM Leaderboard
205
  <font size="4">With the plethora of chatbot LLMs being released week upon week, often with grandiose claims of their performance, it can be hard to filter out the genuine progress that is being made by the open-source community and which chatbot is the current state of the art. The 🤗 Open Chatbot Leaderboard aims to track, rank and evaluate chatbot models as they are released. We evaluate models of 4 key benchmarks from the <a href="https://github.com/EleutherAI/lm-evaluation-harness" target="_blank"> Eleuther AI Language Model Evaluation Harness </a>, a unified framework to test generative language models on a large number of different evaluation tasks. A key advantage of this leaderboard is that anyone from the community can submit a model for automated evaluation on the 🤗 research cluster. As long as it is Transformers model with weights on the 🤗 hub. We also support delta-weights for non-commercial licensed models, such as llama.
206
 
207
  Evaluation is performed against 4 popular benchmarks:
 
221
 
222
  with gr.Row():
223
  gr.Markdown(f"""
224
+ # Evaluation Queue for the 🤗 Open LLM Leaderboard, these models will be automatically evaluated on the 🤗 cluster
225
 
226
  """)
227
  with gr.Accordion("Evaluation Queue", open=False):
utils.py CHANGED
@@ -24,6 +24,19 @@ BENCH_TO_NAME = {
24
  "truthfulqa_mc":"TruthQA (0-shot) ⬆️",
25
  }
26
  def make_clickable_model(model_name):
 
 
 
 
 
 
 
 
 
 
 
 
 
27
  # remove user from model name
28
  #model_name_show = ' '.join(model_name.split('/')[1:])
29
 
 
24
  "truthfulqa_mc":"TruthQA (0-shot) ⬆️",
25
  }
26
  def make_clickable_model(model_name):
27
+ LLAMAS = ["huggingface/llama-7b", "huggingface/llama-13b", "huggingface/llama-30b", "huggingface/llama-65b"]
28
+ if model_name in LLAMAS:
29
+ model = model_name.split("/")[1]
30
+ return f'<a target="_blank" href="https://ai.facebook.com/blog/large-language-model-llama-meta-ai/" style="color: blue; text-decoration: underline;text-decoration-style: dotted;">{model}</a>'
31
+
32
+ if model_name == "HuggingFaceH4/stable-vicuna-13b-2904":
33
+ link = "https://huggingface.co/" + "CarperAI/stable-vicuna-13b-delta"
34
+ return f'<a target="_blank" href="{link}" style="color: blue; text-decoration: underline;text-decoration-style: dotted;">stable-vicuna-13b</a>'
35
+
36
+ if model_name == "HuggingFaceH4/llama-7b-ift-alpaca":
37
+ link = "https://crfm.stanford.edu/2023/03/13/alpaca.html"
38
+ return f'<a target="_blank" href="{link}" style="color: blue; text-decoration: underline;text-decoration-style: dotted;">alpaca-13b</a>'
39
+
40
  # remove user from model name
41
  #model_name_show = ' '.join(model_name.split('/')[1:])
42