Spaces:
Running
on
Zero
Running
on
Zero
initial commit
Browse files- README.md +3 -3
- app.py +194 -0
- requirements.txt +3 -0
README.md
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
---
|
2 |
title: Moonshot Math
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
sdk_version: 5.36.2
|
8 |
app_file: app.py
|
|
|
1 |
---
|
2 |
title: Moonshot Math
|
3 |
+
emoji: ๐๐๐จ๐ปโ๐ฌ
|
4 |
+
colorFrom: red
|
5 |
+
colorTo: purple
|
6 |
sdk: gradio
|
7 |
sdk_version: 5.36.2
|
8 |
app_file: app.py
|
app.py
ADDED
@@ -0,0 +1,194 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import spaces
|
2 |
+
import re
|
3 |
+
import gradio as gr
|
4 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
|
5 |
+
import torch
|
6 |
+
import json
|
7 |
+
|
8 |
+
LEAN4_DEFAULT_HEADER = (
|
9 |
+
"import Mathlib\n"
|
10 |
+
"import Aesop\n\n"
|
11 |
+
"set_option maxHeartbeats 0\n\n"
|
12 |
+
"open BigOperators Real Nat Topology Rat\n"
|
13 |
+
)
|
14 |
+
|
15 |
+
title = "# ๐๐ปโโ๏ธWelcome to ๐Tonic's ๐๐๐จ๐ปโ๐ฌMoonshot Math"
|
16 |
+
|
17 |
+
description = """
|
18 |
+
**Kimina-Prover-72B** is a state-of-the-art large formal reasoning model for Lean 4, achieving **80%+ pass rate** on the miniF2F benchmark, outperforming all prior works.\
|
19 |
+
Trained with Reinforcement Learning, 72B parameters, and a 32K token context window.\
|
20 |
+
- [Kimina-Prover-Preview GitHub](https://github.com/MoonshotAI/Kimina-Prover-Preview)\
|
21 |
+
- [Hugging Face: AI-MO/Kimina-Prover-72B](https://huggingface.co/AI-MO/Kimina-Prover-72B)\
|
22 |
+
- [Kimina Prover blog](https://huggingface.co/blog/AI-MO/kimina-prover)\
|
23 |
+
- [unimath dataset](https://huggingface.co/datasets/introspector/unimath)\
|
24 |
+
"""
|
25 |
+
|
26 |
+
citation = """> **Citation:**
|
27 |
+
> ```
|
28 |
+
> @article{kimina_prover_2025,
|
29 |
+
> title = {Kimina-Prover Preview: Towards Large Formal Reasoning Models with Reinforcement Learning},
|
30 |
+
> author = {Wang, Haiming and Unsal, Mert and ...},
|
31 |
+
> year = {2025},
|
32 |
+
> url = {http://arxiv.org/abs/2504.11354},
|
33 |
+
> }
|
34 |
+
> ```
|
35 |
+
"""
|
36 |
+
|
37 |
+
|
38 |
+
joinus ="""
|
39 |
+
### Join us:
|
40 |
+
๐TeamTonic๐ is always making cool demos! Join our active builder's ๐ ๏ธcommunity ๐ป
|
41 |
+
[](https://discord.gg/qdfnvSPcqP)
|
42 |
+
On ๐คHuggingface: [MultiTransformer](https://huggingface.co/MultiTransformer)
|
43 |
+
On ๐Github: [Tonic-AI](https://github.com/tonic-ai) & contribute to๐ [Build Tonic](https://git.tonic-ai.com/contribute)
|
44 |
+
๐คBig thanks to Yuvi Sharma and all the folks at huggingface for the community grant ๐ค
|
45 |
+
"""
|
46 |
+
|
47 |
+
# Build the initial system prompt
|
48 |
+
SYSTEM_PROMPT = "You are an expert in mathematics and Lean 4."
|
49 |
+
|
50 |
+
# Helper to build a Lean4 code block
|
51 |
+
def build_formal_block(formal_statement, informal_prefix=""):
|
52 |
+
return (
|
53 |
+
f"{LEAN4_DEFAULT_HEADER}\n"
|
54 |
+
f"{informal_prefix}\n"
|
55 |
+
f"{formal_statement}"
|
56 |
+
)
|
57 |
+
|
58 |
+
# Helper to extract the first Lean4 code block from text
|
59 |
+
def extract_lean4_code(text):
|
60 |
+
code_block = re.search(r"```lean4(.*?)(```|$)", text, re.DOTALL)
|
61 |
+
if code_block:
|
62 |
+
code = code_block.group(1)
|
63 |
+
lines = [line for line in code.split('\n') if line.strip()]
|
64 |
+
return '\n'.join(lines)
|
65 |
+
return text.strip()
|
66 |
+
|
67 |
+
# Example problems
|
68 |
+
unimath1 = """Goal:
|
69 |
+
X : UU
|
70 |
+
Y : UU
|
71 |
+
P : UU
|
72 |
+
xp : (X โ P) โ P
|
73 |
+
yp : (Y โ P) โ P
|
74 |
+
X0 : X ร Y โ P
|
75 |
+
x : X
|
76 |
+
============================
|
77 |
+
(Y โ P)"""
|
78 |
+
|
79 |
+
unimath2 = """Goal:
|
80 |
+
R : ring M : module R
|
81 |
+
============================
|
82 |
+
(islinear (idfun M))"""
|
83 |
+
|
84 |
+
unimath3 = """Goal:
|
85 |
+
X : UU i : nat b : hProptoType (i < S i) x : Vector X (S i) r : i = i
|
86 |
+
============================
|
87 |
+
(pr1 lastelement = pr1 (i,, b))"""
|
88 |
+
|
89 |
+
unimath4 = """Goal:
|
90 |
+
X : dcpo CX : continuous_dcpo_struct X x : pr1hSet X y : pr1hSet X
|
91 |
+
============================
|
92 |
+
(x โ y โ (โ i : approximating_family CX x, approximating_family CX x i โ y))"""
|
93 |
+
|
94 |
+
additional_info_prompt = "/-Explain using mathematics-/\n"
|
95 |
+
|
96 |
+
examples = [
|
97 |
+
[unimath1, additional_info_prompt, 2500],
|
98 |
+
[unimath2, additional_info_prompt, 2500],
|
99 |
+
[unimath3, additional_info_prompt, 2500],
|
100 |
+
[unimath4, additional_info_prompt, 2500]
|
101 |
+
]
|
102 |
+
|
103 |
+
model_name = "AI-MO/Kimina-Prover-72B"
|
104 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
105 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)
|
106 |
+
|
107 |
+
# Set generation config
|
108 |
+
model.generation_config = GenerationConfig.from_pretrained(model_name)
|
109 |
+
model.generation_config.pad_token_id = model.generation_config.eos_token_id
|
110 |
+
model.generation_config.do_sample = True
|
111 |
+
model.generation_config.temperature = 0.6
|
112 |
+
model.generation_config.top_p = 0.95
|
113 |
+
|
114 |
+
# Initialize chat history with system prompt
|
115 |
+
def init_chat(formal_statement, informal_prefix):
|
116 |
+
user_prompt = (
|
117 |
+
"Think about and solve the following problem step by step in Lean 4.\n"
|
118 |
+
"# Problem: Provide a formal proof for the following statement.\n"
|
119 |
+
f"# Formal statement:\n```lean4\n{build_formal_block(formal_statement, informal_prefix)}\n```\n"
|
120 |
+
)
|
121 |
+
return [
|
122 |
+
{"role": "system", "content": SYSTEM_PROMPT},
|
123 |
+
{"role": "user", "content": user_prompt}
|
124 |
+
]
|
125 |
+
|
126 |
+
# Gradio chat handler
|
127 |
+
@spaces.GPU
|
128 |
+
def chat_handler(user_message, informal_prefix, max_tokens, chat_history):
|
129 |
+
# If chat_history is empty, initialize with system and first user message
|
130 |
+
if not chat_history or len(chat_history) < 2:
|
131 |
+
chat_history = init_chat(user_message, informal_prefix)
|
132 |
+
display_history = [("user", user_message)]
|
133 |
+
else:
|
134 |
+
# Append new user message
|
135 |
+
chat_history.append({"role": "user", "content": user_message})
|
136 |
+
display_history = []
|
137 |
+
for msg in chat_history:
|
138 |
+
if msg["role"] == "user":
|
139 |
+
display_history.append(("user", msg["content"]))
|
140 |
+
elif msg["role"] == "assistant":
|
141 |
+
display_history.append(("assistant", msg["content"]))
|
142 |
+
# Format prompt using chat template
|
143 |
+
prompt = tokenizer.apply_chat_template(chat_history, tokenize=False, add_generation_prompt=True)
|
144 |
+
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(model.device)
|
145 |
+
attention_mask = torch.ones_like(input_ids)
|
146 |
+
outputs = model.generate(
|
147 |
+
input_ids,
|
148 |
+
attention_mask=attention_mask,
|
149 |
+
max_length=max_tokens + input_ids.shape[1],
|
150 |
+
pad_token_id=model.generation_config.pad_token_id,
|
151 |
+
temperature=model.generation_config.temperature,
|
152 |
+
top_p=model.generation_config.top_p,
|
153 |
+
)
|
154 |
+
result = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
155 |
+
# Extract only the new assistant message (after the prompt)
|
156 |
+
new_response = result[len(prompt):].strip()
|
157 |
+
# Add assistant message to chat history
|
158 |
+
chat_history.append({"role": "assistant", "content": new_response})
|
159 |
+
display_history.append(("assistant", new_response))
|
160 |
+
# Extract Lean4 code
|
161 |
+
code = extract_lean4_code(new_response)
|
162 |
+
# Prepare output
|
163 |
+
output_data = {
|
164 |
+
"model_input": prompt,
|
165 |
+
"model_output": result,
|
166 |
+
"lean4_code": code,
|
167 |
+
"chat_history": chat_history
|
168 |
+
}
|
169 |
+
return display_history, json.dumps(output_data, indent=2), code, chat_history
|
170 |
+
|
171 |
+
def main():
|
172 |
+
with gr.Blocks() as demo:
|
173 |
+
# Title and Model Description
|
174 |
+
gr.Markdown("""# ๐๐ปโโ๏ธWelcome to ๐Tonic's ๐๐๐จ๐ปโ๐ฌMoonshot Math""")
|
175 |
+
gr.Markdown(description)
|
176 |
+
gr.Markdown(joinus)
|
177 |
+
with gr.Row():
|
178 |
+
with gr.Column():
|
179 |
+
chat = gr.Chatbot(label="Chat History")
|
180 |
+
user_input = gr.Textbox(label="Your message or formal statement", lines=4)
|
181 |
+
informal = gr.Textbox(value=additional_info_prompt, label="Optional informal prefix")
|
182 |
+
max_tokens = gr.Slider(minimum=150, maximum=4096, value=2500, label="Max Tokens")
|
183 |
+
submit = gr.Button("Send")
|
184 |
+
with gr.Column():
|
185 |
+
json_out = gr.JSON(label="Full Output")
|
186 |
+
code_out = gr.Code(label="Extracted Lean4 Code", language="lean4")
|
187 |
+
state = gr.State([])
|
188 |
+
# On submit, call chat_handler
|
189 |
+
submit.click(chat_handler, [user_input, informal, max_tokens, state], [chat, json_out, code_out, state])
|
190 |
+
gr.Markdown(citation)
|
191 |
+
demo.launch()
|
192 |
+
|
193 |
+
if __name__ == "__main__":
|
194 |
+
main()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
transformers
|
3 |
+
accelerate
|