6

app.py

@@ -13,6 +13,7 @@ from src.about import (
     get_INTRODUCTION_TEXT,
     LLM_BENCHMARKS_TEXT,
     TITLE,
+    INTRODUCE_BENCHMARK
 )
 from src.display.css_html_js import custom_css
 from src.display.utils import (
@@ -93,6 +94,7 @@ with demo:
     gr.HTML(get_INTRODUCTION_TEXT(LEADERBOARD_DF.shape[0] , datetime.now(pytz.timezone('US/Pacific')).strftime("%Y-%m-%d %H:%M:%S")), elem_classes="markdown-text")
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
+        gr.HTML(INTRODUCE_BENCHMARK)  #TODO
         with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
             
             leaderboard = init_leaderboard(LEADERBOARD_DF)

src/about.py

@@ -44,6 +44,10 @@ def get_INTRODUCTION_TEXT(model_num: int, LAST_UPDATED: str, paper_link="TODO"):
     </div>
     """
 
+#TODO 
+INTRODUCE_BENCHMARK = f"""
+<h2 style="color: #2c3e50;"> Introduction </h2>
+"""
 
 #TODO About
 LLM_BENCHMARKS_TEXT = f"""

src/display/formatting.py

@@ -1,5 +1,5 @@
-def model_hyperlink(link, model_name):
-    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
+def model_hyperlink(link, model_show):
+    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_show}</a>'
 
 
 def make_clickable_model(model_show , model_name , is_open):

src/envs.py

@@ -13,6 +13,7 @@ REPO_ID = f"{OWNER}/CapArena_Auto_1"
 QUEUE_REPO = f"{OWNER}/requests"
 RESULTS_REPO = f"{OWNER}/results"
 
+QUEUE_REPO_TEST = f"{OWNER}/requests-test"
 # If you setup a cache later, just change HF_HOME
 CACHE_PATH=os.getenv("HF_HOME", ".")
 

src/submission/submit.py

@@ -3,7 +3,7 @@ import os
 from datetime import datetime, timezone
 
 from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
+from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO, QUEUE_REPO_TEST
 from src.submission.check_validity import (
     already_submitted_models,
     check_model_card,
@@ -109,7 +109,7 @@ def add_new_eval(
     API.upload_file(
         path_or_fileobj=out_path,
         path_in_repo=out_path.split("eval-queue/")[1],
-        repo_id=QUEUE_REPO,
+        repo_id=QUEUE_REPO_TEST,
         repo_type="dataset",
         commit_message=f"Add {model} to eval queue",
     )