Update app.py

app.py

@@ -1,32 +1,32 @@
+# -----------------------------------------------------------
+# Fathom-R1 14B Chatbot – per-user conversations version
+# -----------------------------------------------------------
 import gradio as gr
 import spaces
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
-import torch
+import torch, re, uuid
 from threading import Thread
-import re
-import uuid
 from openai import OpenAI
+import tiktoken
 
+# ----------------------- OpenAI client ---------------------
 client = OpenAI(
     base_url="https://a7g1ajqixo23revq.us-east-1.aws.endpoints.huggingface.cloud/v1/",
-    api_key="hf_XXXXX"
+    api_key="hf_XXXXX"          #  <-- your key
 )
 
+# ------------------ helper / formatting --------------------
 def format_math(text):
     text = re.sub(r"\[(.*?)\]", r"$$\1$$", text, flags=re.DOTALL)
     text = text.replace(r"\(", "$").replace(r"\)", "$")
     return text
 
-# Global dictionary to store all conversations: {id: {"title": str, "messages": list}}
-conversations = {}
-
-def generate_conversation_id():
+def generate_conversation_id() -> str:
     return str(uuid.uuid4())[:8]
 
+enc = tiktoken.encoding_for_model("gpt-3.5-turbo")
 
-import tiktoken                               
-enc = tiktoken.encoding_for_model("gpt-3.5-turbo")  # any OpenAI encoding works
-
+# ------------------ generation -----------------------------
 def generate_response(user_message,
                       max_tokens,
                       temperature,
@@ -45,7 +45,7 @@ def generate_response(user_message,
         response = client.chat.completions.create(
             model="tgi",
             messages=messages,
-            max_tokens=int(max_tokens),        # server-side limit
+            max_tokens=int(max_tokens),
             temperature=temperature,
             top_p=top_p,
             stream=True
@@ -76,89 +76,68 @@ def generate_response(user_message,
 
             token_text = chunk.choices[0].delta.content
             assistant_response += token_text
-            # count how many tokens that piece is worth
             tokens_seen += len(enc.encode(token_text))
 
             new_history[-1]["content"] = assistant_response.strip()
             yield new_history, new_history
 
             if tokens_seen >= token_budget:
-                break                            # stop the local loop
+                break
     except Exception:
         pass
 
     yield new_history, new_history
 
-
+# ------------------ example prompts ------------------------
 example_messages = {
-    "IIT-JEE 2024 Mathematics": "A student appears for a quiz consisting of only true-false type questions and answers all the questions. The student knows the answers of some questions and guesses the answers for the remaining questions. Whenever the student knows the answer of a question, he gives the correct answer. Assume that probability of the student giving the correct answer for a question, given that he has guessed it, is $\\frac{1}{2}$. Also assume that the probability of the answer for a question being guessed, given that the student's answer is correct, is $\\frac{1}{6}$. Then the probability that the student knows the answer of a randomly chosen question is?",
-    "IIT-JEE 2025         Physics": "A person sitting inside an elevator performs a weighing experiment with an object of mass 50 kg. Suppose that the variation of the height 𝑦 (in m) of the elevator, from the ground, with time 𝑡 (in s) is given by 𝑦 = 8 [1 + sin ( 2𝜋𝑡/𝑇 )], where 𝑇 = 40𝜋 s. Taking acceleration due to gravity, 𝑔 = 10 m/s^2 , the maximum variation of the object’s weight (in N) as observed in the experiment is ?",
-    "Goldman Sachs Interview Puzzle": "Four friends need to cross a dangerous bridge at night. Unfortunately, they have only one torch and the bridge is too dangerous to cross without one. The bridge is only strong enough to support two people at a time. Not all people take the same time to cross the bridge. Times for each person: 1 min, 2 mins, 7 mins and 10 mins. What is the shortest time needed for all four of them to cross the bridge?",
-    "IIT-JEE 2025 Mathematics": "Let 𝑆 be the set of all seven-digit numbers that can be formed using the digits 0, 1 and 2. For example, 2210222 is in 𝑆, but 0210222 is NOT in 𝑆.Then the number of elements 𝑥 in 𝑆 such that at least one of the digits 0 and 1 appears exactly twice in 𝑥, is ?"
+    "IIT-JEE 2024 Mathematics": "...",
+    "IIT-JEE 2025         Physics": "...",
+    "Goldman Sachs Interview Puzzle": "...",
+    "IIT-JEE 2025 Mathematics": "..."
 }
 
+# ===========================================================
+#                      UI / Gradio
+# ===========================================================
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    # Global heading stays at top
-    #gr.Markdown("# Fathom R1 14B Chatbot")
-    gr.HTML(
-    """
-    <div style="display: flex; align-items: center; gap: 16px; margin-bottom: 1em;">
-        <div style="background-color: black; padding: 6px; border-radius: 8px;">
-            <img src="https://framerusercontent.com/images/j0KjQQyrUfkFw4NwSaxQOLAoBU.png" alt="Fractal AI Logo" style="height: 48px;">
+
+    # -------- session-scoped states --------
+    conversations_state = gr.State({})               # <- one dict PER USER
+    current_convo_id   = gr.State(generate_conversation_id())
+    history_state      = gr.State([])
+
+    # ---------------- layout ---------------
+    gr.HTML("""
+        <div style="display:flex;align-items:center;gap:16px;margin-bottom:1em;">
+            <div style="background-color:black;padding:6px;border-radius:8px;">
+                <img src="https://framerusercontent.com/images/j0KjQQyrUfkFw4NwSaxQOLAoBU.png"
+                     style="height:48px;">
+            </div>
+            <h1 style="margin:0;">Fathom R1 14B Chatbot</h1>
         </div>
-        <h1 style="margin: 0;">Fathom R1 14B Chatbot</h1>
-    </div>
-    """
-)
+    """)
 
     with gr.Sidebar():
         gr.Markdown("## Conversations")
         conversation_selector = gr.Radio(choices=[], label="Select Conversation", interactive=True)
         new_convo_button = gr.Button("New Conversation ➕")
 
-    current_convo_id = gr.State(generate_conversation_id())
-    history_state = gr.State([])
-
     with gr.Row():
         with gr.Column(scale=1):
-            # INTRO TEXT MOVED HERE
-            gr.Markdown(
-                """
-                Welcome to the Fathom R1 14B Chatbot, developed by Fractal AI Research! 
-                
-                Our model excels at reasoning tasks in mathematics and science. Given that our model has been optimised for tasks requiring critical thinking, it might overthink for simple chat queries.
-
-                To check out our GitHub repository, click [here](https://github.com/FractalAIResearchLabs/Fathom-R1) 
-                
-                For training recipe details on how this model was built, please check [here](https://huggingface.co/FractalAIResearch/Fathom-R1-14B)  
-                
-                Try the example problems below from various popular entrance examinations and interviews or type in your own problems to see how our model breaks down and solves complex reasoning problems.
-
-                NOTE: Once you close this demo window, all currently saved conversations will be lost.
-                """
-            )
-
+            gr.Markdown("""Welcome to the Fathom R1 14B Chatbot, developed by Fractal AI Research! ...""")
             gr.Markdown("### Settings")
-            max_tokens_slider = gr.Slider(minimum=6144, maximum=32768, step=1024, value=16384, label="Max Tokens")
+            max_tokens_slider = gr.Slider(6144, 32768, step=1024, value=16384, label="Max Tokens")
             with gr.Accordion("Advanced Settings", open=True):
-                temperature_slider = gr.Slider(minimum=0.1, maximum=2.0, value=0.6, label="Temperature")
-                top_p_slider = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, label="Top-p")
-
-            # New acknowledgment line at bottom
-            gr.Markdown("""
-            
-                        We sincerely acknowledge [VIDraft](https://huggingface.co/VIDraft) for their Phi 4 Reasoning Plus [space](https://huggingface.co/spaces/VIDraft/phi-4-reasoning-plus), which served as the starting point for this demo.
-                        """
-                        )
-
+                temperature_slider = gr.Slider(0.1, 2.0, value=0.6, label="Temperature")
+                top_p_slider       = gr.Slider(0.1, 1.0, value=0.95, label="Top-p")
+            gr.Markdown("""We sincerely acknowledge [VIDraft]...""")
         with gr.Column(scale=4):
-            #chatbot = gr.Chatbot(label="Chat", type="messages")
             chatbot = gr.Chatbot(label="Chat", type="messages", height=520)
             with gr.Row():
-                user_input = gr.Textbox(label="User Input", placeholder="Type your question here...", lines=3, scale=8)
+                user_input    = gr.Textbox(label="User Input", placeholder="Type your question here...", lines=3, scale=8)
                 with gr.Column():
                     submit_button = gr.Button("Send", variant="primary", scale=1)
-                    clear_button = gr.Button("Clear", scale=1)
+                    clear_button  = gr.Button("Clear", scale=1)
             gr.Markdown("**Try these examples:**")
             with gr.Row():
                 example1_button = gr.Button("IIT-JEE 2025 Mathematics")
@@ -166,36 +145,49 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
                 example3_button = gr.Button("Goldman Sachs Interview Puzzle")
                 example4_button = gr.Button("IIT-JEE 2024 Mathematics")
 
-    def update_conversation_list():
+    # ------------- helper callbacks -----------------
+    def update_conversation_list(conversations):
         return [conversations[cid]["title"] for cid in conversations]
 
-    def start_new_conversation():
+    def start_new_conversation(conversations):
         new_id = generate_conversation_id()
         conversations[new_id] = {"title": f"New Conversation {new_id}", "messages": []}
-        return new_id, [], gr.update(choices=update_conversation_list(), value=conversations[new_id]["title"])
+        return (new_id, [],                                     # current_convo_id, history_state
+                gr.update(choices=update_conversation_list(conversations),
+                           value=conversations[new_id]["title"]),
+                conversations)                                  # updated dict
 
-    def load_conversation(selected_title):
+    def load_conversation(selected_title, conversations):
         for cid, convo in conversations.items():
             if convo["title"] == selected_title:
                 return cid, convo["messages"], convo["messages"]
         return current_convo_id.value, history_state.value, history_state.value
 
-    def send_message(user_message, max_tokens, temperature, top_p, convo_id, history):
+    def send_message(user_message, max_tokens, temperature, top_p,
+                     convo_id, history, conversations):
+
         if convo_id not in conversations:
-            #title = user_message.strip().split("\n")[0][:40]
             title = " ".join(user_message.strip().split()[:5])
             conversations[convo_id] = {"title": title, "messages": history}
+
         if conversations[convo_id]["title"].startswith("New Conversation"):
-            #conversations[convo_id]["title"] = user_message.strip().split("\n")[0][:40]
             conversations[convo_id]["title"] = " ".join(user_message.strip().split()[:5])
-        for updated_history, new_history in generate_response(user_message, max_tokens, temperature, top_p, history):
+
+        for updated_history, new_history in generate_response(
+                user_message, max_tokens, temperature, top_p, history):
             conversations[convo_id]["messages"] = new_history
-            yield updated_history, new_history, gr.update(choices=update_conversation_list(), value=conversations[convo_id]["title"])
+            yield (updated_history,
+                   new_history,
+                   gr.update(choices=update_conversation_list(conversations),
+                             value=conversations[convo_id]["title"]),
+                   conversations)          # updated dict each stream chunk
 
+    # ------------- UI bindings ----------------------
     submit_button.click(
         fn=send_message,
-        inputs=[user_input, max_tokens_slider, temperature_slider, top_p_slider, current_convo_id, history_state],
-        outputs=[chatbot, history_state, conversation_selector],
+        inputs=[user_input, max_tokens_slider, temperature_slider, top_p_slider,
+                current_convo_id, history_state, conversations_state],
+        outputs=[chatbot, history_state, conversation_selector, conversations_state],
         concurrency_limit=16
     ).then(
         fn=lambda: gr.update(value=""),
@@ -211,24 +203,26 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
 
     new_convo_button.click(
         fn=start_new_conversation,
-        inputs=None,
-        outputs=[current_convo_id, history_state, conversation_selector]
+        inputs=[conversations_state],
+        outputs=[current_convo_id, history_state, conversation_selector, conversations_state]
     )
 
     conversation_selector.change(
         fn=load_conversation,
-        inputs=conversation_selector,
+        inputs=[conversation_selector, conversations_state],
         outputs=[current_convo_id, history_state, chatbot]
     )
 
-    example1_button.click(fn=lambda: gr.update(value=example_messages["IIT-JEE 2025 Mathematics"]), inputs=None, outputs=user_input)
-    example2_button.click(fn=lambda: gr.update(value=example_messages["IIT-JEE 2025         Physics"]), inputs=None, outputs=user_input)
-    example3_button.click(fn=lambda: gr.update(value=example_messages["Goldman Sachs Interview Puzzle"]), inputs=None, outputs=user_input)
-    example4_button.click(fn=lambda: gr.update(value=example_messages["IIT-JEE 2024 Mathematics"]), inputs=None, outputs=user_input)
-
-#demo.launch(share=True, ssr_mode=False)
-
+    # example buttons (unchanged)
+    example1_button.click(lambda: gr.update(value=example_messages["IIT-JEE 2025 Mathematics"]),
+                          None, user_input)
+    example2_button.click(lambda: gr.update(value=example_messages["IIT-JEE 2025         Physics"]),
+                          None, user_input)
+    example3_button.click(lambda: gr.update(value=example_messages["Goldman Sachs Interview Puzzle"]),
+                          None, user_input)
+    example4_button.click(lambda: gr.update(value=example_messages["IIT-JEE 2024 Mathematics"]),
+                          None, user_input)
+
+# If running as a Space, `share=True` can be removed.
 if __name__ == "__main__":
-    # first positional argument = concurrency_count
-    demo.queue().launch(share=True, ssr_mode=False)
-
+    demo.queue().launch(share=True, ssr_mode=False)