Spaces:
Sleeping
Sleeping
added all features
Browse files- app.py +21 -12
- codeexecutor.py +22 -1
- temp.py +4 -0
app.py
CHANGED
|
@@ -2,7 +2,9 @@ import gradio as gr
|
|
| 2 |
import ctranslate2
|
| 3 |
from transformers import AutoTokenizer
|
| 4 |
from huggingface_hub import snapshot_download
|
| 5 |
-
from codeexecutor import get_majority_vote
|
|
|
|
|
|
|
| 6 |
import re
|
| 7 |
import os
|
| 8 |
# Define the model and tokenizer loading
|
|
@@ -41,7 +43,7 @@ def parse_prediction(prediction):
|
|
| 41 |
if answer is None:
|
| 42 |
# If no "Answer:" found, assume last line is the answer
|
| 43 |
answer = lines[-1].strip()
|
| 44 |
-
steps = lines
|
| 45 |
steps_text = '\n'.join(steps).strip()
|
| 46 |
return answer, steps_text
|
| 47 |
|
|
@@ -53,10 +55,16 @@ def majority_vote_with_steps(question, num_iterations=10):
|
|
| 53 |
|
| 54 |
for _ in range(num_iterations):
|
| 55 |
prediction = get_prediction(question)
|
| 56 |
-
answer,
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
# Get the majority voted answer
|
| 62 |
majority_voted_ans = get_majority_vote(all_answers)
|
|
@@ -65,13 +73,14 @@ def majority_vote_with_steps(question, num_iterations=10):
|
|
| 65 |
for i, ans in enumerate(all_answers):
|
| 66 |
if ans == majority_voted_ans:
|
| 67 |
steps_solution = steps_list[i]
|
|
|
|
| 68 |
break
|
| 69 |
else:
|
|
|
|
| 70 |
steps_solution = "No steps found"
|
| 71 |
|
| 72 |
-
return
|
| 73 |
|
| 74 |
-
# Gradio interface for user input and output
|
| 75 |
def gradio_interface(question, correct_answer):
|
| 76 |
final_answer, steps_solution = majority_vote_with_steps(question, iterations)
|
| 77 |
return question, final_answer, steps_solution, correct_answer
|
|
@@ -166,7 +175,7 @@ custom_css = """
|
|
| 166 |
"""
|
| 167 |
|
| 168 |
# Define the directory path
|
| 169 |
-
flagging_dir = "./flagged_data"
|
| 170 |
|
| 171 |
# Create the directory if it doesn't exist
|
| 172 |
if not os.path.exists(flagging_dir):
|
|
@@ -179,9 +188,9 @@ interface = gr.Interface(
|
|
| 179 |
gr.Textbox(label="π§ Math Question", placeholder="Enter your math question here...", elem_id="math_question"),
|
| 180 |
],
|
| 181 |
outputs=[
|
| 182 |
-
gr.Textbox(label="
|
| 183 |
-
gr.Textbox(label="
|
| 184 |
-
gr.Textbox(label="
|
| 185 |
],
|
| 186 |
title="π’ Math Question Solver",
|
| 187 |
description="Enter a math question to get the model's majority-voted answer and steps to solve the problem.",
|
|
|
|
| 2 |
import ctranslate2
|
| 3 |
from transformers import AutoTokenizer
|
| 4 |
from huggingface_hub import snapshot_download
|
| 5 |
+
from codeexecutor import get_majority_vote,type_check
|
| 6 |
+
import codeexecutor
|
| 7 |
+
|
| 8 |
import re
|
| 9 |
import os
|
| 10 |
# Define the model and tokenizer loading
|
|
|
|
| 43 |
if answer is None:
|
| 44 |
# If no "Answer:" found, assume last line is the answer
|
| 45 |
answer = lines[-1].strip()
|
| 46 |
+
steps = lines
|
| 47 |
steps_text = '\n'.join(steps).strip()
|
| 48 |
return answer, steps_text
|
| 49 |
|
|
|
|
| 55 |
|
| 56 |
for _ in range(num_iterations):
|
| 57 |
prediction = get_prediction(question)
|
| 58 |
+
answer,sucess= postprocess_completion(prediction, return_status=True, last_code_block=True)
|
| 59 |
+
if sucess:
|
| 60 |
+
all_predictions.append(prediction)
|
| 61 |
+
all_answers.append(answer)
|
| 62 |
+
steps_list.append(steps)
|
| 63 |
+
else:
|
| 64 |
+
answer, steps = parse_prediction(prediction)
|
| 65 |
+
all_predictions.append(prediction)
|
| 66 |
+
all_answers.append(answer)
|
| 67 |
+
steps_list.append(steps)
|
| 68 |
|
| 69 |
# Get the majority voted answer
|
| 70 |
majority_voted_ans = get_majority_vote(all_answers)
|
|
|
|
| 73 |
for i, ans in enumerate(all_answers):
|
| 74 |
if ans == majority_voted_ans:
|
| 75 |
steps_solution = steps_list[i]
|
| 76 |
+
answer=parse_prediction(steps_solution)
|
| 77 |
break
|
| 78 |
else:
|
| 79 |
+
answer=majority_voted_ans
|
| 80 |
steps_solution = "No steps found"
|
| 81 |
|
| 82 |
+
return answer, steps_solution
|
| 83 |
|
|
|
|
| 84 |
def gradio_interface(question, correct_answer):
|
| 85 |
final_answer, steps_solution = majority_vote_with_steps(question, iterations)
|
| 86 |
return question, final_answer, steps_solution, correct_answer
|
|
|
|
| 175 |
"""
|
| 176 |
|
| 177 |
# Define the directory path
|
| 178 |
+
flagging_dir = "./flagged_data"
|
| 179 |
|
| 180 |
# Create the directory if it doesn't exist
|
| 181 |
if not os.path.exists(flagging_dir):
|
|
|
|
| 188 |
gr.Textbox(label="π§ Math Question", placeholder="Enter your math question here...", elem_id="math_question"),
|
| 189 |
],
|
| 190 |
outputs=[
|
| 191 |
+
gr.Textbox(label="Question", interactive=False), # Non-editable
|
| 192 |
+
gr.Textbox(label="Answer", interactive=False), # Non-editable
|
| 193 |
+
gr.Textbox(label="Solution", interactive=True), # Editable textbox for correct solution
|
| 194 |
],
|
| 195 |
title="π’ Math Question Solver",
|
| 196 |
description="Enter a math question to get the model's majority-voted answer and steps to solve the problem.",
|
codeexecutor.py
CHANGED
|
@@ -97,7 +97,7 @@ def execute_completion(executor, completion, return_status, last_code_block):
|
|
| 97 |
success = successes[-1]
|
| 98 |
if return_status:
|
| 99 |
return output, success
|
| 100 |
-
return output
|
| 101 |
|
| 102 |
|
| 103 |
def postprocess_completion(text, return_status, last_code_block):
|
|
@@ -113,3 +113,24 @@ def get_majority_vote(answers):
|
|
| 113 |
c = Counter(answers)
|
| 114 |
value, _ = c.most_common()[0]
|
| 115 |
return value
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
success = successes[-1]
|
| 98 |
if return_status:
|
| 99 |
return output, success
|
| 100 |
+
return output ,False
|
| 101 |
|
| 102 |
|
| 103 |
def postprocess_completion(text, return_status, last_code_block):
|
|
|
|
| 113 |
c = Counter(answers)
|
| 114 |
value, _ = c.most_common()[0]
|
| 115 |
return value
|
| 116 |
+
|
| 117 |
+
|
| 118 |
+
def type_check(self,expr_str):
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
expr = sp.sympify(expr_str)
|
| 122 |
+
|
| 123 |
+
# Check if the expression is a real number
|
| 124 |
+
if expr.is_real:
|
| 125 |
+
return "Real"
|
| 126 |
+
|
| 127 |
+
# Check if the expression is a complex number
|
| 128 |
+
if expr.is_complex:
|
| 129 |
+
return "Complex"
|
| 130 |
+
|
| 131 |
+
# Check if the expression is a polynomial
|
| 132 |
+
if expr.is_polynomial():
|
| 133 |
+
return "Polynomial"
|
| 134 |
+
|
| 135 |
+
# Otherwise, classify as other
|
| 136 |
+
return "Other"
|
temp.py
CHANGED
|
@@ -3,6 +3,7 @@ import ctranslate2
|
|
| 3 |
from transformers import AutoTokenizer
|
| 4 |
from huggingface_hub import snapshot_download
|
| 5 |
from codeexecutor import get_majority_vote
|
|
|
|
| 6 |
import re
|
| 7 |
|
| 8 |
# Define the model and tokenizer loading
|
|
@@ -11,6 +12,9 @@ tokenizer = AutoTokenizer.from_pretrained("AI-MO/NuminaMath-7B-TIR")
|
|
| 11 |
model_path = snapshot_download(repo_id="Makima57/deepseek-math-Numina")
|
| 12 |
generator = ctranslate2.Generator(model_path, device="cpu", compute_type="int8")
|
| 13 |
iterations = 10
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
# Function to generate predictions using the model
|
| 16 |
def get_prediction(question):
|
|
|
|
| 3 |
from transformers import AutoTokenizer
|
| 4 |
from huggingface_hub import snapshot_download
|
| 5 |
from codeexecutor import get_majority_vote
|
| 6 |
+
import codeexecutor
|
| 7 |
import re
|
| 8 |
|
| 9 |
# Define the model and tokenizer loading
|
|
|
|
| 12 |
model_path = snapshot_download(repo_id="Makima57/deepseek-math-Numina")
|
| 13 |
generator = ctranslate2.Generator(model_path, device="cpu", compute_type="int8")
|
| 14 |
iterations = 10
|
| 15 |
+
executor = PythonREPL()
|
| 16 |
+
|
| 17 |
+
|
| 18 |
|
| 19 |
# Function to generate predictions using the model
|
| 20 |
def get_prediction(question):
|