feat: updating files, adding chat model basics

app.py

@@ -1,45 +0,0 @@
-import gradio as gr
-from transformers import pipeline
-
-gbert_pipeline = pipeline("text-classification", model="LennardZuendorf/bert-base-german-cased", top_k=None, token="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")
-leolm_pipeline = pipeline("text-classification", model="LennardZuendorf/interpretor", top_k=None, token="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")
-chat_pipeline = pipeline("conversational", model="meta-llama/Llama-2-7b-chat-hf", top_k=None, token="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")
-
-with gr.Blocks() as ui:
-    with gr.Row():
-        gr.Markdown(
-            """
-            # Thesis Model Demos
-            Select between tabs below for try the different models.
-            """)
-    with gr.Tab("GBERT HateSpeech Detection"):
-        with gr.Row():
-            gr.Markdown(
-                """
-                ### GBERT (German Language BERT by Deepset) Demo
-                #### Model finetuned on German Hate Speech dataset (~3,5k sequences)
-                """)
-        with gr.Row():
-            gr.Interface.from_pipeline(gbert_pipeline)
-
-    with gr.Tab("LeoLM HateSpeech Detection"):
-        with gr.Row():
-            gr.Markdown(
-                """
-                ### LeoLM (German Language FineTuned LlaMa2 Model) Demo
-                #### Model finetuned on German Hate Speech dataset (~3,5k sequences)
-                """)
-        with gr.Row():
-            gr.Button("New Tiger")
-
-    with gr.Tab("Chat Model Interface"):
-        with gr.Row():
-            gr.Markdown(
-                """
-                ### LlaMa 2 Chat Demo
-                """)
-        with gr.Row():
-            gr.Interface.from_pipeline(chat_pipeline)
-
-if __name__ == "__main__":
-    ui.launch()

chatmodel.py

@@ -0,0 +1,60 @@
+from transformers import pipeline
+import torch
+from transformers import AutoTokenizer
+import os
+
+token = os.environ.get("HGFTOKEN")
+
+model = "meta-llama/Llama-2-7b-chat-hf"
+tokenizer = AutoTokenizer.from_pretrained(model, token=token)
+
+llama_pipeline = pipeline(
+    "text-generation",
+    model=model,
+    torch_dtype=torch.float32,
+    device_map="auto",
+)
+
+# Formatting function for message and history
+def format_message(message: str, history: list, system_prompt:str, memory_limit: int = 3) -> str:
+
+    if len(history) > memory_limit:
+        history = history[-memory_limit:]
+
+    system_prompt="<s>[INST] <<SYS>>\n"+system_prompt+"\n<</SYS>>"
+
+    if len(history) == 0:
+        return system_prompt + f"{message} [/INST]"
+
+    formatted_message = system_prompt + f"{history[0][0]} [/INST] {history[0][1]} </s>"
+
+    # Handle conversation history
+    for user_msg, model_answer in history[1:]:
+        formatted_message += f"<s>[INST] {user_msg} [/INST] {model_answer} </s>"
+
+    # Handle the current message
+    formatted_message += f"<s>[INST] {message} [/INST]"
+
+    return formatted_message
+
+# Generate a response from the Llama model
+def interference(message: str, history: list, ) -> str:
+    system_prompt="You are a helpful assistant providing reasonable answers."
+
+    query = format_message(message, history, system_prompt)
+    response = ""
+
+    sequences = llama_pipeline(
+        query,
+        do_sample=True,
+        top_k=10,
+        num_return_sequences=1,
+        eos_token_id=tokenizer.eos_token_id,
+        max_length=1024,
+    )
+
+    generated_text = sequences[0]['generated_text']
+    response = generated_text[len(query):]  # Remove the prompt from the output
+
+    print("Chatbot:", response.strip())
+    return response.strip()

ui.py

@@ -0,0 +1,40 @@
+import gradio as gr
+import chatmodel as chat
+import interpret as shap
+import visualize as viz
+
+with gr.Blocks() as ui:
+    with gr.Row():
+        gr.Markdown(
+            """
+            # Thesis Demo - AI Chat Application with XAI
+            Select between tabs below for the different views.
+            """)
+    with gr.Tab("LlaMa 2 ChatBot"):
+        with gr.Row():
+            gr.Markdown(
+                """
+                ### ChatBot Demo
+                #### Standard LlaMa 2 7B Model fine-tuned for chat and transformed to huggingface format.
+                """)
+        with gr.Row():
+            gr.ChatInterface(chat.interference)
+
+    with gr.Tab("SHAP Backend"):
+        with gr.Row():
+            gr.Markdown(
+                """
+                ### SHAP Dashboard
+                #### SHAP Visualization Dashboard adopted from shapash
+                """)
+
+    with gr.Tab("LlaMa 2 Model Overview"):
+        with gr.Row():
+            gr.Markdown(
+                """
+                ### LlaMa 2 Model Overview for Transparency
+                """)
+
+
+if __name__ == "__main__":
+    ui.launch(debug=True)