Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
|
3 |
+
st.set_page_config(page_title="HAERAE Open Research Questions", layout="wide")
|
4 |
+
|
5 |
+
st.title("HAERAE Open Research Questions")
|
6 |
+
|
7 |
+
st.write("""
|
8 |
+
HAERAE is a non-profit research lab focused on the interpretability and evaluation of Korean language models.
|
9 |
+
Our mission is to advance the field with insightful benchmarks and tools. Below is an overview of our projects.
|
10 |
+
|
11 |
+
We've been doing most of our projects internally, but for those that have been unsolvable,
|
12 |
+
we are planning to open them to get help from the open-source community.
|
13 |
+
""")
|
14 |
+
|
15 |
+
st.header("HAERAE-Math Challenge")
|
16 |
+
|
17 |
+
st.write("""
|
18 |
+
Today we are introducing our first challenge: HAERAE-Math. We've created high-quality instructions on math
|
19 |
+
but don't have an idea on how to generate high-quality answers for them. We are looking for solutions that
|
20 |
+
use open-source models with openly available licenses.
|
21 |
+
|
22 |
+
We have created a total of 20,000 instructions already and are generating more. We've opened up a preview
|
23 |
+
of 50 of them in this link: [HAERAE-Math Samples](https://huggingface.co/datasets/HAERAE-HUB/HAERAE-Math-samples)
|
24 |
+
|
25 |
+
For those who generate answers for the 50 and share the methodology/results with us, we'll share the
|
26 |
+
remaining instructions and credit for the resulting dataset.
|
27 |
+
""")
|
28 |
+
|
29 |
+
st.subheader("Example Question")
|
30 |
+
|
31 |
+
example_question = """
|
32 |
+
νκ΅μ 보μ μ λ¬Έκ°κ° κ³ λνλ λ°μ΄ν° λ³΄νΈ μμ€ν
μ κ°λ°νκ³ μμ΅λλ€. μ΄ μμ€ν
μ 3μ°¨μ κΈ°ννμ μ κΈ λ©μ»€λμ¦μ μ¬μ©νλλ°, μ κΈ μ₯μΉλ μλΏ λͺ¨μμΌλ‘ λμ΄ μκ³ , λ°λ©΄μ λ°μ§λ¦μ 6cm, λμ΄λ 8cmμ
λλ€. μ΄ μλΏ λͺ¨μμ μ κΈ μ₯μΉμλ μν΅ λͺ¨μμ μ΄μ κ° λ± λ§κ² λ€μ΄κ°κ² μ€κ³λμ΄ μμ΅λλ€.
|
33 |
+
|
34 |
+
보μ μ λ¬Έκ°λ λ λμ μμ€μ 보μμ μν΄ μν΅ λͺ¨μμ μ΄μ μμ ꡬ λͺ¨μμ μ κΈ μ₯μΉλ₯Ό μΆκ°νλ €κ³ ν©λλ€. μ΄ κ΅¬λ μν΅ μμ λ± λ€μ΄κ°λλ‘ μ€κ³λμ΄ μμ΅λλ€.
|
35 |
+
|
36 |
+
λ€μμ μ§λ¬Έλ€μ ν΄κ²°νμκΈ° λ°λλλ€:
|
37 |
+
|
38 |
+
1. μλΏ μμ λ± λ€μ΄κ°κ² μ€κ³λ μν΅μ λ°μ§λ¦μ μΌλ§μΈκ°μ?
|
39 |
+
2. μν΅ μμ λ± λ€μ΄κ°κ² μ€κ³λ ꡬμ λΆνΌλ μΌλ§μΈκ°μ?
|
40 |
+
3. μλΏ, μν΅, κ΅¬κ° λͺ¨λ κ°μ μ€μ¬μΆμ 곡μ νκ³ μμΌλ©° μλΏμ κΌλκΈ°μ κ³Ό μν΅, ꡬμ μ€μ¬μ μ΄ λμΌνλ€κ³ κ°μ νλ©΄, μλΏμμ μν΅μ΄ μ°¨μ§νλ λΉμ¨μ ꡬνμμ€.
|
41 |
+
4. μ΄μ μλΏμ λμ΄λ₯Ό 2λ°°λ‘ λ리μ. μλΏμ λμ΄κ° 16cmκ° λμμ λ, μν΅κ³Ό ꡬμ ν¬κΈ°μ λΆνΌλ μ΄λ»κ² λ³νλμ?
|
42 |
+
5. μλΏμ λμ΄μ λ°λ©΄μ λ°μ§λ¦μ κ°κ° hμ rμ΄λΌκ³ ν λ, μν΅κ³Ό ꡬμ μ΅λ λΆνΌλ₯Ό rκ³Ό hλ‘ νννμμ€.
|
43 |
+
|
44 |
+
μλΏ, μν΅, ꡬμ λΆνΌ 곡μμ μ¬μ©νμ¬ λ¬Έμ λ₯Ό ν΄κ²°νμκΈ° λ°λλλ€:
|
45 |
+
|
46 |
+
μλΏμ λΆνΌ: V = 1/3ΟrΒ²h
|
47 |
+
μν΅μ λΆνΌ: V = ΟrΒ²h
|
48 |
+
ꡬμ λΆνΌ: V = 4/3ΟrΒ³
|
49 |
+
"""
|
50 |
+
|
51 |
+
st.code(example_question, language="markdown")
|
52 |
+
|
53 |
+
st.header("How to Participate")
|
54 |
+
|
55 |
+
st.write("""
|
56 |
+
1. Access the 50 sample questions from the provided Hugging Face dataset link.
|
57 |
+
2. Generate high-quality answers for these questions using open-source models.
|
58 |
+
3. Document your methodology and results.
|
59 |
+
4. Share your findings with us through [contact information or submission form].
|
60 |
+
5. If your approach is promising, we'll provide access to the full dataset of 20,000 instructions.
|
61 |
+
6. Collaborate with us to refine and improve the answer generation process.
|
62 |
+
7. Receive credit as a contributor to the final HAERAE-Math dataset.
|
63 |
+
""")
|
64 |
+
|
65 |
+
st.header("Why Participate?")
|
66 |
+
|
67 |
+
st.write("""
|
68 |
+
- Contribute to advancing Korean language model research
|
69 |
+
- Gain access to a large, high-quality dataset of math instructions
|
70 |
+
- Collaborate with HAERAE researchers
|
71 |
+
- Receive recognition in the field of NLP and math education
|
72 |
+
- Potential for co-authorship on related publications
|
73 |
+
""")
|
74 |
+
|
75 |
+
st.header("Contact Us")
|
76 |
+
|
77 |
+
st.write("""
|
78 |
+
For more information or to submit your results, please contact us at:
|
79 |
+
[Your contact information or a link to a submission form]
|
80 |
+
""")
|
81 |
+
|
82 |
+
st.sidebar.title("About HAERAE")
|
83 |
+
st.sidebar.info("""
|
84 |
+
HAERAE is a non-profit research lab dedicated to advancing the field of
|
85 |
+
Korean language model interpretability and evaluation. Our work focuses on
|
86 |
+
creating insightful benchmarks and tools to push the boundaries of NLP research.
|
87 |
+
""")
|