Update app.py

app.py

@@ -3,16 +3,36 @@ import gradio as gr
 from transformers import pipeline, AutoTokenizer
 import torch
 import spaces
+import json
+from huggingface_hub import HfApi, upload_file
 
-# Shared dropdown options
+# --- Constants ---
+HF_TOKEN = os.environ.get("HF_TOKEN")
+DATASET_REPO = "Pisethan/khmer-lesson-dataset-generated"
+LOCAL_JSONL = "generated_lessons.jsonl"
+
+# --- Options ---
 grade_options = ["1", "2", "3", "4", "5", "6"]
 topic_options = ["Addition", "Subtraction", "Counting", "Number Recognition", "Multiplication", "Division"]
 level_options = ["Beginner", "Intermediate", "Advanced"]
 
-# Load tokenizer separately so we can access eos_token_id
-HF_TOKEN = os.environ.get("HF_TOKEN")
+# --- Tokenizer (global) ---
 tokenizer = AutoTokenizer.from_pretrained("Pisethan/khmer-lesson-model", token=HF_TOKEN)
 
+# --- Helper to save and upload ---
+def save_to_jsonl(record):
+    with open(LOCAL_JSONL, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, ensure_ascii=False) + "\n")
+
+    upload_file(
+        path_or_fileobj=LOCAL_JSONL,
+        path_in_repo="generated_lessons.jsonl",
+        repo_id=DATASET_REPO,
+        repo_type="dataset",
+        token=HF_TOKEN
+    )
+
+# --- Generation for one lesson ---
 @spaces.GPU
 def generate_lesson(grade, topic, level):
     device = 0 if torch.cuda.is_available() else -1
@@ -41,8 +61,20 @@ TaRL Level: {level}
 """
 
     output = pipe(prompt, max_new_tokens=300, temperature=0.7, do_sample=True, eos_token_id=tokenizer.eos_token_id)
-    return output[0]['generated_text']
-
+    result = output[0]['generated_text']
+
+    # Save to dataset
+    record = {
+        "grade": grade,
+        "topic": topic,
+        "level": level,
+        "prompt": prompt.strip(),
+        "completion": result.strip()
+    }
+    save_to_jsonl(record)
+    return result
+
+# --- Generation for all combinations ---
 @spaces.GPU
 def generate_all_lessons():
     device = 0 if torch.cuda.is_available() else -1
@@ -64,10 +96,21 @@ Grade: {grade}
 Topic: {topic}
 TaRL Level: {level}"""
                 output = pipe(prompt, max_new_tokens=200, temperature=0.7, do_sample=True)
-                results += f"🔹 ថ្នាក់ {grade} | {topic} | {level}\n{output[0]['generated_text']}\n\n{'-'*50}\n\n"
+                result = output[0]['generated_text']
+
+                record = {
+                    "grade": grade,
+                    "topic": topic,
+                    "level": level,
+                    "prompt": prompt.strip(),
+                    "completion": result.strip()
+                }
+                save_to_jsonl(record)
+
+                results += f"🔹 ថ្នាក់ {grade} | {topic} | {level}\n{result}\n\n{'-'*50}\n\n"
     return results
 
-# Gradio UI
+# --- UI ---
 with gr.Blocks() as demo:
     gr.Markdown("## 🤖 អ្នកជំនួយបង្កើតមេរៀនគណិតវិទ្យា")
     gr.Markdown("ជ្រើសរើសថ្នាក់ ប្រធានបទ និងកម្រិតសិស្ស រួចចុចបង្កើតមេរៀន។ ឬចុចប៊ូតុងខាងក្រោមសម្រាប់បង្កើតមេរៀនទាំងអស់។")