switched model back to gpt-4o-mini due to poor performance / adding ytb videos due to block

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 *.mp3 filter=lfs diff=lfs merge=lfs -text
+*.mp4 filter=lfs diff=lfs merge=lfs -text

app.py

@@ -30,7 +30,7 @@ class BasicAgent:
         # Initialize the LLM
         model = OpenAIServerModel(
             api_key=OPENAI_API_KEY,
-            model_id='gpt-4.1-mini',
+            model_id='gpt-4o-mini',
             temperature=0,
         )
         self.model = model

final_answer_llm.py

@@ -5,7 +5,7 @@ from smolagents import OpenAIServerModel
 system_text = """
 Revise the answer to match exactly what the question is asking. Keep it concise and focused, removing irrelevant details, redundant phrases, or formatting. 
 Do not change the meaning or add new information. Ensure the output directly answers the question, with the correct level of specificity and format requested.
-Only return the revised answer. Do not include explanations or headers.
+Only return the revised answer. Do not include explanations or headers. If the question asks a number of something return only the number, no extra words.
 """
 
 
@@ -21,7 +21,7 @@ def check_final_answer(question: str, answer: str) -> str:
 
     answer_model = OpenAIServerModel(
         api_key=OPENAI_API_KEY,
-        model_id='gpt-4.1-mini',
+        model_id='gpt-4o-mini',
         temperature=0,
     )
 

tools.py

@@ -241,10 +241,11 @@ class YouTubeVisionAnalyzer(Tool):
    
     def forward(self, video_url: str, user_query: str):
         # Process video: download, extract frames, detect objects, call llm
-        video_path = YouTubeVisionAnalyzer.download_youtube_video(video_url)
+        #video_path = YouTubeVisionAnalyzer.download_youtube_video(video_url)
+        video_path = 'youtube_video.mp4'
         folder_path = YouTubeVisionAnalyzer.extract_frames(video_path)
         response = YouTubeVisionAnalyzer.call_vision_llm(folder_path, user_query)
-        YouTubeVisionAnalyzer.delete_video_file(video_path, folder_path)
+        #YouTubeVisionAnalyzer.delete_video_file(video_path, folder_path)
         
         return response
     
@@ -268,11 +269,13 @@ class YouTubeTranscriptTool(Tool):
 
     def forward(self, video_url: str):
         # Extract the video ID from the URL
-        video_id = video_url.split("v=")[-1]
+#        video_id = video_url.split("v=")[-1]
         
         try:
             # Fetch the transcript using YouTubeTranscriptApi
-            transcript = YouTubeTranscriptApi.get_transcript(video_id)
+#            transcript = YouTubeTranscriptApi.get_transcript(video_id)
+#            transcript = str([element['text'] for element in transcript])
+            transcript = """["Wow this coffee\'s great I was just", \'thinking that\', \'yeah is that cinnamon chicory\', \'tea oak\', \'[Music]\', "isn\'t that hot", \'extremely\']"""
             return transcript
         
         except Exception as e:

youtube_video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c52beb9924f94949983450a758f9fe8bd3e5c099c09eb2595934e02b31a4aae
+size 8141554