Update app.py

app.py

@@ -6,173 +6,166 @@ from PIL import Image
 import pytesseract
 import fitz
 import pathlib
-
+import threading
+import time
 
 def resolve_token(ui_token):
     if ui_token and ui_token.strip():
         return ui_token.strip()
-
-    env_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
+    env_token = os.getenv("hf")
     if env_token:
         return env_token.strip()
-
     return None
 
-
 SUPPORTED_EXT = (
     ".pdf", ".docx", ".txt", ".png", ".jpg", ".jpeg", ".webp", ".bmp", ".tiff"
 )
 
-
 def extract_text_from_file(filepath):
     if not filepath:
         return ""
-
     if isinstance(filepath, dict) and "name" in filepath:
         filepath = filepath["name"]
-
     ext = pathlib.Path(filepath).suffix.lower()
-
     try:
         if ext == ".pdf":
             doc = fitz.open(filepath)
-            return "\n".join([page.get_text() for page in doc])
-
+            return "\n".join(page.get_text() for page in doc)
         elif ext == ".docx":
             doc = Document(filepath)
-            return "\n".join([p.text for p in doc.paragraphs])
-
+            return "\n".join(p.text for p in doc.paragraphs)
         elif ext == ".txt":
             with open(filepath, "r", encoding="utf-8", errors="ignore") as f:
                 return f.read()
-
         elif ext in (".png", ".jpg", ".jpeg", ".webp", ".bmp", ".tiff"):
             img = Image.open(filepath)
             return pytesseract.image_to_string(img)
-
         else:
             return "Unsupported file type"
-
     except Exception as e:
         return f"Error reading file: {str(e)}"
 
-
 MODELS = {
-    "Qwen 3.5 0.8B (Fastest)": "Qwen/Qwen3.5-0.8B",
-    "DeepSeek R1 1.5B": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
-    "Qwen 3.5 2B (Balanced Speed)": "Qwen/Qwen3.5-2B",
-    "Gemma 4 5B": "google/gemma-4-E2B-it",
-    "Qwen 3.5 4B": "Qwen/Qwen3.5-4B",
-    "Gemma 4 9B": "google/gemma-4-9b-it",
-    "DeepSeek Qwen 7B": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
-    "DeepSeek R1 Llama 8B": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
-    "Gemma 8B": "google/gemma-4-E4B-it",
-    "Qwen 3.5 9B (Balanced Thinking)": "Qwen/Qwen3.5-9B",
-    "Qwen 3.6 27B (Best)": "Qwen/Qwen3.6-27B",
+    "DeepSeek-Qwen 1.5B (Fastest)": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
+    "Qwen 3 4B (Good Speed)": "Qwen/Qwen3-4B-Thinking-2507:nscale",
+    "Gemma 3 4B": "google/gemma-3n-E4B-it:together",
+    "DeepSeek-Llama 8B (Recommended)": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B:nscale",
+    "Qwen 3 8B": "Qwen/Qwen3-8B:nscale",
+    "OpenAI GPT OSS 120B": "openai/gpt-oss-120b:novita",
+    "Gemma 4 31B (Best)": "google/gemma-4-31B-it:novita"
 }
 
 ALL_MODEL_NAMES = list(MODELS.keys())
 
-SYSTEM_MSG = """You are a precise educational assistant.
-Follow instructions strictly.
-Use simple language.
-Be structured.
-Avoid repetition.
-Output markdown.
-"""
-
+SYSTEM_MSG = """You are an expert Academic Instructor and Curriculum Designer, highly familiar with SASTRA University's academic structure.
+- STRICTLY adhere to the provided syllabus/topic. Do not invent concepts, formulas, or fake exam papers.
+- Use clean, structured Markdown. Enforce exact formatting where specified.
+- Keep outputs practical, concise, and immediately usable. Avoid overcomplication.
+- SASTRA RESOURCES: Actively identify and recommend SASTRA-specific resources. Mention SASTRA Material Hub, Material Base, previous year CIA/End-Sem papers, and standard textbooks prescribed by SASTRA faculty. Frame teaching points and questions to match SASTRA's typical assessment patterns (CIA 1, CIA 2, End Semester).
+- GENERAL RESOURCES: If a specific standard textbook or verified platform (like NPTEL, standard authors) is widely used for this topic, cite it accurately.
+- NEVER hallucinate citations or paper questions. If unsure about a specific SASTRA paper, generalize based on typical university patterns while maintaining the SASTRA difficulty curve."""
 
-def make_prompts(topic):
+def make_prompts(syllabus):
     base = f"""
-Topic:
-{topic}
-Instructions:
-- Be clear and concise
-- Use bullet points
-- Avoid long paragraphs
+Academic Context:
+{syllabus}
+
+General Requirements:
+- Maintain strict alignment with the syllabus
+- Format strictly in Markdown
+- Keep explanations concise, practical, and exam/teaching ready
+- Cite SASTRA-specific resources (Material Hub, Material Base, previous papers) or widely recognized academic platforms
 """
-
     return {
         "lesson": base + """
-Create a lesson plan with:
-1. Objectives (3-5 points)
-2. Short introduction
-3. Key concepts (bullet points)
-4. 2 simple examples
-5. 1 case study (short)
-6. 1 classroom activity
-7. 5 assessment questions
+Generate a period-wise teaching schedule (Lesson Plan):
+- Assume 10-12 periods of 45-50 minutes each.
+- Output MUST be a single Markdown table with exactly these columns:
+  | Period | Topic/Subtopic | Key Teaching Points | SASTRA Resource/Activity Hint |
+- In the 'SASTRA Resource/Activity Hint' column, specify relevant SASTRA Material Hub modules, standard prescribed book chapters, or typical CIA question types for that topic.
+- Keep descriptions brief. Focus on logical progression.
+- Reserve the final period for SASTRA end-semester pattern revision.
+- Do not add any introductory or concluding text outside the table.
 """,
         "qa": base + """
-Generate 10 short exam questions with answers.
-Keep answers brief (2-3 lines each).
+Generate 8 exam-style short questions with precise answers:
+- Mirror SASTRA CIA and End-Semester phrasing and difficulty.
+- Cover definitions, direct applications, derivations, and 1-step reasoning.
+- Format strictly:
+  **Q1.** [Question]
+  **A:** [Concise answer, max 2 lines]
+- If the question matches a known pattern from SASTRA Material Base or a prescribed textbook exercise, note it briefly at the end of the answer.
 """,
         "mcq": base + """
-Generate 10 MCQs:
-- 4 options each
-- mark correct answer
+Generate 8 multiple-choice questions:
+- 4 options (A-D), one clearly correct, three plausible distractors.
+- Format strictly:
+  **Q1.** [Stem]
+  A) ... B) ... C) ... D) ...
+  **Correct:** [Letter] | **Why:** [1-line rationale]
+- Focus on core concepts, formula application, and common student mistakes seen in SASTRA exams.
 """,
         "summary": base + """
-Write a 200-word summary.
-Use simple sentences.
+Produce a rapid-revision summary:
+- List 5 core concepts/formulas
+- Provide 3 high-yield takeaways for SASTRA exams
+- Include 1 quick self-check question based on typical CIA patterns
+- Keep strictly under 150 words. Use bullet points only. No paragraphs.
 """
     }
 
-
 def stream_llm(model_id, prompt, hf_token):
     if not hf_token:
         yield "❌ No Hugging Face API key found."
         return
-
     try:
-        client = InferenceClient(model=model_id, token=hf_token)
-
-        stream = client.text_generation(
-            SYSTEM_MSG + "\n\n" + prompt,
-            max_new_tokens=1024,
+        client = InferenceClient(token=hf_token)
+        stream = client.chat.completions.create(
+            model=model_id,
+            messages=[
+                {"role": "system", "content": SYSTEM_MSG},
+                {"role": "user", "content": prompt},
+            ],
+            max_tokens=2048,
             temperature=0.7,
             top_p=0.95,
-            repetition_penalty=1.1,
             stream=True,
         )
-
         partial = ""
         for chunk in stream:
-            if hasattr(chunk, "token"):
-                token_text = chunk.token.text
-            else:
-                token_text = str(chunk)
-
-            partial += token_text
-            yield partial
-
+            if chunk.choices and chunk.choices[0].delta:
+                token = chunk.choices[0].delta.get("content", "")
+                if token:
+                    partial += token
+                    yield partial
     except Exception as e:
-        err = str(e)
-        if "401" in err or "Unauthorized" in err:
-            yield f"❌ Invalid Hugging Face API Key {err}"
-        else:
-            yield f"❌ API Error:\n{err}"
-
+        yield f"❌ API Error:\n{str(e)}"
 
 def generate_content(text, file, model_label, token):
+    hf_token = resolve_token(token)
     file_text = extract_text_from_file(file) if file else ""
     syllabus = (text + "\n\n" + file_text).strip()
-
     if not syllabus:
         yield ("Provide topic or file", "", "", "")
         return
-
     prompts = make_prompts(syllabus)
     model_id = MODELS[model_label]
-
     outputs = ["", "", "", ""]
     keys = ["lesson", "qa", "mcq", "summary"]
-
+    lock = threading.Lock()
+    def run_stream(i, key):
+        for chunk in stream_llm(model_id, prompts[key], hf_token):
+            with lock:
+                outputs[i] = chunk
+    threads = []
     for i, key in enumerate(keys):
-        for chunk in stream_llm(model_id, prompts[key], resolve_token(token)):
-            outputs[i] = chunk
-            yield tuple(outputs)
-
+        t = threading.Thread(target=run_stream, args=(i, key))
+        t.start()
+        threads.append(t)
+    while any(t.is_alive() for t in threads):
+        time.sleep(0.1)
+        yield tuple(outputs)
+    yield tuple(outputs)
 
 CSS = """
 body,.gradio-container{
@@ -180,48 +173,29 @@ body,.gradio-container{
 }
 """
 
-with gr.Blocks(css=CSS) as demo:
+with gr.Blocks() as demo:
     gr.Markdown("# 🎓 AI Study Material Generator (Streaming)")
-
     with gr.Row():
         with gr.Column():
-            text_input = gr.Textbox(
-                placeholder="Paste syllabus or topic",
-                lines=6
-            )
-            file_input = gr.File(label="Upload syllabus file")
-
+            text_input = gr.Textbox(placeholder="Paste syllabus or topic", lines=6)
+            file_input = gr.File(label="Upload syllabus file", file_types=list(SUPPORTED_EXT))
         with gr.Column():
-            model_selector = gr.Dropdown(
-                choices=ALL_MODEL_NAMES,
-                value=ALL_MODEL_NAMES[0],
-                label="Model"
-            )
-
-            token_box = gr.Textbox(
-                label="HF API Key (optional)",
-                type="password"
-            )
-
+            model_selector = gr.Dropdown(choices=ALL_MODEL_NAMES, value="DeepSeek-Llama 8B (Recommended)", label="Model")
+            token_box = gr.Textbox(label="HF API Key (optional)", type="password")
             btn = gr.Button("Generate")
-
     with gr.Tabs():
         with gr.TabItem("Lesson Plan"):
             lesson = gr.Markdown()
-
-        with gr.TabItem("Q&A"):
+        with gr.TabItem("Question and Anwer"):
             qa = gr.Markdown()
-
         with gr.TabItem("MCQ"):
             mcq = gr.Markdown()
-
         with gr.TabItem("Summary"):
             summary = gr.Markdown()
-
     btn.click(
         fn=generate_content,
         inputs=[text_input, file_input, model_selector, token_box],
         outputs=[lesson, qa, mcq, summary]
     )
 
-demo.launch()
+demo.launch(css=CSS)