Commit ·
67eefaa
1
Parent(s): 51e1a5a
small
Browse files
app.py
CHANGED
|
@@ -66,6 +66,31 @@ SHEET_COLUMNS = [
|
|
| 66 |
]
|
| 67 |
SHEET_HEADERS = [label for _, label in SHEET_COLUMNS]
|
| 68 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
def get_model_name():
|
| 71 |
return os.environ.get("OPENAI_MODEL_NAME", DEFAULT_MODEL_NAME)
|
|
@@ -1812,12 +1837,7 @@ def build_practice_scoring_note(submission_result):
|
|
| 1812 |
total_cases = submission_result["total_cases"]
|
| 1813 |
return (
|
| 1814 |
f"This practice score is based on {total_cases} hidden calibration cases. "
|
| 1815 |
-
"Each case is run 3 times to check prompt consistency.
|
| 1816 |
-
"LLM 1 can earn up to 1 point for correct routing and target-firm "
|
| 1817 |
-
"normalization, and LLM 3 can earn up to 1 point for a correct final "
|
| 1818 |
-
"JSON answer with supported citations. Step 2 is not scored directly, "
|
| 1819 |
-
"but it affects the LLM 3 score because LLM 3 relies on the snippet-level "
|
| 1820 |
-
"analysis. The final submission uses a separate hidden holdout set."
|
| 1821 |
)
|
| 1822 |
|
| 1823 |
|
|
@@ -2023,7 +2043,7 @@ def handle_submission(submission_type, email, name, s1, s2, s3):
|
|
| 2023 |
|
| 2024 |
|
| 2025 |
def build_interface():
|
| 2026 |
-
with gr.Blocks() as demo:
|
| 2027 |
gr.Markdown(
|
| 2028 |
"""
|
| 2029 |
# Applicant Task: Target Company & Law Firm Identification
|
|
@@ -2170,7 +2190,7 @@ def build_interface():
|
|
| 2170 |
|
| 2171 |
gr.Markdown(
|
| 2172 |
"""
|
| 2173 |
-
<div
|
| 2174 |
<b>Please note:</b><br>
|
| 2175 |
Each run may take a couple of minutes.<br>
|
| 2176 |
After you click a button, wait for the result and do not click it again.
|
|
|
|
| 66 |
]
|
| 67 |
SHEET_HEADERS = [label for _, label in SHEET_COLUMNS]
|
| 68 |
|
| 69 |
+
APP_CSS = """
|
| 70 |
+
.submission-note {
|
| 71 |
+
background: var(--submission-note-bg, #fff7e6);
|
| 72 |
+
border: 1px solid var(--submission-note-border, #ffe5b4);
|
| 73 |
+
border-radius: 8px;
|
| 74 |
+
color: var(--submission-note-text, #3d2b00);
|
| 75 |
+
margin-bottom: 1em;
|
| 76 |
+
padding: 16px;
|
| 77 |
+
}
|
| 78 |
+
|
| 79 |
+
@media (prefers-color-scheme: dark) {
|
| 80 |
+
.submission-note {
|
| 81 |
+
--submission-note-bg: #2b2111;
|
| 82 |
+
--submission-note-border: #6b4b18;
|
| 83 |
+
--submission-note-text: #f7e8c5;
|
| 84 |
+
}
|
| 85 |
+
}
|
| 86 |
+
|
| 87 |
+
.dark .submission-note {
|
| 88 |
+
--submission-note-bg: #2b2111;
|
| 89 |
+
--submission-note-border: #6b4b18;
|
| 90 |
+
--submission-note-text: #f7e8c5;
|
| 91 |
+
}
|
| 92 |
+
"""
|
| 93 |
+
|
| 94 |
|
| 95 |
def get_model_name():
|
| 96 |
return os.environ.get("OPENAI_MODEL_NAME", DEFAULT_MODEL_NAME)
|
|
|
|
| 1837 |
total_cases = submission_result["total_cases"]
|
| 1838 |
return (
|
| 1839 |
f"This practice score is based on {total_cases} hidden calibration cases. "
|
| 1840 |
+
"Each case is run 3 times to check prompt consistency."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1841 |
)
|
| 1842 |
|
| 1843 |
|
|
|
|
| 2043 |
|
| 2044 |
|
| 2045 |
def build_interface():
|
| 2046 |
+
with gr.Blocks(css=APP_CSS) as demo:
|
| 2047 |
gr.Markdown(
|
| 2048 |
"""
|
| 2049 |
# Applicant Task: Target Company & Law Firm Identification
|
|
|
|
| 2190 |
|
| 2191 |
gr.Markdown(
|
| 2192 |
"""
|
| 2193 |
+
<div class="submission-note">
|
| 2194 |
<b>Please note:</b><br>
|
| 2195 |
Each run may take a couple of minutes.<br>
|
| 2196 |
After you click a button, wait for the result and do not click it again.
|