whyumesh commited on
Commit
6de7312
·
verified ·
1 Parent(s): 8dea557

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +21 -11
app.py CHANGED
@@ -36,26 +36,29 @@ vision_model, vision_processor, code_model, code_tokenizer = load_models()
36
 
37
  VISION_SYSTEM_PROMPT = """You are an AI assistant specialized in analyzing images and videos of code editors. Your task is to:
38
 
39
- 1. Focus exclusively on frames containing code snippets or development environments.
40
- 2. Extract any visible code snippets, ignoring non-code content.
41
  3. Identify any error messages, warnings, or highlighting that indicates bugs within the code.
42
 
43
  Important:
44
- - Ignore frames showing irrelevant content such as the Eterniq dashboard, other window tabs, or non-code related screens.
45
- - Provide a thorough and accurate description of the code-specific content only.
46
  - If multiple code snippets are visible in different frames, describe each separately.
 
47
 
48
- Your analysis will be used to understand and potentially fix the code, so maintain a high level of detail and accuracy in your descriptions of code-related content.
49
  """
50
 
51
  CODE_SYSTEM_PROMPT = """You are an expert code debugging assistant. Based on the description of code and errors provided, your task is to:
 
52
  1. Identify the bugs and issues in the code
53
  2. Provide a corrected version of the code
54
  3. Explain the fixes made and why they resolve the issues
 
55
  Be thorough in your explanation and ensure the corrected code is complete and functional.
 
56
 
57
  Note: Please provide the output in a well-structured Markdown format. Remove all unnecessary information and exclude any additional code formatting such as triple backticks or language identifiers. The response should be ready to be rendered as Markdown content.
58
-
59
  """
60
 
61
  def process_image_for_code(image):
@@ -147,11 +150,18 @@ def process_video_for_code(video_path, max_frames=16, frame_interval=30):
147
 
148
  cap.release()
149
 
150
- # Process the first frame for now (you could extend this to handle multiple frames)
151
- if frames:
152
- return process_image_for_code(frames[0])
153
- else:
154
- return "No frames could be extracted from the video.", "No code could be analyzed."
 
 
 
 
 
 
 
155
 
156
  @spaces.GPU
157
  def process_content(content):
 
36
 
37
  VISION_SYSTEM_PROMPT = """You are an AI assistant specialized in analyzing images and videos of code editors. Your task is to:
38
 
39
+ 1. Focus EXCLUSIVELY on frames containing code snippets or development environments.
40
+ 2. Extract and describe ONLY the visible code snippets, completely ignoring any non-code content.
41
  3. Identify any error messages, warnings, or highlighting that indicates bugs within the code.
42
 
43
  Important:
44
+ - Completely disregard any frames showing the Eterniq dashboard, other window tabs, or non-code related screens.
45
+ - Provide descriptions ONLY for code-specific content.
46
  - If multiple code snippets are visible in different frames, describe each separately.
47
+ - Do not mention or describe any user interface elements, buttons, or non-code visuals.
48
 
49
+ Your analysis will be used to understand and potentially fix the code, so maintain a high level of detail and accuracy in your descriptions of code-related content only.
50
  """
51
 
52
  CODE_SYSTEM_PROMPT = """You are an expert code debugging assistant. Based on the description of code and errors provided, your task is to:
53
+
54
  1. Identify the bugs and issues in the code
55
  2. Provide a corrected version of the code
56
  3. Explain the fixes made and why they resolve the issues
57
+
58
  Be thorough in your explanation and ensure the corrected code is complete and functional.
59
+ Focus solely on the code and its issues, ignoring any mentions of user interfaces or non-code elements.
60
 
61
  Note: Please provide the output in a well-structured Markdown format. Remove all unnecessary information and exclude any additional code formatting such as triple backticks or language identifiers. The response should be ready to be rendered as Markdown content.
 
62
  """
63
 
64
  def process_image_for_code(image):
 
150
 
151
  cap.release()
152
 
153
+ # Process all extracted frames
154
+ vision_outputs = []
155
+ for frame in frames:
156
+ vision_output, _ = process_image_for_code(frame)
157
+ if vision_output.strip() and "no code" not in vision_output.lower():
158
+ vision_outputs.append(vision_output)
159
+
160
+ # Combine vision outputs and process with code model
161
+ combined_vision_output = "\n\n".join(vision_outputs)
162
+ _, code_output = process_image_for_code(combined_vision_output)
163
+
164
+ return combined_vision_output, code_output
165
 
166
  @spaces.GPU
167
  def process_content(content):