Upload 12 files

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+faiss_index/index.faiss filter=lfs diff=lfs merge=lfs -text
+SQL_s.docx filter=lfs diff=lfs merge=lfs -text

Retrieval.ipynb

@@ -0,0 +1,588 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "aed563e2",
+   "metadata": {},
+   "source": [
+    "# Retrieval"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "3c99cd9f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain_community.vectorstores import FAISS\n",
+    "from langchain_ollama import OllamaEmbeddings\n",
+    "from langchain_ollama import ChatOllama\n",
+    "from langchain_core.prompts import PromptTemplate"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "002e9bea",
+   "metadata": {},
+   "source": [
+    "#### Load Vectore_store"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "20eb543e",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
+   "id": "93a1dad2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "embedding = OllamaEmbeddings(model=\"nomic-embed-text\")\n",
+    "\n",
+    "vector_store = FAISS.load_local(\n",
+    "    \"faiss_index\",\n",
+    "    embedding,\n",
+    "    allow_dangerous_deserialization=True\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 22,
+   "id": "2aa58c7c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "retriever = vector_store.as_retriever(search_type=\"similarity\", search_kwargs={\"k\": 1})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 23,
+   "id": "4ddc68fb",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "VectorStoreRetriever(tags=['FAISS', 'OllamaEmbeddings'], vectorstore=<langchain_community.vectorstores.faiss.FAISS object at 0x000002832218BD10>, search_kwargs={'k': 1})"
+      ]
+     },
+     "execution_count": 23,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "retriever"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "id": "af9e08ab",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[Document(id='d19b5983-6c58-4baa-9878-d1edcc1330f0', metadata={'topic': 'Distinct and Count Statement', 'question_no': 1}, page_content='1.Identify count of orders by counting ORDER_ID from ORDERS table. To count order_Id from table, you can write the following statement.\\n\\n\\n\\nselect \\n\\n count(order_id) \\n\\n from orders')]"
+      ]
+     },
+     "execution_count": 24,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "retriever.invoke('Identify count of orders by counting ORDER_ID from ORDERS table. To count order_Id from table, you can write the following statement')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "18a0d111",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "id": "18e9e4bd",
+   "metadata": {},
+   "source": [
+    "# Step 3 - Augmentation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 45,
+   "id": "bd69d69e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "llm = ChatOllama(\n",
+    "    model=\"llama3.2\",\n",
+    "    temperature=0\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 46,
+   "id": "e4cef64b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "prompt = PromptTemplate(\n",
+    "    template=\"\"\"\n",
+    "You are an SQL assistant.\n",
+    "\n",
+    "Use ONLY the provided context.\n",
+    "\n",
+    "Rules:\n",
+    "1. If the SQL query exists in the context, return ONLY that SQL query.\n",
+    "2. Do not explain.\n",
+    "3. Do not rewrite the question.\n",
+    "4. Do not generate a new query.\n",
+    "5. If the SQL query is missing, reply exactly:\n",
+    "I don't know.\n",
+    "\n",
+    "Context:\n",
+    "{context}\n",
+    "\n",
+    "Question:\n",
+    "{question}\n",
+    "\n",
+    "SQL:\n",
+    "\"\"\",\n",
+    "    input_variables=[\"context\", \"question\"],\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 47,
+   "id": "708dbc11",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "question = \"You can see that counting customer_id is going to give you the same result. It means that even if you use count(*), you are going to get the same number of rows. Try using COUNT(*) this time to count the number of rows in the table \"\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 48,
+   "id": "dba263ce",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "context_text = retriever.invoke(question)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 49,
+   "id": "0f98f8d6",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[Document(id='579858b1-f5c5-475c-8958-b16098c1a20b', metadata={'topic': 'Distinct and Count Statement', 'question_no': 3}, page_content='3.You can see that counting customer_id is going to give you the same result. It means that even if you use count(*), you are going to get the same number of rows. Try using COUNT(*) this time to count the number of rows in the table\\n\\n\\n\\n select \\n\\n count(*) \\n\\n from orders')]"
+      ]
+     },
+     "execution_count": 49,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "context_text"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 50,
+   "id": "d5878f7b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "final_prompt = prompt.invoke({\"context\": context_text, \"question\": question})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a31f0f2e",
+   "metadata": {},
+   "source": [
+    "# Step 4 - Generation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 51,
+   "id": "9166fe55",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "select count(*) from orders\n"
+     ]
+    }
+   ],
+   "source": [
+    "answer = llm.invoke(final_prompt)\n",
+    "print(answer.content)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ffae664d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "prompt= PromptTemplate(\n",
+    "    template=\"\"\"you have to extract the SQL query from the image and all the other dat from the  . If the SQL query is not present in the context, return \"I don't know.\" Do not provide any explanations or additional information.\n",
+    "    \n",
+    "                                  \"\"\"\n",
+    "    \n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "13b8ce68",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "response = chat(\n",
+    "        model=\"qwen2.5vl:7b\",\n",
+    "        messages=[\n",
+    "            {\n",
+    "                \"role\": \"user\",\n",
+    "                \"content\": prompt,\n",
+    "                \"images\": [image_path]   # User uploaded image\n",
+    "            }\n",
+    "        ]\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "deec7ff4",
+   "metadata": {},
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6a8080fa",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b020203d",
+   "metadata": {},
+   "source": [
+    "# Building a Chain"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 52,
+   "id": "f31cfbd8",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain_core.runnables import RunnableParallel, RunnablePassthrough, RunnableLambda\n",
+    "from langchain_core.output_parsers import StrOutputParser"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 53,
+   "id": "3580a60a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "parallel_chain = RunnableParallel({\n",
+    "    'context': retriever ,\n",
+    "    'question': RunnablePassthrough()\n",
+    "})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 55,
+   "id": "277f666d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "parser = StrOutputParser()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 56,
+   "id": "aec127e2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "main_chain = parallel_chain | prompt | llm | parser"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 58,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'select count(*) from orders'"
+      ]
+     },
+     "execution_count": 58,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "question = \"You can see that counting customer_id is going to give you the same result. It means that even if you use count(*), you are going to get the same number of rows. Try using COUNT(*) this time to count the number of rows in the table \"\n",
+    "main_chain.invoke(question)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f4372959",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1956a2ec",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 59,
+   "id": "f9c5db7b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain_core.prompts import ChatPromptTemplate\n",
+    "from langchain_core.messages import HumanMessage\n",
+    "\n",
+    "# Define the extraction instructions\n",
+    "system_instructions = \"\"\"\n",
+    "Act as an expert OCR and data extraction assistant. I will provide you with an image of a user asking a coding question on a learning platform.\n",
+    "\n",
+    "Please format your response clearly using the following headings:\n",
+    "### User Statement\n",
+    "Extract the introductory text written by the user at the very top of the image.\n",
+    "\n",
+    "### Question Description\n",
+    "Extract the problem statement from the left-hand panel under the \"Description\" tab.\n",
+    "\n",
+    "### SQL Query\n",
+    "Extract the exact SQL code written inside the dark-themed code editor on the right. Preserve the line breaks and formatting.\n",
+    "\n",
+    "### Error Message\n",
+    "Extract the red error message displayed in the terminal/result area at the bottom right.\n",
+    "\"\"\"\n",
+    "\n",
+    "# Create the ChatPromptTemplate for multimodal input\n",
+    "multimodal_prompt = ChatPromptTemplate.from_messages([\n",
+    "    (\"system\", system_instructions),\n",
+    "    (\"user\", [\n",
+    "        {\"type\": \"text\", \"text\": \"Please extract the requested data from this image.\"},\n",
+    "        {\"type\": \"image_url\", \"image_url\": \"data:image/png;base64,{image_base64}\"}\n",
+    "    ])\n",
+    "])\n",
+    "\n",
+    "# Example usage (assuming 'image_base64' is your base64 encoded string of the image):\n",
+    "# chain = multimodal_prompt | vision_llm\n",
+    "# result = chain.invoke({\"image_base64\": base64_string})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 61,
+   "id": "9f054cc3",
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "ValidationError",
+     "evalue": "1 validation error for Message\ncontent\n  Input should be a valid string [type=string_type, input_value=ChatPromptTemplate(input_... additional_kwargs={})]), input_type=ChatPromptTemplate]\n    For further information visit https://errors.pydantic.dev/2.13/v/string_type",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mValidationError\u001b[39m                           Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[61]\u001b[39m\u001b[32m, line 2\u001b[39m\n\u001b[32m      1\u001b[39m \u001b[38;5;28;01mfrom\u001b[39;00m ollama \u001b[38;5;28;01mimport\u001b[39;00m chat\n\u001b[32m----> \u001b[39m\u001b[32m2\u001b[39m response = chat(\n\u001b[32m      3\u001b[39m         model=\u001b[33m\"qwen2.5vl:7b\"\u001b[39m,\n\u001b[32m      4\u001b[39m         messages=[\n\u001b[32m      5\u001b[39m             {\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\p4pri\\anaconda3\\envs\\langchain311\\Lib\\site-packages\\ollama\\_client.py:393\u001b[39m, in \u001b[36mClient.chat\u001b[39m\u001b[34m(self, model, messages, tools, stream, think, logprobs, top_logprobs, format, options, keep_alive)\u001b[39m\n\u001b[32m    340\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mchat\u001b[39m(\n\u001b[32m    341\u001b[39m   \u001b[38;5;28mself\u001b[39m,\n\u001b[32m    342\u001b[39m   model: \u001b[38;5;28mstr\u001b[39m = \u001b[33m'\u001b[39m\u001b[33m'\u001b[39m,\n\u001b[32m   (...)\u001b[39m\u001b[32m    352\u001b[39m   keep_alive: Optional[Union[\u001b[38;5;28mfloat\u001b[39m, \u001b[38;5;28mstr\u001b[39m]] = \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[32m    353\u001b[39m ) -> Union[ChatResponse, Iterator[ChatResponse]]:\n\u001b[32m    354\u001b[39m \u001b[38;5;250m  \u001b[39m\u001b[33;03m\"\"\"\u001b[39;00m\n\u001b[32m    355\u001b[39m \u001b[33;03m  Create a chat response using the requested model.\u001b[39;00m\n\u001b[32m    356\u001b[39m \n\u001b[32m   (...)\u001b[39m\u001b[32m    385\u001b[39m \u001b[33;03m  Returns `ChatResponse` if `stream` is `False`, otherwise returns a `ChatResponse` generator.\u001b[39;00m\n\u001b[32m    386\u001b[39m \u001b[33;03m  \"\"\"\u001b[39;00m\n\u001b[32m    387\u001b[39m   \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._request(\n\u001b[32m    388\u001b[39m     ChatResponse,\n\u001b[32m    389\u001b[39m     \u001b[33m'\u001b[39m\u001b[33mPOST\u001b[39m\u001b[33m'\u001b[39m,\n\u001b[32m    390\u001b[39m     \u001b[33m'\u001b[39m\u001b[33m/api/chat\u001b[39m\u001b[33m'\u001b[39m,\n\u001b[32m    391\u001b[39m     json=ChatRequest(\n\u001b[32m    392\u001b[39m       model=model,\n\u001b[32m--> \u001b[39m\u001b[32m393\u001b[39m       messages=\u001b[30;43mlist\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m_copy_messages\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mmessages\u001b[39;49m\u001b[30;43m)\u001b[39;49m\u001b[30;43m)\u001b[39;49m,\n\u001b[32m    394\u001b[39m       tools=\u001b[38;5;28mlist\u001b[39m(_copy_tools(tools)),\n\u001b[32m    395\u001b[39m       stream=stream,\n\u001b[32m    396\u001b[39m       think=think,\n\u001b[32m    397\u001b[39m       logprobs=logprobs,\n\u001b[32m    398\u001b[39m       top_logprobs=top_logprobs,\n\u001b[32m    399\u001b[39m       \u001b[38;5;28mformat\u001b[39m=\u001b[38;5;28mformat\u001b[39m,\n\u001b[32m    400\u001b[39m       options=options,\n\u001b[32m    401\u001b[39m       keep_alive=keep_alive,\n\u001b[32m    402\u001b[39m     ).model_dump(exclude_none=\u001b[38;5;28;01mTrue\u001b[39;00m),\n\u001b[32m    403\u001b[39m     stream=stream,\n\u001b[32m    404\u001b[39m   )\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\p4pri\\anaconda3\\envs\\langchain311\\Lib\\site-packages\\ollama\\_client.py:1323\u001b[39m, in \u001b[36m_copy_messages\u001b[39m\u001b[34m(messages)\u001b[39m\n\u001b[32m   1321\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34m_copy_messages\u001b[39m(messages: Optional[Sequence[Union[Mapping[\u001b[38;5;28mstr\u001b[39m, Any], Message]]]) -> Iterator[Message]:\n\u001b[32m   1322\u001b[39m   \u001b[38;5;28;01mfor\u001b[39;00m message \u001b[38;5;129;01min\u001b[39;00m messages \u001b[38;5;129;01mor\u001b[39;00m []:\n\u001b[32m-> \u001b[39m\u001b[32m1323\u001b[39m     \u001b[38;5;28;01myield\u001b[39;00m \u001b[30;43mMessage\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mmodel_validate\u001b[39;49m\u001b[30;43m(\u001b[39;49m\n\u001b[32m   1324\u001b[39m \u001b[30;43m      \u001b[39;49m\u001b[30;43m{\u001b[39;49m\u001b[30;43mk\u001b[39;49m\u001b[30;43m:\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mlist\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m_copy_images\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mv\u001b[39;49m\u001b[30;43m)\u001b[39;49m\u001b[30;43m)\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01mif\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mk\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m==\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m'\u001b[39;49m\u001b[30;43mimages\u001b[39;49m\u001b[30;43m'\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01melse\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mv\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01mfor\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mk\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mv\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01min\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mdict\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mmessage\u001b[39;49m\u001b[30;43m)\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mitems\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m)\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01mif\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mv\u001b[39;49m\u001b[30;43m}\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m   1325\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\p4pri\\anaconda3\\envs\\langchain311\\Lib\\site-packages\\pydantic\\main.py:732\u001b[39m, in \u001b[36mBaseModel.model_validate\u001b[39m\u001b[34m(cls, obj, strict, extra, from_attributes, context, by_alias, by_name)\u001b[39m\n\u001b[32m    726\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m by_alias \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mFalse\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m by_name \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mTrue\u001b[39;00m:\n\u001b[32m    727\u001b[39m     \u001b[38;5;28;01mraise\u001b[39;00m PydanticUserError(\n\u001b[32m    728\u001b[39m         \u001b[33m'\u001b[39m\u001b[33mAt least one of `by_alias` or `by_name` must be set to True.\u001b[39m\u001b[33m'\u001b[39m,\n\u001b[32m    729\u001b[39m         code=\u001b[33m'\u001b[39m\u001b[33mvalidate-by-alias-and-name-false\u001b[39m\u001b[33m'\u001b[39m,\n\u001b[32m    730\u001b[39m     )\n\u001b[32m--> \u001b[39m\u001b[32m732\u001b[39m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mcls\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m__pydantic_validator__\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mvalidate_python\u001b[39;49m\u001b[30;43m(\u001b[39;49m\n\u001b[32m    733\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mobj\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    734\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mstrict\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mstrict\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    735\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mextra\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mextra\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    736\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mfrom_attributes\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mfrom_attributes\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    737\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mcontext\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mcontext\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    738\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mby_alias\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mby_alias\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    739\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43mby_name\u001b[39;49m\u001b[30;43m=\u001b[39;49m\u001b[30;43mby_name\u001b[39;49m\u001b[30;43m,\u001b[39;49m\n\u001b[32m    740\u001b[39m \u001b[30;43m\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[31mValidationError\u001b[39m: 1 validation error for Message\ncontent\n  Input should be a valid string [type=string_type, input_value=ChatPromptTemplate(input_... additional_kwargs={})]), input_type=ChatPromptTemplate]\n    For further information visit https://errors.pydantic.dev/2.13/v/string_type"
+     ]
+    }
+   ],
+   "source": [
+    "from ollama import chat\n",
+    "response = chat(\n",
+    "        model=\"qwen2.5vl:7b\",\n",
+    "        messages=[\n",
+    "            {\n",
+    "                \"role\": \"user\",\n",
+    "                \"content\": multimodal_prompt,\n",
+    "                \"images\": ['sql_eorror.png']   # User uploaded image\n",
+    "            }\n",
+    "        ]\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "58284ff3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "response"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 62,
+   "id": "2c273a9c",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "### User Statement\n",
+      "No user statement is provided in the image.\n",
+      "\n",
+      "### Question Description\n",
+      "The SQL script attempts to alter a table named 'CAL' by setting a column as NOT NULL and adding a PRIMARY KEY constraint, followed by creating an index on that column. However, it encounters errors due to constraints related to nullability of columns.\n",
+      "\n",
+      "### SQL Query\n",
+      "```sql\n",
+      "IF NOT EXISTS(\n",
+      "    SELECT *\n",
+      "    FROM INFORMATION_SCHEMA.TABLE_CONSTRAINTS\n",
+      "    WHERE CONSTRAINT_TYPE = 'PRIMARY KEY' AND TABLE_NAME = 'CAL'\n",
+      ")\n",
+      "BEGIN\n",
+      "    DROP INDEX IF EXISTS CAL$01 ON dbo.CAL;\n",
+      "    \n",
+      "    ALTER TABLE BTS.dbo.CAL\n",
+      "    ALTER COLUMN Intern \n",
+      "        INT NOT NULL;\n",
+      "\n",
+      "    ALTER TABLE BTS.dbo.CAL\n",
+      "    ADD CONSTRAINT PK_CAL_Intern PRIMARY KEY (Intern);\n",
+      "\n",
+      "    CREATE INDEX CAL$01\n",
+      "    ON CAL (Intern);\n",
+      "END\n",
+      "```\n",
+      "\n",
+      "### Error Message\n",
+      "Msg 8111, Level 16, State 1, Line 111  \n",
+      "Cannot define PRIMARY KEY constraint on nullable column in table 'CAL'.  \n",
+      "\n",
+      "Msg 1750, Level 16, State 0, Line 111  \n",
+      "Could not create constraint or index. See previous errors.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from ollama import chat\n",
+    "\n",
+    "system_prompt = \"\"\"\n",
+    "Act as an expert OCR and data extraction assistant.\n",
+    "\n",
+    "Please format your response clearly using the following headings:\n",
+    "\n",
+    "### User Statement\n",
+    "Extract the introductory text written by the user at the very top of the image.\n",
+    "\n",
+    "### Question Description\n",
+    "Extract the problem statement from the left-hand panel under the Description tab.\n",
+    "\n",
+    "### SQL Query\n",
+    "Extract the exact SQL code written inside the dark-themed code editor on the right.\n",
+    "\n",
+    "### Error Message\n",
+    "Extract the red error message displayed at the bottom right.\n",
+    "\"\"\"\n",
+    "\n",
+    "response = chat(\n",
+    "    model=\"qwen2.5vl:7b\",\n",
+    "    messages=[\n",
+    "        {\n",
+    "            \"role\": \"system\",\n",
+    "            \"content\": system_prompt\n",
+    "        },\n",
+    "        {\n",
+    "            \"role\": \"user\",\n",
+    "            \"content\": \"Extract all requested information from this image.\",\n",
+    "            \"images\": [\"sql_eorror.png\"]\n",
+    "        }\n",
+    "    ]\n",
+    ")\n",
+    "\n",
+    "print(response.message.content)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ee298127",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "langchain311",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

SQL_s.docx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c12f86dcdfeed5535a737ba9085c0c24d58920b41d603a3e7db81a72e83476ff
+size 368494

app.py

@@ -0,0 +1,192 @@
+# import streamlit as st
+# from PIL import Image
+# import help 
+
+
+# st.set_page_config(page_title="Multimodal Chat", page_icon="🤖")
+
+# st.title(" Multimodal Chat")
+
+# # -------------------------
+# # User Inputs
+# # -------------------------
+# text_input = st.text_area(
+#     "Enter your question (Optional)",
+#     placeholder="Ask something..."
+# )
+
+# uploaded_image = st.file_uploader(
+#     "Upload an image (Optional)",
+#     type=["png", "jpg", "jpeg"]
+# )
+
+# image = None
+
+# if uploaded_image is not None:
+#     image = Image.open(uploaded_image)
+#     st.image(image, caption="Uploaded Image", width="stretch")
+
+# # -------------------------
+# # Submit Button
+# # -------------------------
+# if st.button("Submit"):
+
+#     # Check if at least one input is provided
+#     if not text_input and image is None:
+#         st.warning("Please provide either a text question or an image.")
+#         st.stop()
+
+#     # -------------------------
+#     # Case 1 : Text + Image
+#     # -------------------------
+#     if text_input and image is not None:
+#         st.success("Received both Text and Image")
+
+#         # Replace this with your model
+#         response = f"""
+#         Text:
+#         {text_input}
+
+#         Image received successfully.
+
+#         (Call your VLM here.)
+#         """
+
+#     # -------------------------
+#     # Case 2 : Only Text
+#     # -------------------------
+#     elif text_input:
+#         st.success("Received only Text")
+
+#         # Replace this with your LLM
+#         response = help.inital_step(text_input)
+
+#     # -------------------------
+#     # Case 3 : Only Image
+#     # -------------------------
+#     else:
+#         st.success("Received only Image")
+
+#         # Replace this with your Vision Model
+        
+#         import tempfile
+
+#         if uploaded_image is not None:
+
+#             with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp:
+#                 tmp.write(uploaded_image.getbuffer())
+#                 temp_path = tmp.name
+
+#             response = help.image_input(temp_path)
+
+
+
+#     st.subheader("Response")
+#     st.write(response)
+
+
+
+
+
+import streamlit as st
+from PIL import Image
+import tempfile
+import help
+
+st.set_page_config(page_title="Error Assistant", page_icon="🤖")
+
+st.title("🤖 SQL / Excel Error Assistant")
+
+# ----------------------------------------
+# Select Task
+# ----------------------------------------
+task = st.selectbox(
+    "Choose Task",
+    [
+        "SQL Error",
+        "Excel Error"
+    ]
+)
+
+# ----------------------------------------
+# User Input
+# ----------------------------------------
+text_input = st.text_area(
+    "Enter your question (Optional)",
+    placeholder="Ask something..."
+)
+
+uploaded_image = st.file_uploader(
+    "Upload Screenshot (Optional)",
+    type=["png", "jpg", "jpeg"]
+)
+
+image = None
+
+if uploaded_image:
+    image = Image.open(uploaded_image)
+    st.image(image, caption="Uploaded Image", width="stretch")
+
+# ----------------------------------------
+# Submit
+# ----------------------------------------
+if st.button("Submit"):
+
+    if not text_input and image is None:
+        st.warning("Please provide text or image.")
+        st.stop()
+
+    # ==========================
+    # SQL
+    # ==========================
+    if task == "SQL Error":
+
+        if text_input and image:
+            st.success("Received both Text and Image")
+
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp:
+                tmp.write(uploaded_image.getbuffer())
+                img_path = tmp.name
+
+            response = help.sql_text_image(text_input, img_path)
+
+        elif text_input:
+            st.success("Received only Text")
+            response = help.sql_text(text_input)
+
+        else:
+            st.success("Received only Image")
+
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp:
+                tmp.write(uploaded_image.getbuffer())
+                img_path = tmp.name
+
+            response = help.sql_image(img_path)
+
+    # ==========================
+    # Excel
+    # ==========================
+    else:
+
+        if text_input and image:
+            st.success("Received both Text and Image")
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp:
+                tmp.write(uploaded_image.getbuffer())
+                img_path = tmp.name
+
+            response = help.excel_text_image(text_input, img_path)
+
+        elif text_input:
+            st.success("Received only Text")
+            response = help.excel_text(text_input)
+
+        else:
+            st.success("Received only Image")
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp:
+                tmp.write(uploaded_image.getbuffer())
+                img_path = tmp.name
+
+            response = help.excel_image(img_path)
+
+    st.subheader("Response")
+    st.write(response)

faiss_index/index.faiss

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40a5dbcae30c6bcb7dba6651b5dcdf33863dbb571025bb284687ed3de8ee411a
+size 583725

faiss_index/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b3267176abd0e48b42fde9793bdd528a2c0db2d4c5d65150bfdcf4234e878cc
+size 105142

help.py

@@ -0,0 +1,541 @@
+from click import prompt
+from langchain_core.runnables import RunnableParallel
+from langchain_core.runnables import RunnablePassthrough
+from langchain_core.output_parsers import StrOutputParser
+from streamlit import text_input
+from langchain_community.vectorstores import FAISS
+from langchain_ollama import OllamaEmbeddings
+from langchain_ollama import ChatOllama
+from langchain_core.prompts import PromptTemplate
+from ollama import chat
+from langchain_core.runnables import RunnableParallel 
+
+#### ************************************************************************************SQL***********************************************************************************
+
+def sql_text(text_input):
+    embedding = OllamaEmbeddings(model="nomic-embed-text")
+
+    vector_store = FAISS.load_local(
+        "faiss_index",
+        embedding,
+        allow_dangerous_deserialization=True
+    )
+
+    retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 1})
+
+
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+
+    prompt = PromptTemplate(
+        template="""
+    You are an SQL assistant.
+
+    Use ONLY the provided context.
+
+    Rules:
+    1. If the SQL query exists in the context, return ONLY that SQL query.
+    2. Do not explain.
+    3. Do not rewrite the question.
+    4. Do not generate a new query.
+    5. If the SQL query is missing, reply exactly:
+    I don't know.
+
+    Context:
+    {context}
+
+    Question:
+    {question}
+
+    SQL:
+    """,
+        input_variables=["context", "question"],
+    )
+
+
+    parallel_chain = RunnableParallel({
+        'context': retriever ,
+        'question': RunnablePassthrough()
+    })
+
+    parser = StrOutputParser()
+
+    main_chain = parallel_chain | prompt | llm | parser
+
+
+    response = main_chain.invoke(text_input)
+    return response
+
+
+
+
+
+
+# def image_input( image_path):
+
+#     system_prompt = """
+#     Act as an expert OCR and data extraction assistant.
+
+#     Please format your response clearly using the following headings:
+
+#     ### User Statement
+#     Extract the introductory text written by the user at the very top of the image.
+
+#     ### Question Description
+#     Extract the problem statement from the left-hand panel under the Description tab.
+
+#     ### SQL Query
+#     Extract the exact SQL code written inside the dark-themed code editor on the right.
+
+#     ### Error Message
+#     Extract the red error message displayed at the bottom right.
+#     """
+
+#     response = chat(
+#         model="qwen2.5vl:7b",
+#         messages=[
+#             {
+#                 "role": "system",
+#                 "content": system_prompt
+#             },
+#             {
+#                 "role": "user",
+#                 "content": "Extract all requested information from this image.",
+#                 "images": ["sql_eorror.png"]
+#             }
+#         ]
+#     )
+
+#     response_image_test =response.message.content
+
+
+#     llm = ChatOllama(
+#         model="llama3.2",
+#         temperature=0
+#     )
+
+
+#     embedding = OllamaEmbeddings(model="nomic-embed-text")
+
+#     vector_store = FAISS.load_local(
+#         "faiss_index",
+#         embedding,
+#         allow_dangerous_deserialization=True
+#     )
+
+#     retriever = vector_store.as_retriever(
+#         search_type="similarity",
+#         search_kwargs={"k": 1}
+#     )
+
+
+
+
+#     prompt = PromptTemplate(
+#         template="""
+#     You are an SQL assistant.
+
+#     Use ONLY the provided context.
+
+#     Rules:
+#     1. If the SQL query exists in the context, return ONLY that SQL query.
+#     2. Do not explain.
+#     3. Do not rewrite the question.
+#     4. Do not generate a new query.
+#     5. If the SQL query is missing, reply exactly:
+#     I don't know.
+
+#     Context:
+#     {context}
+
+#     Question:
+#     {question}
+
+#     SQL:
+#     """,
+#         input_variables=["context", "question"],
+#     )
+
+
+#     parallel_chain = RunnableParallel({
+#         'context': retriever ,
+#         'question': RunnablePassthrough()
+#     })
+
+#     parser = StrOutputParser()
+
+#     main_chain = parallel_chain | prompt | llm | parser
+
+
+#     response = main_chain.invoke(response_image_test)
+#     return response
+
+
+
+
+def sql_image(image_path):
+
+    # --------------------------------------------------
+    # STEP 1 : Extract ONLY the SQL question
+    # --------------------------------------------------
+    system_prompt = """
+You are an OCR assistant.
+
+Your job is to read the screenshot and extract ONLY the SQL problem statement.
+
+Rules:
+- Ignore the SQL editor.
+- Ignore the error message.
+- Ignore buttons.
+- Ignore the database output.
+- Ignore everything except the Question Description.
+
+Return ONLY the question text.
+Do not add headings.
+Do not explain anything.
+"""
+
+    response = chat(
+        model="qwen2.5vl:7b",
+        messages=[
+            {
+                "role": "system",
+                "content": system_prompt
+            },
+            {
+                "role": "user",
+                "content": "Extract only the SQL question.",
+                "images": [image_path]
+            }
+        ]
+    )
+
+    extracted_question = response.message.content.strip()
+
+    print("\nExtracted Question:\n")
+    print(extracted_question)
+
+    # --------------------------------------------------
+    # STEP 2 : Load LLM
+    # --------------------------------------------------
+
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+
+    # --------------------------------------------------
+    # STEP 3 : Load Embeddings
+    # --------------------------------------------------
+
+    embedding = OllamaEmbeddings(
+        model="nomic-embed-text"
+    )
+
+    vector_store = FAISS.load_local(
+        "faiss_index",
+        embedding,
+        allow_dangerous_deserialization=True
+    )
+
+    retriever = vector_store.as_retriever(
+        search_type="similarity",
+        search_kwargs={"k":3}
+    )
+
+    # --------------------------------------------------
+    # STEP 4 : Prompt
+    # --------------------------------------------------
+
+    prompt = PromptTemplate(
+        template="""
+You are an SQL assistant.
+
+Use ONLY the retrieved context.
+
+Rules:
+
+1. Return ONLY the SQL query.
+2. No explanation.
+3. No markdown.
+4. Do not rewrite the question.
+5. If the answer is not present in the context, reply exactly:
+
+I don't know.
+
+Context:
+{context}
+
+Question:
+{question}
+
+SQL:
+""",
+        input_variables=["context", "question"],
+    )
+
+    # --------------------------------------------------
+    # STEP 5 : Chain
+    # --------------------------------------------------
+
+    parallel_chain = RunnableParallel(
+        {
+            "context": retriever,
+            "question": RunnablePassthrough()
+        }
+    )
+
+    parser = StrOutputParser()
+
+    main_chain = (
+        parallel_chain
+        | prompt
+        | llm
+        | parser
+    )
+
+    # --------------------------------------------------
+    # STEP 6 : Retrieve using ONLY the extracted question
+    # --------------------------------------------------
+
+    response = main_chain.invoke(extracted_question)
+
+    return response
+
+
+
+
+
+def sql_text_image(question, image_path):
+    # --------------------------------------------------
+    # STEP 1 : Extract ONLY the SQL question from the image
+    # --------------------------------------------------
+    sql_answer1 = sql_image(image_path)
+    sql_answer2 = sql_text(question)
+
+    # --------------------------------------------------
+    # STEP 2 : Combine the extracted question and the text input
+    # --------------------------------------------------
+    # Build the Parallel Chain
+    prompt = f"""
+You are an SQL expert.
+
+Image SQL:
+{sql_answer1}
+
+User Question:
+{sql_answer2}
+
+Use BOTH pieces of information to produce the final SQL answer.
+
+Output exactly in this format:
+
+Kindly go through the provided information.
+<final_answer> -> By Prince Kushwaha
+
+Do not include any explanation or markdown.
+Replace <final_answer> with the actual SQL answer.
+"""
+    
+
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+    response = llm.invoke(prompt)
+
+    return response.content
+
+
+
+
+
+### *************************************************************************Excel**************************************************************************
+
+
+
+def excel_text(question):
+        # Load LLM
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+
+    # Prompt
+    prompt = PromptTemplate(
+        template="""
+    You are an Excel expert.
+
+    Answer the following Excel question.
+
+    Rules:
+    1. Return only the final answer.
+    2. If the answer is an Excel formula, return only the formula.
+    3. Do not use markdown.
+    4. Keep the answer concise.
+
+    Question:
+    {question}
+
+    Answer:
+    """,
+        input_variables=["question"],
+    )
+
+    # Output Parser
+    parser = StrOutputParser()
+
+    # Chain
+    main_chain = prompt | llm | parser
+
+    # Generate Response
+    response = main_chain.invoke({
+        "question": text_input
+    })
+
+    return response.content.strip()
+
+
+
+
+
+
+
+
+
+def excel_image(image_path):
+    # --------------------------------------------------
+    # STEP 1 : Extract ONLY the Excel question
+    # --------------------------------------------------
+
+    system_prompt = """
+You are an OCR assistant.
+
+Your task is to read the screenshot and extract ONLY the Excel question.
+
+Rules:
+- Ignore the Excel ribbon.
+- Ignore row and column headers.
+- Ignore formulas already written.
+- Ignore buttons and menus.
+- Ignore answer choices if not part of the question.
+- Return ONLY the question text.
+
+Do not explain anything.
+Do not add headings.
+"""
+
+    response = chat(
+        model="qwen2.5vl:7b",
+        messages=[
+            {
+                "role": "system",
+                "content": system_prompt
+            },
+            {
+                "role": "user",
+                "content": "Extract only the Excel question.",
+                "images": [image_path]
+            }
+        ]
+    )
+
+    extracted_question = response.message.content.strip()
+
+    print("\nExtracted Question:\n")
+    print(extracted_question)
+
+    # --------------------------------------------------
+    # STEP 2 : Load LLM
+    # --------------------------------------------------
+
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+
+    # --------------------------------------------------
+    # STEP 3 : Prompt
+    # --------------------------------------------------
+
+    prompt = PromptTemplate(
+        template="""
+You are an Excel expert.
+
+Answer the following Excel question.
+
+Rules:
+1. Provide the correct answer only.
+2. If a formula is required, return only the formula.
+3. If an explanation is required, keep it concise.
+4. Do not use markdown.
+
+Question:
+{question}
+
+Answer:
+""",
+        input_variables=["question"],
+    )
+
+    parser = StrOutputParser()
+
+    chain = prompt | llm | parser
+
+    # --------------------------------------------------
+    # STEP 4 : Generate Answer
+    # --------------------------------------------------
+
+    answer = chain.invoke({
+        "question": extracted_question
+    })
+
+    return answer 
+
+def excel_text_image(question, image_path):
+    """
+    question   : Extracted text from OCR
+    image_path : Path to the Excel screenshot (optional, not used here)
+    """
+
+    # Load LLM
+    llm = ChatOllama(
+        model="llama3.2",
+        temperature=0
+    )
+
+    # Prompt
+    prompt = PromptTemplate(
+        template="""
+You are an Excel expert.
+
+Answer the following Excel question.
+
+Rules:
+1. Return only the final answer.
+2. If the answer is an Excel formula, return only the formula.
+3. If the question asks for a value, return only the value.
+4. If an explanation is needed, keep it brief.
+5. Do not use markdown.
+
+Question:
+{question}
+
+Answer:
+""",
+        input_variables=["question"],
+    )
+
+    parser = StrOutputParser()
+
+    # Chain
+    chain = prompt | llm | parser
+
+    # Generate answer
+    response = chain.invoke({
+        "question": question
+    })
+
+    return response