Skip to content

Commit

Permalink
Update monitor_md.py
Browse files Browse the repository at this point in the history
  • Loading branch information
infwinston authored Nov 20, 2024
1 parent e59fd55 commit 4a6024c
Showing 1 changed file with 7 additions and 5 deletions.
12 changes: 7 additions & 5 deletions fastchat/serve/monitor/monitor_md.py
Original file line number Diff line number Diff line change
Expand Up @@ -19,11 +19,12 @@
"math": "Math",
"if": "Instruction Following",
"multiturn": "Multi-Turn",
"creative_writing": "Creative Writing",
"coding": "Coding",
"coding_style_control": "Coding w/ Style Control",
"hard_6": "Hard Prompts (Overall)",
"hard_6": "Hard Prompts",
"hard_english_6": "Hard Prompts (English)",
"hard_6_style_control": "Hard Prompts (Overall) w/ Style Control",
"hard_6_style_control": "Hard Prompts w/ Style Control",
"long_user": "Longer Query",
"english": "English",
"chinese": "Chinese",
Expand All @@ -48,8 +49,8 @@
"Multi-Turn": "Multi-Turn Conversation (>= 2 turns)",
"Coding": "Coding: whether conversation contains code snippets",
"Coding w/ Style Control": "Coding with Style Control",
"Hard Prompts (Overall)": "Hard Prompts (Overall): details in [blog post](https://lmsys.org/blog/2024-05-17-category-hard/)",
"Hard Prompts (Overall) w/ Style Control": "Hard Prompts with Style Control. See details in [blog post](https://lmsys.org/blog/2024-08-28-style-control/).",
"Hard Prompts": "Hard Prompts: details in [blog post](https://lmsys.org/blog/2024-05-17-category-hard/)",
"Hard Prompts w/ Style Control": "Hard Prompts with Style Control. See details in [blog post](https://lmsys.org/blog/2024-08-28-style-control/).",
"Hard Prompts (English)": "Hard Prompts (English), note: the delta is to English Category. details in [blog post](https://lmsys.org/blog/2024-05-17-category-hard/)",
"Longer Query": "Longer Query (>= 500 tokens)",
"English": "English Prompts",
Expand All @@ -65,6 +66,7 @@
"Exclude Refusal": 'Exclude model responses with refusal (e.g., "I cannot answer")',
"overall_limit_5_user_vote": "overall_limit_5_user_vote",
"Overall (Deprecated)": "Overall without De-duplicating Top Redundant Queries (top 0.1%). See details in [blog post](https://lmsys.org/blog/2024-05-17-category-hard/#note-enhancing-quality-through-de-duplication).",
"Creative Writing": "Creative Writing",
}
cat_name_to_baseline = {
"Hard Prompts (English)": "English",
Expand All @@ -82,7 +84,7 @@ def make_default_md_1(mirror=False):
link_color = "#1976D2" # This color should be clear in both light and dark mode
leaderboard_md = f"""
# 🏆 Chatbot Arena LLM Leaderboard: Community-driven Evaluation for Best LLM and AI chatbots
[Blog](https://blog.lmarena.ai/blog/2023/arena/) | [GitHub](https://github.com/lm-sys/FastChat) | [Paper](https://arxiv.org/abs/2403.04132) | [Dataset](https://github.com/lm-sys/FastChat/blob/main/docs/dataset_release.md) | [Twitter](https://twitter.com/lmsysorg) | [Discord](https://discord.gg/6GXcFg3TH8) | [Kaggle Competition](https://www.kaggle.com/competitions/lmsys-chatbot-arena)
[Twitter](https://twitter.com/lmarena_ai) | [Discord](https://discord.gg/6GXcFg3TH8) | [Blog](https://blog.lmarena.ai/) | [GitHub](https://github.com/lm-sys/FastChat) | [Paper](https://arxiv.org/abs/2403.04132) | [Dataset](https://github.com/lm-sys/FastChat/blob/main/docs/dataset_release.md) | [Kaggle Competition](https://www.kaggle.com/competitions/wsdm-cup-multilingual-chatbot-arena)
"""

return leaderboard_md
Expand Down

0 comments on commit 4a6024c

Please sign in to comment.