Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -3,7 +3,6 @@ import gradio as gr
|
|
3 |
import pandas as pd
|
4 |
import torch
|
5 |
import logging
|
6 |
-
import gc
|
7 |
from transformers import pipeline
|
8 |
|
9 |
# Setup logging
|
@@ -13,15 +12,9 @@ logging.basicConfig(
|
|
13 |
)
|
14 |
logger = logging.getLogger(__name__)
|
15 |
|
16 |
-
# Device configuration
|
17 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
18 |
logger.info(f"Using device: {DEVICE}")
|
19 |
|
20 |
-
def clear_gpu_memory():
|
21 |
-
if DEVICE == "cuda":
|
22 |
-
torch.cuda.empty_cache()
|
23 |
-
gc.collect()
|
24 |
-
|
25 |
class FinancialAnalyzer:
|
26 |
def __init__(self):
|
27 |
self.analysis_model = None
|
@@ -30,12 +23,14 @@ class FinancialAnalyzer:
|
|
30 |
|
31 |
def load_models(self):
|
32 |
try:
|
|
|
33 |
self.analysis_model = pipeline(
|
34 |
"text-generation",
|
35 |
model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
|
36 |
torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32
|
37 |
)
|
38 |
|
|
|
39 |
self.sentiment_model = pipeline(
|
40 |
"text-classification",
|
41 |
model="ProsusAI/finbert",
|
@@ -47,115 +42,105 @@ class FinancialAnalyzer:
|
|
47 |
logger.error(f"Error loading models: {str(e)}")
|
48 |
raise
|
49 |
|
50 |
-
def
|
|
|
51 |
try:
|
52 |
-
|
53 |
-
number = float(number.replace(',', '').replace('$', '').strip())
|
54 |
-
return f"${number:,.0f}"
|
55 |
-
except:
|
56 |
-
return str(number)
|
57 |
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
-
return
|
70 |
except Exception as e:
|
71 |
-
logger.error(f"Error
|
72 |
raise
|
73 |
|
74 |
-
def analyze_financials(self,
|
75 |
try:
|
76 |
-
#
|
77 |
-
|
78 |
-
|
79 |
|
80 |
-
|
81 |
-
|
82 |
|
83 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
84 |
sentiment = self.sentiment_model(
|
85 |
-
|
86 |
truncation=True
|
87 |
)[0]
|
88 |
|
89 |
-
# Generate analysis
|
90 |
-
|
91 |
|
92 |
-
{
|
93 |
|
94 |
Market Sentiment: {sentiment['label']} ({sentiment['score']:.2%})
|
95 |
|
96 |
-
Provide concise analysis
|
97 |
-
1. Financial Health
|
98 |
-
2.
|
99 |
-
3.
|
100 |
[/INST]"""
|
101 |
|
102 |
-
|
103 |
-
|
104 |
max_new_tokens=500,
|
105 |
temperature=0.7,
|
106 |
num_return_sequences=1,
|
107 |
truncation=True
|
108 |
)
|
109 |
|
110 |
-
return self.format_response(
|
111 |
|
112 |
except Exception as e:
|
113 |
logger.error(f"Analysis error: {str(e)}")
|
114 |
return f"Error in analysis: {str(e)}"
|
115 |
|
116 |
-
def
|
117 |
-
try:
|
118 |
-
# Extract latest year metrics
|
119 |
-
latest_metrics = {
|
120 |
-
'Revenue': income_df.loc[income_df['year'] == 'Total Net Revenue', '2025'].iloc[0],
|
121 |
-
'Net_Income': income_df.loc[income_df['year'] == 'Net Income', '2025'].iloc[0],
|
122 |
-
'Assets': balance_df.loc[balance_df['year'] == 'Total Assets', '2025'].iloc[0],
|
123 |
-
'Liabilities': balance_df.loc[balance_df['year'] == 'Total Liabilities', '2025'].iloc[0],
|
124 |
-
'Equity': balance_df.loc[balance_df['year'] == "Shareholder's Equity", '2025'].iloc[0]
|
125 |
-
}
|
126 |
-
|
127 |
-
return f"""Financial Metrics (2025):
|
128 |
-
Revenue: {self.format_number(latest_metrics['Revenue'])}
|
129 |
-
Net Income: {self.format_number(latest_metrics['Net_Income'])}
|
130 |
-
Total Assets: {self.format_number(latest_metrics['Assets'])}
|
131 |
-
Total Liabilities: {self.format_number(latest_metrics['Liabilities'])}
|
132 |
-
Shareholder's Equity: {self.format_number(latest_metrics['Equity'])}"""
|
133 |
-
|
134 |
-
except Exception as e:
|
135 |
-
logger.error(f"Error creating context: {str(e)}")
|
136 |
-
raise
|
137 |
-
|
138 |
-
def format_response(self, analysis_text, sentiment):
|
139 |
try:
|
140 |
sections = [
|
141 |
"# Financial Analysis Report\n\n",
|
142 |
-
f"## Market Sentiment: {sentiment['label'].upper()} ({sentiment['score']:.2%})\n\n"
|
|
|
|
|
|
|
|
|
143 |
]
|
144 |
|
145 |
-
current_section = None
|
146 |
for line in analysis_text.split('\n'):
|
147 |
line = line.strip()
|
148 |
if not line:
|
149 |
continue
|
150 |
|
151 |
-
if "Financial Health"
|
152 |
-
sections.append("
|
153 |
-
elif "Key Insights" in line:
|
154 |
-
sections.append("\n## Key Insights\n")
|
155 |
-
elif "Strategic Recommendations" in line:
|
156 |
-
sections.append("\n## Strategic Recommendations\n")
|
157 |
elif line:
|
158 |
-
|
|
|
|
|
159 |
|
160 |
return "".join(sections)
|
161 |
except Exception as e:
|
@@ -167,13 +152,22 @@ def analyze_statements(income_statement, balance_sheet):
|
|
167 |
if not income_statement or not balance_sheet:
|
168 |
return "Please upload both financial statements."
|
169 |
|
|
|
|
|
170 |
income_df = pd.read_csv(income_statement)
|
171 |
balance_df = pd.read_csv(balance_sheet)
|
172 |
|
|
|
|
|
|
|
|
|
|
|
173 |
analyzer = FinancialAnalyzer()
|
174 |
-
result = analyzer.analyze_financials(
|
175 |
|
176 |
-
|
|
|
|
|
177 |
return result
|
178 |
|
179 |
except Exception as e:
|
@@ -181,31 +175,26 @@ def analyze_statements(income_statement, balance_sheet):
|
|
181 |
return f"""Analysis Error: {str(e)}
|
182 |
|
183 |
Please check:
|
184 |
-
1.
|
185 |
-
2.
|
186 |
3. Files are not corrupted"""
|
187 |
|
188 |
# Create Gradio interface
|
189 |
iface = gr.Interface(
|
190 |
fn=analyze_statements,
|
191 |
inputs=[
|
192 |
-
gr.File(
|
193 |
-
|
194 |
-
file_types=[".csv"]
|
195 |
-
),
|
196 |
-
gr.File(
|
197 |
-
label="Balance Sheet",
|
198 |
-
file_types=[".csv"]
|
199 |
-
)
|
200 |
],
|
201 |
outputs=gr.Markdown(),
|
202 |
-
title="Financial Statement Analyzer",
|
203 |
-
description="Upload financial statements for AI analysis
|
|
|
204 |
theme="default",
|
205 |
-
|
206 |
)
|
207 |
|
208 |
-
# Launch
|
209 |
if __name__ == "__main__":
|
210 |
iface.launch(
|
211 |
server_name="0.0.0.0",
|
|
|
3 |
import pandas as pd
|
4 |
import torch
|
5 |
import logging
|
|
|
6 |
from transformers import pipeline
|
7 |
|
8 |
# Setup logging
|
|
|
12 |
)
|
13 |
logger = logging.getLogger(__name__)
|
14 |
|
|
|
15 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
16 |
logger.info(f"Using device: {DEVICE}")
|
17 |
|
|
|
|
|
|
|
|
|
|
|
18 |
class FinancialAnalyzer:
|
19 |
def __init__(self):
|
20 |
self.analysis_model = None
|
|
|
23 |
|
24 |
def load_models(self):
|
25 |
try:
|
26 |
+
logger.info("Loading TinyLlama model...")
|
27 |
self.analysis_model = pipeline(
|
28 |
"text-generation",
|
29 |
model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
|
30 |
torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32
|
31 |
)
|
32 |
|
33 |
+
logger.info("Loading FinBERT model...")
|
34 |
self.sentiment_model = pipeline(
|
35 |
"text-classification",
|
36 |
model="ProsusAI/finbert",
|
|
|
42 |
logger.error(f"Error loading models: {str(e)}")
|
43 |
raise
|
44 |
|
45 |
+
def extract_and_analyze(self, statement_text, statement_type):
|
46 |
+
"""Extract information from financial statement text"""
|
47 |
try:
|
48 |
+
prompt = f"""[INST] As a financial analyst, analyze this {statement_type}:
|
|
|
|
|
|
|
|
|
49 |
|
50 |
+
{statement_text}
|
51 |
+
|
52 |
+
Extract and summarize:
|
53 |
+
1. Key financial numbers for 2025
|
54 |
+
2. Notable trends
|
55 |
+
3. Important metrics
|
56 |
+
|
57 |
+
Focus on the most recent year (2025) and key financial indicators.
|
58 |
+
[/INST]"""
|
59 |
+
|
60 |
+
response = self.analysis_model(
|
61 |
+
prompt,
|
62 |
+
max_new_tokens=300,
|
63 |
+
temperature=0.3,
|
64 |
+
num_return_sequences=1,
|
65 |
+
truncation=True
|
66 |
+
)
|
67 |
|
68 |
+
return response[0]['generated_text']
|
69 |
except Exception as e:
|
70 |
+
logger.error(f"Error extracting data from {statement_type}: {str(e)}")
|
71 |
raise
|
72 |
|
73 |
+
def analyze_financials(self, income_text, balance_text):
|
74 |
try:
|
75 |
+
# First, extract key information from each statement
|
76 |
+
logger.info("Analyzing Income Statement...")
|
77 |
+
income_analysis = self.extract_and_analyze(income_text, "Income Statement")
|
78 |
|
79 |
+
logger.info("Analyzing Balance Sheet...")
|
80 |
+
balance_analysis = self.extract_and_analyze(balance_text, "Balance Sheet")
|
81 |
|
82 |
+
# Combine the analyses
|
83 |
+
combined_analysis = f"""Income Statement Analysis:
|
84 |
+
{income_analysis}
|
85 |
+
|
86 |
+
Balance Sheet Analysis:
|
87 |
+
{balance_analysis}"""
|
88 |
+
|
89 |
+
# Get sentiment
|
90 |
sentiment = self.sentiment_model(
|
91 |
+
combined_analysis[:512],
|
92 |
truncation=True
|
93 |
)[0]
|
94 |
|
95 |
+
# Generate final analysis
|
96 |
+
final_prompt = f"""[INST] Based on this financial analysis:
|
97 |
|
98 |
+
{combined_analysis}
|
99 |
|
100 |
Market Sentiment: {sentiment['label']} ({sentiment['score']:.2%})
|
101 |
|
102 |
+
Provide a concise analysis with:
|
103 |
+
1. Overall Financial Health (2-3 key points)
|
104 |
+
2. Main Business Insights (2-3 insights)
|
105 |
+
3. Key Recommendations (2-3 recommendations)
|
106 |
[/INST]"""
|
107 |
|
108 |
+
final_response = self.analysis_model(
|
109 |
+
final_prompt,
|
110 |
max_new_tokens=500,
|
111 |
temperature=0.7,
|
112 |
num_return_sequences=1,
|
113 |
truncation=True
|
114 |
)
|
115 |
|
116 |
+
return self.format_response(final_response[0]['generated_text'], sentiment, combined_analysis)
|
117 |
|
118 |
except Exception as e:
|
119 |
logger.error(f"Analysis error: {str(e)}")
|
120 |
return f"Error in analysis: {str(e)}"
|
121 |
|
122 |
+
def format_response(self, analysis_text, sentiment, raw_analysis):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
try:
|
124 |
sections = [
|
125 |
"# Financial Analysis Report\n\n",
|
126 |
+
f"## Market Sentiment: {sentiment['label'].upper()} ({sentiment['score']:.2%})\n\n",
|
127 |
+
"## Extracted Financial Data\n```\n",
|
128 |
+
raw_analysis,
|
129 |
+
"\n```\n\n",
|
130 |
+
"## Analysis\n\n"
|
131 |
]
|
132 |
|
|
|
133 |
for line in analysis_text.split('\n'):
|
134 |
line = line.strip()
|
135 |
if not line:
|
136 |
continue
|
137 |
|
138 |
+
if any(header in line for header in ["Financial Health", "Business Insights", "Recommendations"]):
|
139 |
+
sections.append(f"\n### {line}\n")
|
|
|
|
|
|
|
|
|
140 |
elif line:
|
141 |
+
if not line.startswith('-'):
|
142 |
+
line = f"- {line}"
|
143 |
+
sections.append(f"{line}\n")
|
144 |
|
145 |
return "".join(sections)
|
146 |
except Exception as e:
|
|
|
152 |
if not income_statement or not balance_sheet:
|
153 |
return "Please upload both financial statements."
|
154 |
|
155 |
+
logger.info("Reading financial statements...")
|
156 |
+
# Read files as text
|
157 |
income_df = pd.read_csv(income_statement)
|
158 |
balance_df = pd.read_csv(balance_sheet)
|
159 |
|
160 |
+
# Convert to string while preserving format
|
161 |
+
income_text = income_df.to_string(index=False)
|
162 |
+
balance_text = balance_df.to_string(index=False)
|
163 |
+
|
164 |
+
logger.info("Initializing analysis...")
|
165 |
analyzer = FinancialAnalyzer()
|
166 |
+
result = analyzer.analyze_financials(income_text, balance_text)
|
167 |
|
168 |
+
if DEVICE == "cuda":
|
169 |
+
torch.cuda.empty_cache()
|
170 |
+
|
171 |
return result
|
172 |
|
173 |
except Exception as e:
|
|
|
175 |
return f"""Analysis Error: {str(e)}
|
176 |
|
177 |
Please check:
|
178 |
+
1. Files are readable CSV files
|
179 |
+
2. Files contain financial data
|
180 |
3. Files are not corrupted"""
|
181 |
|
182 |
# Create Gradio interface
|
183 |
iface = gr.Interface(
|
184 |
fn=analyze_statements,
|
185 |
inputs=[
|
186 |
+
gr.File(label="Income Statement (CSV)", file_types=[".csv"]),
|
187 |
+
gr.File(label="Balance Sheet (CSV)", file_types=[".csv"])
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
],
|
189 |
outputs=gr.Markdown(),
|
190 |
+
title="AI Financial Statement Analyzer",
|
191 |
+
description="""Upload your financial statements for AI analysis.
|
192 |
+
The model will extract and analyze key financial information automatically.""",
|
193 |
theme="default",
|
194 |
+
flagging_mode="never"
|
195 |
)
|
196 |
|
197 |
+
# Launch
|
198 |
if __name__ == "__main__":
|
199 |
iface.launch(
|
200 |
server_name="0.0.0.0",
|