update

app.py

@@ -1,70 +1,225 @@
+import os
+import csv
+import uuid
+from datetime import datetime
+import torch
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
-    """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-    """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
-
-    messages = [{"role": "system", "content": system_message}]
-
-    messages.extend(history)
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-chatbot = gr.ChatInterface(
-    respond,
-    type="messages",
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
+from filelock import FileLock
+from huggingface_hub import HfApi
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+from peft import PeftModel
+
+# =========================
+# ⚙️ Config
+# =========================
+MAX_HISTORY_TURNS = 10
+MAX_PROMPT_TOKENS = 1024
+MAX_NEW_TOKENS = 60
+
+LOG_DIR = "/tmp/chat_logs"
+os.makedirs(LOG_DIR, exist_ok=True)
+LOCK_PATH = os.path.join(LOG_DIR, ".lock")
+
+HF_TOKEN = os.environ.get("HF_TOKEN")
+PRIVATE_LOG_REPO = "hparten/math_chat_logs"  # Private dataset repo
+HF_API = HfApi()
+
+MODEL_ID = "hparten/prob1_qlora_math_student"
+
+# =========================
+# 🔠 Model + Tokenizer
+# =========================
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+tokenizer.pad_token = tokenizer.eos_token
+
+pipe = pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    dtype=torch.float16,
+    device_map="auto",
 )
 
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
+# =========================
+# 🧩 Strategy Explanations
+# =========================
+strategy_explanations = {
+    "friendly": "You add on from 41 until you get to 84, usually by counting by 10s, 20s, or 40, then ones.",
+    "differencing": "You difference the ones or tens separately during any part of your answer.",
+    "subtraction": "You turn the problem into a subtraction: 84 minus 41 equals blank to find the missing addend.",
+}
+
+# =========================
+# 🧠 System Prompt
+# =========================
+def build_system_block(problem_prefix, strategy):
+    problem_text = "41 plus blank equals 84"
+    strat_key = strategy.lower()
+    strat_expl = strategy_explanations.get(strat_key, "Use the named strategy to explain your steps clearly.")
+    strategy_tag = f"<strategy_{strat_key}>"
+    problem_tag = f"<{problem_prefix.lower()}>"
+
+    system_text = (
+        f"<system>\n"
+        f"You are the student in a math dialogue.\n"
+        f"PROBLEM: {problem_tag} - {problem_text}\n"
+        f"STRATEGY: {strategy_tag} — {strat_expl}\n"
+        f"When you answer, think step by step, like a student explaining their work out loud.\n"
+        f"Keep your answers short and natural—1 sentence. Let the teacher ask follow-up questions.\n"
+        f"Reply exactly to the teacher questions using <student> ... </student>. Never include any teacher text in your answer.\n"
+        f"</system>\n"
+    )
+    return system_text.strip()
+
+# =========================
+# 🧾 Logging (Private Upload)
+# =========================
+CSV_HEADERS = ["timestamp", "session_id", "username", "strategy", "teacher", "student"]
+
+def log_turn(session_id, username, strategy, teacher_msg, student_msg):
+    path = os.path.join(LOG_DIR, f"chat_{session_id}.csv")
+    file_exists = os.path.exists(path)
+
+    with FileLock(LOCK_PATH):
+        with open(path, "a", newline="", encoding="utf-8") as f:
+            writer = csv.writer(f)
+            if not file_exists:
+                writer.writerow(CSV_HEADERS)
+            writer.writerow([
+                datetime.now().isoformat(timespec="seconds"),
+                session_id,
+                username,
+                strategy,
+                teacher_msg,
+                student_msg,
+            ])
+
+    # --- Try uploading to private dataset repo ---
+    try:
+        HF_API.upload_file(
+            path_or_fileobj=path,
+            path_in_repo=f"{os.path.basename(path)}",
+            repo_id=PRIVATE_LOG_REPO,
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+        print(f"✅ Uploaded log to private dataset: {PRIVATE_LOG_REPO}")
+    except Exception as e:
+        print(f"⚠️ Could not push log: {e}")
+
+# =========================
+# 🧩 Prompt Builder
+# =========================
+def build_prompt(strategy, history, teacher_question, tokenizer, problem_prefix="Problem_1"):
+    base_system_prompt = build_system_block(problem_prefix, strategy)
+    turns = [f"<teacher> {tq} </teacher> <student> {sa} </student>" for tq, sa in history[-MAX_HISTORY_TURNS:]]
+    full_prompt = base_system_prompt + "\n" + " ".join(turns)
+    full_prompt += f"<teacher> {teacher_question} </teacher>\n"
+
+    while len(tokenizer.encode(full_prompt, add_special_tokens=False)) > MAX_PROMPT_TOKENS and len(turns) > 0:
+        turns.pop(0)
+        convo_block = " ".join(turns)
+        full_prompt = base_system_prompt + convo_block + f"<teacher> {teacher_question} </teacher>"
+
+    return full_prompt.strip()
+
+# =========================
+# 🤖 Generation
+# =========================
+def generate_response(teacher_question, username, history, session_id, strategy):
+    prompt = build_prompt(strategy, history, teacher_question, tokenizer)
+    out = pipe(
+        prompt,
+        max_new_tokens=MAX_NEW_TOKENS,
+        do_sample=True,
+        temperature=0.5,
+        top_p=0.9,
+        repetition_penalty=1.05,
+        pad_token_id=tokenizer.pad_token_id or tokenizer.eos_token_id,
+        return_full_text=False,
+    )
+    out_text = out[0]["generated_text"]
+
+    if "<student>" in out_text and "</student>" in out_text:
+        student_reply = out_text.split("<student>", 1)[1].split("</student>", 1)[0].strip()
+    else:
+        student_reply = out_text.strip()
+
+    history.append((teacher_question, student_reply))
+    log_turn(session_id, username, strategy, teacher_question, student_reply)
+    return student_reply, history
+
+# =========================
+# 🖥 Gradio UI
+# =========================
+def on_send(teacher_question, username, strategy_choice, history, session_id):
+    if not session_id:
+        session_id = uuid.uuid4().hex[:12]
+    if history is None:
+        history = []
+    if not username.strip():
+        gr.Warning("Please enter your name before starting the chat.")
+        return history, history, "", session_id
+    if not teacher_question.strip():
+        gr.Warning("Please type a question for the student before sending.")
+        return history, history, "", session_id
+
+    student_reply, history = generate_response(
+        teacher_question.strip(),
+        username.strip(),
+        history,
+        session_id,
+        strategy_choice.lower(),
+    )
+
+    msgs = []
+    for t, s in history[-MAX_HISTORY_TURNS:]:
+        msgs.append({"role": "user", "content": t})
+        msgs.append({"role": "assistant", "content": s})
+
+    return msgs, history, "", session_id
+
+def on_reset():
+    return [], [], "", uuid.uuid4().hex[:12]
+
+# =========================
+# 🚀 Gradio App
+# =========================
+with gr.Blocks(title="Elementary Math Student Chatbot") as demo:
+    gr.Markdown("## 🧮 Practice Eliciting Student Thinking (Prototype)")
+    gr.Markdown(
+        "You are an elementary math teacher exploring a student's reasoning for **41 + ___ = 84**.\n"
+        "Ask questions and see how the student explains their thinking."
+    )
+
+    with gr.Row():
+        username = gr.Textbox(label="👤 Your Name", placeholder="Enter your name...")
+        strategy_choice = gr.Dropdown(
+            ["friendly", "differencing", "subtraction"],
+            value="friendly",
+            label="🧩 Student Strategy",
+        )
+        reset_btn = gr.Button("🔄 Start Over", variant="secondary")
+
+    teacher_q = gr.Textbox(label="👩‍🏫 Teacher Question", placeholder="Ask the student a question…")
+    chat = gr.Chatbot(label="💬 Chat", type="messages")
+    state_history = gr.State([])
+    state_session = gr.State("")
+    send = gr.Button("Send", variant="primary")
+
+    send.click(
+        on_send,
+        inputs=[teacher_q, username, strategy_choice, state_history, state_session],
+        outputs=[chat, state_history, teacher_q, state_session],
+    )
 
+    reset_btn.click(
+        on_reset,
+        inputs=[],
+        outputs=[chat, state_history, teacher_q, state_session],
+    )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue()
+    demo.launch()

requirements.txt

@@ -0,0 +1,8 @@
+torch>=2.1.0
+transformers>=4.44.0
+peft>=0.10.0
+accelerate>=0.30.0
+bitsandbytes>=0.43.0
+huggingface_hub>=0.23.0
+gradio>=4.40.0
+filelock>=3.12.0