chore: file moving and renaming

app.py

@@ -1,51 +1,67 @@
+# main application file initializing the gradio based ui and calling other modules
+
+# external package imports
 import gradio as gr
-import mistral
-import interpret as shap
-import visualize as viz
 import markdown
 
-def load_md(filename):
-    path = "./public/"+str(filename)
+
+# internal imports
+from model import mistral
+from explanation import interpret as shap
+
+
+# function to load md files in pthon as a string
+def load_md(path):
 
     # credit: official python-markdown documentation (https://python-markdown.github.io/reference/)
     with open(path, "r") as file:
         text = file.read()
 
+    # return markdown as a string
     return markdown.markdown(text)
 
+# ui interface based on Gradio Blocks (see documentation: https://www.gradio.app/docs/interface)
 with gr.Blocks() as ui:
+    # header row with markdown based text
     with gr.Row():
         gr.Markdown(
             """
             # Thesis Demo - AI Chat Application with XAI
             ### Select between tabs below for the different views.
             """)
-    with gr.Tab("Mistral AI ChatBot"):
+    # ChatBot tab used to chat with the AI chatbot
+    with gr.Tab("AI ChatBot"):
         with gr.Row():
             gr.Markdown(
                 """
                 ### ChatBot Demo
                 Mitral AI 7B Model fine-tuned for instruction and fully open source (see at [HGF](https://huggingface.co/mistralai/Mistral-7B-v0.1))
                 """)
-
+        # row with chatbot ui displaying "conversation" with the model (see documentation: https://www.gradio.app/docs/chatbot)
         with gr.Row():
             chatbot = gr.Chatbot(layout="panel", show_copy_button=True,avatar_images=("./public/human.jpg","./public/bot.jpg"))
+        # disclaimer information row - data is recorded for shap dashboard and model explanability
         with gr.Row():
             gr.Markdown(
                 """
                 ##### ⚠️ All Conversations are recorded for qa assurance and explanation functionality!
                 """)
+        # row with input textbox
         with gr.Row():
                 prompt = gr.Textbox(label="Input Message")
+        # row with columns for buttons to submit and clear content
         with gr.Row():
             with gr.Column(scale=1):
+                # default clear button which clearn the given components (see documentation: https://www.gradio.app/docs/clearbutton)
                 clear_btn = gr.ClearButton([prompt, chatbot])
             with gr.Column(scale=1):
                 submit_btn = gr.Button("Submit")
 
+        # two functions performing the same action (triggered the model response), when the button is used or the textbox submit function is used (clicking enter).
         submit_btn.click(mistral.chat, [prompt, chatbot], [prompt, chatbot])
         prompt.submit(mistral.chat, [prompt, chatbot], [prompt, chatbot])
 
+    # explanations tab used to provide explanations for a specific conversation
     with gr.Tab("Explanations"):
         with gr.Row():
             gr.Markdown(
@@ -54,7 +70,7 @@ with gr.Blocks() as ui:
                 SHAP Visualization Dashboard adopted from [shapash](https://github.com/MAIF/shapash)
                 """)
 
-
+    # shap dashboard tab for shapash inspired dashboard
     with gr.Tab("SHAP Dashboard"):
         with gr.Row():
             gr.Markdown(
@@ -63,6 +79,7 @@ with gr.Blocks() as ui:
                 SHAP Visualization Dashboard adopted from [shapash](https://github.com/MAIF/shapash)
                 """)
 
+    # visualize dashboard to display global visualization provided by the BERTViz adoption
     with gr.Tab("Visualize Dashboard"):
         with gr.Row():
             gr.Markdown(
@@ -71,17 +88,18 @@ with gr.Blocks() as ui:
                 Visualization Dashboard adopted from [BERTViz](https://github.com/jessevig/bertviz)
                 """)
 
-    with gr.Tab("Mitral Model Overview"):
-        with gr.Row():
-            gr.Markdown(
-                """
-                ### Mistral 7B Model & Data Overview for Transparency
-                Adopted from official [model paper](https://arxiv.org/abs/2310.06825) by Mistral AI
-                """)
+    # model overview tab for transparency
+    with gr.Tab("Model Overview"):
+        with gr.Tab("Mistral 7B Instruct"):
+            gr.Markdown(value=load_md("./model/mistral.md"))
+        with gr.Tab("LlaMa 2 7B Chat"):
+            gr.Markdown(value=load_md("./model/llama2.md"))
 
+    # final row to show legal information - credits, data protection and link to the LICENSE on GitHub
     with gr.Row():
         with gr.Accordion("Credits, Data Protection and License", open=False):
-            gr.Markdown(value=load_md("credits_dataprotection_license.md"))
+            gr.Markdown(value=load_md("./public/credits_dataprotection_license.md"))
 
+# launch function for Gradio Interface
 if __name__ == "__main__":
     ui.launch(debug=True)

model/llama2.md

@@ -0,0 +1,6 @@
+### Credits
+
+### Data Protection
+
+### License
+This Product is licensed under the MIT license.

model/llama2.py

@@ -0,0 +1,66 @@
+from huggingface_hub import InferenceClient
+import os
+
+# huggingface token used to load closed off models
+token = os.environ.get("HGFTOKEN")
+
+# interference client created from mistral 7b instruction fine tuned model
+# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
+interference = InferenceClient(
+    "mistralai/Mistral-7B-Instruct-v0.1"
+)
+
+# default model settings
+model_temperature = 0.7
+model_max_new_tokens = 320
+model_top_p = 0.95
+model_repetition_penalty = 1.1
+
+# chat function - basically the main function calling other functions and returning a response to showcase in chatbot ui
+def chat (prompt, history,):
+
+    # creating formatted prompt and calling for an answer from the model
+    formatted_prompt = format_prompt(prompt, history)
+    answer=respond(formatted_prompt)
+
+    # updating the chat history with the new answer
+    history.append((prompt, answer))
+
+    # returning the chat history to be displayed in the chatbot ui
+    return "",history
+
+# function to format prompt in a way that is understandable for the text generation model
+# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
+def format_prompt(message, history):
+    prompt = "<s>"
+
+    # labeling each message in the history as bot or user
+    for user_prompt, bot_response in history:
+        prompt += f"[INST] {user_prompt} [/INST]"
+        prompt += f" {bot_response}</s> "
+    prompt += f"[INST] {message} [/INST]"
+    return prompt
+
+# function to get the response
+# credit: minimally changed from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
+def respond(formatted_prompt):
+
+    # setting model temperature and
+    temperature = float(model_temperature)
+    if temperature < 1e-2:
+        temperature = 1e-2
+    top_p = float(model_top_p)
+
+    # creating model arguments/settings
+    generate_kwargs = dict(
+        temperature=temperature,
+        max_new_tokens=model_max_new_tokens,
+        top_p=top_p,
+        repetition_penalty=model_repetition_penalty,
+        do_sample=True,
+        seed=42,
+    )
+
+    # calling for model output and returning it
+    output = interference.text_generation(formatted_prompt, **generate_kwargs, stream=False, details=True, return_full_text=False).generated_text
+    return output

model/mistral.md

@@ -0,0 +1,6 @@
+### Credits
+
+### Data Protection
+
+### License
+This Product is licensed under the MIT license.

mistral.py → model/mistral.py

@@ -1,41 +1,57 @@
 from huggingface_hub import InferenceClient
 import os
-import gradio as gr
 
+# huggingface token used to load closed off models
 token = os.environ.get("HGFTOKEN")
 
+# interference client created from mistral 7b instruction fine tuned model
+# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
 interference = InferenceClient(
     "mistralai/Mistral-7B-Instruct-v0.1"
 )
 
+# default model settings
 model_temperature = 0.7
-model_max_new_tokens = 256
+model_max_new_tokens = 320
 model_top_p = 0.95
 model_repetition_penalty = 1.1
 
+# chat function - basically the main function calling other functions and returning a response to showcase in chatbot ui
 def chat (prompt, history,):
 
+    # creating formatted prompt and calling for an answer from the model
     formatted_prompt = format_prompt(prompt, history)
     answer=respond(formatted_prompt)
 
+    # updating the chat history with the new answer
     history.append((prompt, answer))
 
+    # returning the chat history to be displayed in the chatbot ui
     return "",history
 
+# function to format prompt in a way that is understandable for the text generation model
+# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
 def format_prompt(message, history):
     prompt = "<s>"
+
+    # labeling each message in the history as bot or user
     for user_prompt, bot_response in history:
         prompt += f"[INST] {user_prompt} [/INST]"
         prompt += f" {bot_response}</s> "
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
+# function to get the response
+# credit: minimally changed from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
 def respond(formatted_prompt):
+
+    # setting model temperature and
     temperature = float(model_temperature)
     if temperature < 1e-2:
         temperature = 1e-2
     top_p = float(model_top_p)
 
+    # creating model arguments/settings
     generate_kwargs = dict(
         temperature=temperature,
         max_new_tokens=model_max_new_tokens,
@@ -45,5 +61,6 @@ def respond(formatted_prompt):
         seed=42,
     )
 
+    # calling for model output and returning it
     output = interference.text_generation(formatted_prompt, **generate_kwargs, stream=False, details=True, return_full_text=False).generated_text
     return output