Anna

.gitignore

@@ -1,10 +1,13 @@
+
 __pycache__/
 *.py[cod]
 *$py.class
 model_files/
 medscribe_env/
 **/.DS_Store
 chat_data/
+.env/
 finetunedmodels/
+.env/
 src/geval/mistral-7b-instruct-v0.2.Q4_K_M.gguf
 eval_files/

CHANGELOG.md

@@ -18,7 +18,6 @@
 - Support multiple models for evaluation and finetuning
 - Support multiple models for chat
 
-
 ## 1-31
 - Added Evaluation
 - Added Evaluation via GUI

pegasus.py

@@ -0,0 +1,27 @@
+# import pegasus
+from transformers import PegasusForConditionalGeneration, PegasusTokenizer
+from GUI import *
+
+class Pegasus:
+    def __init__ (self):
+        # load pegasus for summarization
+        self.model_name = "google/pegasus-xsum"
+        self.tokenizer  = PegasusTokenizer.from_pretrained(self.model_name) # initialize pegasus' tokenizer
+        self.model  = PegasusForConditionalGeneration.from_pretrained(self.model_name) # initialize pegasus model
+
+
+    def tokenize(self, input_text): 
+        # tokenize text 
+        tokenized = self.tokenizer(input_text, truncation = True, padding = "longest", return_tensors = "pt")
+        return tokenized
+
+    def summarizer(self, tokenized_text):
+        #   summarize
+        summarized = self.model.generate(**tokenized_text)
+        return summarized
+
+    def detokenize(self, summarized_text):
+        # detokenize
+        summary = self.tokenizer.batch_decode (summarized_text)
+        return summary
+

requirements.txt

@@ -7,4 +7,4 @@ rouge-score>=0.1.2
 nltk>=3.8.1
 bert-score>=0.3.13
 evaluate>=0.4.0
-llama-cpp-python>=0.2.11
+llama-cpp-python>=0.2.11

sources.txt

@@ -0,0 +1,18 @@
+
+Works Cited
+
+Avinash. “LLM Evaluation Metrics — BLEU, ROGUE and METEOR Explained.” Medium, 7 Aug. 2024, avinashselvam.medium.com/llm-evaluation-metrics-bleu-rogue-and-meteor-explained-a5d2b129e87f. Accessed 11 Feb. 2025.
+Bais, Gourav. “LLM Evaluation for Text Summarization.” Neptune.ai, 25 Sept. 2024, neptune.ai/blog/llm-evaluation-text-summarization. Accessed 7 Feb. 2025. (photo)
+Banerjee, Satanjeev, and Alon Lavie. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. 2005.
+Bajaj, Ganesh. “When to Use BLEU Score: Evaluating Text Generation with N-Gram Precision.” Medium, Artificial Intelligence in Plain English, 26 Sept. 2024, ai.plainenglish.io/when-to-use-bleu-score-evaluating-text-generation-with-n-gram-precision-3431829a641e. Accessed 7 Feb. 2025. (photo)
+“BERT Score - a Hugging Face Space by Evaluate-Metric.” Huggingface.co, huggingface.co/spaces/evaluate-metric/bertscore.
+courtzc. “Evaluating the Performance of LLM Summarization Prompts with G-Eval.” Microsoft.com, 25 June 2024, learn.microsoft.com/en-us/ai/playbook/technology-guidance/generative-ai/working-with-llms/evaluation/g-eval-metric-for-summarization. Accessed 11 Feb. 2025.
+Falcão, Fabiano. “Metrics for Evaluating Summarization of Texts Performed by Transformers: How to Evaluate The….” Medium, 22 Apr. 2023, fabianofalcao.medium.com/metrics-for-evaluating-summarization-of-texts-performed-by-transformers-how-to-evaluate-the-b3ce68a309c3.
+Issiaka Faissal Compaore, et al. “AI-Driven Generation of News Summaries: Leveraging GPT and Pegasus Summarizer for Efficient Information Extraction.” Hal.science, 5 Feb. 2024, hal.science/hal-04437765, https://hal.science/hal-04437765. Accessed 11 Feb. 2025.
+Liu, Yang, et al. G-EVAL: NLG Evaluation Using GPT-4 with Better Human Alignment.
+Otten, Neri Van, and Neri Van Otten. “METEOR Metric in NLP: How It Works & How to Tutorial in Python.” Spot Intelligence, 26 Aug. 2024, spotintelligence.com/2024/08/26/meteor-metric-in-nlp-how-it-works-how-to-tutorial-in-python/.
+---. “ROUGE Metric in NLP: Complete Guide & How to Tutorial in Python.” Spot Intelligence, 12 Aug. 2024, spotintelligence.com/2024/08/12/rouge-metric-in-nlp/.
+Ruman. “BERT Score Explained | Medium.” Medium, 17 May 2024, rumn.medium.com/bert-score-explained-8f384d37bb06. Accessed 7 Feb. 2025. (photo)
+Yang, Xinyu, et al. “Navigating Dataset Documentations in AI: A Large-Scale Analysis of Dataset Cards on Hugging Face.” ArXiv.org, 24 Jan. 2024, arxiv.org/abs/2401.13822. Accessed 7 Feb. 2025.
+
+

src/components/__init__.py

@@ -6,4 +6,3 @@
 from components.llm_input import LLMInput
 from components.llm_list import LLMList
 from components.evaluation_visualizer import EvaluationVisualizer
-from components.llm_details import LLMDetails

src/components/evaluation_form.py

@@ -60,6 +60,5 @@ def start_evaluation(self):
             dataset_path, 
             model_path,
             start_idx,
-            end_idx,
-            self.use_geval.get()  # Pass the G-EVAL toggle state
-        )
+            end_idx
+        )

src/components/llm_details.py

@@ -4,33 +4,40 @@
 import os
 from itertools import islice
 from collections import OrderedDict
+
 class LLMDetails(tk.Frame):
     def __init__(self, parent, llm, **kwargs):
         super().__init__(parent, bg="#D2E9FC", **kwargs)
         self.llm = llm
         self.setup_input_area()
+
     def setup_input_area(self):
         self.grid_columnconfigure(0, weight=1)
         self.grid_rowconfigure(0, weight=1)
+
         # Container frame for LLM Name, LLM Details
         self.container = tk.Frame(self, bg="#D2E9FC")
         self.container.grid(row=0, column=0, sticky="nsew", padx=0, pady=15)
         self.container.grid_rowconfigure(0, weight=1)
         self.container.grid_rowconfigure(1, weight=1)
         self.container.grid_columnconfigure(0, weight=1)  # for name frame
         self.container.grid_columnconfigure(1, weight=1)  # for details frame
+
         # LLM Name container
         self.llm_name_container = tk.Frame(self.container, bg="#D2E9FC")
         self.llm_name_container.grid(row=0, column=0, sticky="ew", padx=(5, 10), pady=(20, 5))
         self.llm_name_container.grid_columnconfigure(0, weight=1)
-        # LLM details container frame with rounded corners
+
+	# LLM details container frame with rounded corners
         self.llm_details_container = RoundedFrame(self.container, "#FFFFFF", radius=50)
         self.llm_details_container.grid(row=1, column=0, sticky="nsew", padx=(20, 5), pady=10)
         self.llm_details_container.grid_rowconfigure(0, weight=1)
         self.llm_details_container.grid_columnconfigure(0, weight=1)
+
         # "LLM Name" string variable
         string = tk.StringVar()
         string.set("LLM: " + self.llm)
+
         # "Import LLM" area
         self.name_label = tk.Label(
             self.llm_name_container,
@@ -42,6 +49,7 @@ def setup_input_area(self):
             justify="left"
         )
         self.name_label.grid(row=0, column=0, sticky="w", padx=20, pady=(20, 10))
+
         # LLM detail area
         self.details = tk.Text(
             self.llm_details_container,
@@ -54,9 +62,11 @@ def setup_input_area(self):
             highlightthickness=0,
             relief="flat"
         )
+
         self.details.grid(row=0, column=0, sticky="nsew", padx=20, pady=20)
         self.details.grid_rowconfigure(0, weight=1)
         config = self.get_model_info()
+
         # get first 10 in config
         first_twenty = dict(islice(config.items(), 20))
         rev = OrderedDict(reversed(list(first_twenty.items())))
@@ -66,6 +76,8 @@ def setup_input_area(self):
             self.details.insert("1.0", "\n")
             self.details.insert("1.0", str_key + ": " + str_value)
             self.details.insert("1.0", "\n")
+
+
     def get_model_info(self):
         # cd to model_files
         cwd = os.getcwd()
@@ -74,4 +86,5 @@ def get_model_info(self):
         model_dir = os.path.join(parent, file)
         f = open(model_dir) # read config file of selected model
         config = json.load(f) # load configurations
-        return config
+        return config
+

src/components/llm_input.py

@@ -208,4 +208,4 @@ def get_input(self):
 
     def clear_input(self):
         self.input_text.delete("1.0", "end")
-        self.input_text.focus()
+        self.input_text.focus()

src/components/llm_list.py

@@ -75,16 +75,11 @@ def get_models(self):
         cwd = os.getcwd()  # current directory
         parent = os.path.dirname(cwd)  # parent directory
         model_dir = os.path.join(parent, "model_files")
-
-        # Create model_files directory if it doesn't exist
-        if not os.path.exists(model_dir):
-            os.makedirs(model_dir)
-
-        # check all folders in model_files folder
-        for folder in os.listdir(model_dir):
+        # check if model_files
+        for folder in os.listdir(model_dir):  # list all folders in model_files folder
             if os.path.isdir(os.path.join(model_dir, folder)):
                 folder_list.append(folder)
-                
+
         return folder_list
 
     def write_list(self, list):
@@ -101,4 +96,4 @@ def mouse_scroll(self, event):
             self.list.yview_scroll(-1, "units")
         elif event.num == 5:
             self.list.yview_scroll(1, "units")
-        return "break"
+        return "break"

src/components/navbar.py

@@ -4,18 +4,18 @@
 
 # Navigation items
 nav_items = [
-    ("Home","home"),
+    ("Home", "home"),
     ("Evaluations", "evaluations"),
     ("LLMs", "llms"),
-    ("Finetune", "finetune")
+    ("Finetune", "finetune"),
 ]
 
 class Navbar(tk.Frame):
     def __init__(self, parent, show_page_callback, **kwargs):
         super().__init__(parent, bg="#FFFFFF", **kwargs)
         # Set callback function
         self.show_page_callback = show_page_callback
-        self.current_page = "llms"  # Default page
+        self.current_page = "home"  # Default page
         self.buttons = {}  # Store button references
         self.logo_photo = None  
         self._setup_navbar()
@@ -94,7 +94,7 @@ def _setup_navbar(self):
             btn.bind("<Enter>", lambda e, b=btn_container, p=page: self._on_hover(b, p))
             btn.bind("<Leave>", lambda e, b=btn_container, p=page: self._on_leave(b, p))
 
-        self.set_active_page("llms")
+        self.set_active_page("home")
 
     def _handle_click(self, page):
         if page != self.current_page:
@@ -137,4 +137,4 @@ def _on_leave(self, container, page):
         else:
             container.configure(bg="#E8F0FE")
             for child in container.winfo_children():
-                child.configure(bg="#E8F0FE")
+                child.configure(bg="#E8F0FE")

src/download_pegasus.py

@@ -20,10 +20,19 @@ def download_model():
 
 """
 Model Details:
+<<<<<<< HEAD
+├── config.json              (Architecture blueprint)
+├── generation_config.json   (Controls how the model generates text: beam search)
+├── model.safetensors       (Weights that change when doing finetuning)
+├── special_tokens_map.json  (Maps special tokens like [PAD], [EOS], [UNK] to their IDs, Used for handling start/end of text, padding, unknown words)
+├── spiece.model            (The tokenizer's vocabulary and rules, Defines how to split text into tokens)
+└── tokenizer_config.json    (Configuration for the tokenizer's behavior)
+
+=======
 ├── config.json              (Model architecture configuration)
 ├── generation_config.json   (Text generation parameters)
 ├── model.safetensors       (Model weights)
 ├── special_tokens_map.json  (Mapping of special tokens like [PAD], [EOS])
 ├── tokenizer.json          (Tokenizer vocabulary and rules)
 └── tokenizer_config.json    (Tokenizer configuration)
-"""
+"""

src/main.py

@@ -6,6 +6,7 @@
 from pages.finetune_page import FinetunePage
 from pages.home_page import HomePage
 
+
 # Set environment variable
 os.environ['TOKENIZERS_PARALLELISM'] = 'false'
 
@@ -20,7 +21,6 @@ def setup_gui(self):
         self._configure_root()
         self._configure_grid()
         self._setup_navbar()
-        self.show_page("llms")  # Start with llm page
 
     def _configure_root(self):
         self.root.title("Assess.ai")
@@ -45,14 +45,16 @@ def _clear_content(self):
     def show_page(self, page_name):
         # Show the new page
         self._clear_content()
+
         if page_name == "evaluations":
             self.current_page = EvaluationPage(self.root)
         elif page_name == "llms":
             self.current_page = LLMsPage(self.root)
         elif page_name == "finetune":
             self.current_page = FinetunePage(self.root)
         elif page_name == "home":
-            self.current_page = HomePage(self.root)
+            self.current_page = HomePage(self.root, self.show_page)
+
 
 def main():
     # Initialize root and components
@@ -66,9 +68,10 @@ def main():
 
     # Initialize GUI
     app = AssessAIGUI(root)
+    app.show_page("home")
 
     # Start main loop
     root.mainloop()
 
 if __name__ == "__main__":
-    main()
+    main()