update text

app.py

@@ -16,21 +16,31 @@ from src.display.css_html_js import custom_css
 from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
 
 
-
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
+
 try:
     print(EVAL_REQUESTS_PATH)
     snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=QUEUE_REPO,
+        local_dir=EVAL_REQUESTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
 try:
     print(EVAL_RESULTS_PATH)
     snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=RESULTS_REPO,
+        local_dir=EVAL_RESULTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
@@ -42,10 +52,17 @@ with demo:
     gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        show_result_page(root_path='MC', title='📝 Exam', index=0)
-        show_result_page(root_path='LLM', title='🤖 LLM as Judge', index=1)
-        show_result_page(root_path='NLU', title='🕵️ NLU', index=2)
-        show_result_page(root_path='NLG', title='🖊️ NLG', index=3)
+        show_result_page(root_path="MC", title="📝 Exam", index=0)
+        show_result_page(
+            root_path="LLM",
+            title="🤖 LLM-as-a-Judge",
+            index=1,
+            extra=lambda: gr.Markdown(
+                "* Gemini result is only reported in MT-Bench because the other benchmark was mistakenly classified as unsafe"
+            ),
+        )
+        show_result_page(root_path="NLU", title="🕵️ NLU", index=2)
+        show_result_page(root_path="NLG", title="🖊️ NLG", index=3)
         show_about_page(index=4)
         show_submit_page(index=5)
 
@@ -58,10 +75,9 @@ with demo:
                 elem_id="citation-button",
                 show_copy_button=True,
             )
-        gr.Markdown('* Gemini result is only reported in MT-Bench because the other benchmark was mistakenly classified as unsafe')
-        
+
 
 scheduler = BackgroundScheduler()
 scheduler.add_job(restart_space, "interval", seconds=1800)
 scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+demo.queue(default_concurrency_limit=40).launch()

src/about.py

@@ -2,8 +2,7 @@ from dataclasses import dataclass
 from enum import Enum
 
 
-
-NUM_FEWSHOT = 0 # Change with your few shot
+NUM_FEWSHOT = 0  # Change with your few shot
 # ---------------------------------------------------
 
 TITLE = """<h1>🇹🇭 Thai LLM Leaderboard</h1>"""
@@ -14,7 +13,7 @@ TITLE = """<h1>🇹🇭 Thai LLM Leaderboard</h1>"""
 INTRODUCTION_TEXT = """
 The Thai-LLM Leaderboard 🇹🇭 focused on standardizing evaluation methods for large language models (LLMs) in the Thai language based on <a href="https://github.com/SEACrowd">SEACrowd</a>,
 As part of an open community project, we welcome you to submit new evaluation tasks or models. 
-This leaderboard is developed in collaboration with <a href="https://www.scb10x.com">SCB 10X</a>, <a href="https://www.vistec.ac.th/">Vistec</a>, and <a href="https://github.com/SEACrowd">SEACrowd</a>. 
+This leaderboard is developed in collaboration with <a href="https://www.scb10x.com">SCB 10X</a>, <a href="https://www.vistec.ac.th/">Vistec</a>, and <a href="https://github.com/SEACrowd">SEACrowd</a>. Read more on <a href="https://blog.opentyphoon.ai/introducing-the-thaillm-leaderboard-thaillm-evaluation-ecosystem-508e789d06bf">Introduction Blog</a>
 """
 
 LLM_BENCHMARKS_TEXT = f"""
@@ -23,8 +22,8 @@ The leaderboard currently consists of the following benchmarks:
 - Exam
   - <a href="https://huggingface.co/datasets/scb10x/thai_exam">ThaiExam</a>: ThaiExam is a Thai language benchmark based on examinations for high-school students and investment professionals in Thailand.
   - <a href="https://arxiv.org/abs/2306.05179">M3Exam</a>: M3Exam is a novel benchmark sourced from real and official human exam questions for evaluating LLMsin a multilingual, multimodal, and multilevel context. Here is Thai subset of M3Exam.
-- LLM as Judge
-  - Thai MT-Bench: <a href="https://arxiv.org/abs/2306.05685">MT-Bench</a> inspired varient of LLM as judges specifically developed by Vistec for Thai language and cultural.
+- LLM-as-a-Judge
+  - <a href="https://huggingface.co/datasets/ThaiLLM-Leaderboard/mt-bench-thai">Thai MT-Bench</a>: <a href="https://arxiv.org/abs/2306.05685">MT-Bench</a> inspired varient of LLM-as-a-Judge specifically developed by Vistec for Thai language and cultural.
 - NLU
   - <a href="https://huggingface.co/datasets/facebook/belebele">Belebele</a>: Belebele is a multiple-choice machine reading comprehension (MRC) dataset spanning 122 language variants. Here is Thai subset of Belebele.
   - <a href="https://huggingface.co/datasets/facebook/xnli">XNLI</a>: XNLI is an evaluation corpus for language transfer and cross-lingual sentence classification in 15 languages. Here is Thai subset of XNLI.
@@ -37,9 +36,10 @@ The leaderboard currently consists of the following benchmarks:
 
 
 Metrics Implementations
+- Multiple-choice accuracy is calculated using the <a href="https://github.com/SEACrowd/seacrowd-experiments/blob/048536fc0d4614734d479b298ea00a1f520da42b/evaluation/main_nlu_prompt_batch.py#L71">SEACrowd implementation</a> of logits comparison, similar to the method used by the <a href="https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard">Open LLM Leaderboard</a> (<a href="https://github.com/EleutherAI/lm-evaluation-harness">EleutherAI Harness</a>). <a href="https://huggingface.co/blog/open-llm-leaderboard-mmlu">explain</a>
 - BLEU is calculated using flores200 tokenizer using huggingface evaluate <a href="https://huggingface.co/spaces/evaluate-metric/sacrebleu">implementation</a>.
 - ROUGEL is calculated using pythainlp newmm tokenizer using huggingface evaluate <a href="https://huggingface.co/spaces/evaluate-metric/rouge">implementation</a>.
-- LLM as judge rating is judged by OpenAI gpt-4o-2024-05-13 using prompt specific by <a href="https://github.com/lm-sys/FastChat/blob/main/fastchat/llm_judge/data/judge_prompts.jsonl">lmsys MT-Bench</a>.
+- LLM-as-a-Judge rating is judged by OpenAI gpt-4o-2024-05-13 using prompt specific by <a href="https://github.com/lm-sys/FastChat/blob/main/fastchat/llm_judge/data/judge_prompts.jsonl">lmsys MT-Bench</a>.
 
 Reproducibility
 
@@ -51,7 +51,15 @@ We're grateful to community members for task and model submitting. To contribute
 """
 
 CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
-CITATION_BUTTON_TEXT = r"""@misc{lovenia2024seacrowdmultilingualmultimodaldata,
+CITATION_BUTTON_TEXT = r"""@misc{thaillm-leaderboard,
+  author = {SCB 10X and Vistec and SEACrowd},
+  title = {Thai LLM Leaderboard},
+  year = {2024},
+  publisher = {Hugging Face},
+  howpublished = "\url{https://huggingface.co/spaces/ThaiLLM-Leaderboard/leaderboard}",
+}
+
+@misc{lovenia2024seacrowdmultilingualmultimodaldata,
       title={SEACrowd: A Multilingual Multimodal Data Hub and Benchmark Suite for Southeast Asian Languages}, 
       author={Holy Lovenia and Rahmad Mahendra and Salsabil Maulana Akbar and Lester James V. Miranda and Jennifer Santoso and Elyanah Aco and Akhdan Fadhilah and Jonibek Mansurov and Joseph Marvin Imperial and Onno P. Kampman and Joel Ruben Antony Moniz and Muhammad Ravi Shulthan Habibi and Frederikus Hudi and Railey Montalan and Ryan Ignatius and Joanito Agili Lopo and William Nixon and Börje F. Karlsson and James Jaya and Ryandito Diandaru and Yuze Gao and Patrick Amadeus and Bin Wang and Jan Christian Blaise Cruz and Chenxi Whitehouse and Ivan Halim Parmonangan and Maria Khelli and Wenyu Zhang and Lucky Susanto and Reynard Adha Ryanda and Sonny Lazuardi Hermawan and Dan John Velasco and Muhammad Dehan Al Kautsar and Willy Fitra Hendria and Yasmin Moslem and Noah Flynn and Muhammad Farid Adilazuarda and Haochen Li and Johanes Lee and R. Damanhuri and Shuo Sun and Muhammad Reza Qorib and Amirbek Djanibekov and Wei Qi Leong and Quyet V. Do and Niklas Muennighoff and Tanrada Pansuwan and Ilham Firdausi Putra and Yan Xu and Ngee Chia Tai and Ayu Purwarianti and Sebastian Ruder and William Tjhi and Peerat Limkonchotiwat and Alham Fikri Aji and Sedrick Keh and Genta Indra Winata and Ruochen Zhang and Fajri Koto and Zheng-Xin Yong and Samuel Cahyawijaya},
       year={2024},
@@ -59,4 +67,4 @@ CITATION_BUTTON_TEXT = r"""@misc{lovenia2024seacrowdmultilingualmultimodaldata,
       archivePrefix={arXiv},
       primaryClass={cs.CL},
       url={https://arxiv.org/abs/2406.10118}, 
-}"""
+}"""

src/pages/result_table.py

@@ -1,3 +1,4 @@
+from typing import Callable
 import gradio as gr
 from gradio_leaderboard import Leaderboard, SearchColumns, SelectColumns, ColumnFilter
 from src.leaderboard.read_evals import FILTERED_ONLY_FIELD
@@ -8,11 +9,12 @@ from src.display.utils import (
     fields,
 )
 
-def show_result_page(root_path: str, title: str, index: int):
+
+def show_result_page(root_path: str, title: str, index: int, extra: Callable[[None], None] = None):
     raw_data, original_df = get_leaderboard_df(EVAL_RESULTS_PATH + f"/{root_path}")
     leaderboard_df = original_df.copy()
     with gr.TabItem(title, elem_id="llm-benchmark-tab-table", id=index):
-        return Leaderboard(
+        Leaderboard(
             value=leaderboard_df,
             datatype=[c.type for c in fields(AutoEvalColumn)],
             select_columns=SelectColumns(
@@ -29,7 +31,7 @@ def show_result_page(root_path: str, title: str, index: int):
             ),
             filter_columns=[
                 ColumnFilter(
-                    'params',
+                    "params",
                     type="slider",
                     min=0.01,
                     max=150,
@@ -39,3 +41,7 @@ def show_result_page(root_path: str, title: str, index: int):
             bool_checkboxgroup_label="Hide models",
             interactive=False,
         )
+        if extra is not None:
+            with gr.Column():
+                extra()
+