Spaces:
Sleeping
Sleeping
Upload multimodal_tools.py
Browse files- tools/multimodal_tools.py +177 -0
tools/multimodal_tools.py
ADDED
@@ -0,0 +1,177 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import base64
|
2 |
+
import os
|
3 |
+
from langchain_core.messages import AnyMessage, SystemMessage, HumanMessage
|
4 |
+
from langchain_google_genai import ChatGoogleGenerativeAI
|
5 |
+
from langchain.tools import Tool
|
6 |
+
from langchain_core.tools import tool
|
7 |
+
|
8 |
+
api_key = os.getenv("GEMINI_API_KEY")
|
9 |
+
|
10 |
+
# Create LLM class
|
11 |
+
vision_llm = ChatGoogleGenerativeAI(
|
12 |
+
model= "gemini-2.5-flash-preview-05-20",
|
13 |
+
temperature=0,
|
14 |
+
max_retries=2,
|
15 |
+
google_api_key=api_key
|
16 |
+
)
|
17 |
+
|
18 |
+
@tool("extract_text_tool", parse_docstring=True)
|
19 |
+
def extract_text(img_path: str) -> str:
|
20 |
+
"""Extract text from an image file using a multimodal model.
|
21 |
+
|
22 |
+
Args:
|
23 |
+
img_path (str): The path to the image file from which to extract text.
|
24 |
+
|
25 |
+
Returns:
|
26 |
+
str: The extracted text from the image, or an empty string if an error occurs.
|
27 |
+
"""
|
28 |
+
all_text = ""
|
29 |
+
try:
|
30 |
+
# Read image and encode as base64
|
31 |
+
with open(img_path, "rb") as image_file:
|
32 |
+
image_bytes = image_file.read()
|
33 |
+
|
34 |
+
image_base64 = base64.b64encode(image_bytes).decode("utf-8")
|
35 |
+
|
36 |
+
# Prepare the prompt including the base64 image data
|
37 |
+
message = [
|
38 |
+
HumanMessage(
|
39 |
+
content=[
|
40 |
+
{
|
41 |
+
"type": "text",
|
42 |
+
"text": (
|
43 |
+
"Extract all the text from this image. "
|
44 |
+
"Return only the extracted text, no explanations."
|
45 |
+
),
|
46 |
+
},
|
47 |
+
{
|
48 |
+
"type": "image_url",
|
49 |
+
"image_url": {
|
50 |
+
"url": f"data:image/png;base64,{image_base64}"
|
51 |
+
},
|
52 |
+
},
|
53 |
+
]
|
54 |
+
)
|
55 |
+
]
|
56 |
+
|
57 |
+
# Call the vision-capable model
|
58 |
+
response = vision_llm.invoke(message)
|
59 |
+
|
60 |
+
# Append extracted text
|
61 |
+
all_text += response.content + "\n\n"
|
62 |
+
|
63 |
+
return all_text.strip()
|
64 |
+
except Exception as e:
|
65 |
+
# A butler should handle errors gracefully
|
66 |
+
error_msg = f"Error extracting text: {str(e)}"
|
67 |
+
print(error_msg)
|
68 |
+
return ""
|
69 |
+
|
70 |
+
@tool("analyze_image_tool", parse_docstring=True)
|
71 |
+
def analyze_image_tool(user_query: str, img_path: str) -> str:
|
72 |
+
"""Answer the question reasoning on the image.
|
73 |
+
|
74 |
+
Args:
|
75 |
+
user_query (str): The question to be answered based on the image.
|
76 |
+
img_path (str): Path to the image file to be analyzed.
|
77 |
+
|
78 |
+
Returns:
|
79 |
+
str: The answer to the query based on image content, or an empty string if an error occurs.
|
80 |
+
"""
|
81 |
+
all_text = ""
|
82 |
+
try:
|
83 |
+
# Read image and encode as base64
|
84 |
+
with open(img_path, "rb") as image_file:
|
85 |
+
image_bytes = image_file.read()
|
86 |
+
|
87 |
+
image_base64 = base64.b64encode(image_bytes).decode("utf-8")
|
88 |
+
|
89 |
+
# Prepare the prompt including the base64 image data
|
90 |
+
message = [
|
91 |
+
HumanMessage(
|
92 |
+
content=[
|
93 |
+
{
|
94 |
+
"type": "text",
|
95 |
+
"text": (
|
96 |
+
f"User query: {user_query}"
|
97 |
+
),
|
98 |
+
},
|
99 |
+
{
|
100 |
+
"type": "image_url",
|
101 |
+
"image_url": {
|
102 |
+
"url": f"data:image/png;base64,{image_base64}"
|
103 |
+
},
|
104 |
+
},
|
105 |
+
]
|
106 |
+
)
|
107 |
+
]
|
108 |
+
|
109 |
+
# Call the vision-capable model
|
110 |
+
response = vision_llm.invoke(message)
|
111 |
+
|
112 |
+
# Append extracted text
|
113 |
+
all_text += response.content + "\n\n"
|
114 |
+
|
115 |
+
return all_text.strip()
|
116 |
+
except Exception as e:
|
117 |
+
# A butler should handle errors gracefully
|
118 |
+
error_msg = f"Error analyzing image: {str(e)}"
|
119 |
+
print(error_msg)
|
120 |
+
return ""
|
121 |
+
|
122 |
+
@tool("analyze_audio_tool", parse_docstring=True)
|
123 |
+
def analyze_audio_tool(user_query: str, audio_path: str) -> str:
|
124 |
+
"""Answer the question by reasoning on the provided audio file.
|
125 |
+
|
126 |
+
Args:
|
127 |
+
user_query (str): The question to be answered based on the audio content.
|
128 |
+
audio_path (str): Path to the audio file (e.g., .mp3, .wav, .flac, .aac, .ogg).
|
129 |
+
|
130 |
+
Returns:
|
131 |
+
str: The answer to the query based on audio content, or an error message/empty string if an error occurs.
|
132 |
+
"""
|
133 |
+
try:
|
134 |
+
# Determine MIME type from file extension
|
135 |
+
_filename, file_extension = os.path.splitext(audio_path)
|
136 |
+
file_extension = file_extension.lower()
|
137 |
+
|
138 |
+
supported_formats = {
|
139 |
+
".mp3": "audio/mp3", ".wav": "audio/wav", ".flac": "audio/flac",
|
140 |
+
".aac": "audio/aac", ".ogg": "audio/ogg"
|
141 |
+
}
|
142 |
+
|
143 |
+
if file_extension not in supported_formats:
|
144 |
+
return (f"Error: Unsupported audio file format '{file_extension}'. "
|
145 |
+
f"Supported extensions: {', '.join(supported_formats.keys())}.")
|
146 |
+
mime_type = supported_formats[file_extension]
|
147 |
+
|
148 |
+
# Read audio file and encode as base64
|
149 |
+
with open(audio_path, "rb") as audio_file:
|
150 |
+
audio_bytes = audio_file.read()
|
151 |
+
audio_base64 = base64.b64encode(audio_bytes).decode("utf-8")
|
152 |
+
|
153 |
+
# Prepare the prompt including the base64 audio data
|
154 |
+
message = [
|
155 |
+
HumanMessage(
|
156 |
+
content=[
|
157 |
+
{
|
158 |
+
"type": "text",
|
159 |
+
"text": f"User query: {user_query}",
|
160 |
+
},
|
161 |
+
{
|
162 |
+
"type": "audio",
|
163 |
+
"source_type": "base64",
|
164 |
+
"mime_type": mime_type,
|
165 |
+
"data": audio_base64
|
166 |
+
},
|
167 |
+
]
|
168 |
+
)
|
169 |
+
]
|
170 |
+
|
171 |
+
# Call the vision-capable model
|
172 |
+
response = vision_llm.invoke(message)
|
173 |
+
return response.content.strip()
|
174 |
+
except Exception as e:
|
175 |
+
error_msg = f"Error analyzing audio: {str(e)}"
|
176 |
+
print(error_msg)
|
177 |
+
return ""
|