{"repo": "artyomza312/doziscolar", "files": {"Sin.py": {"18febb45253e997e413049cd33aa2d2560611d17": {"t": 1785674365.0749898, "n": 5291, "h": "3be964466416c3d4", "c": "import numpy as np\nimport requests\nfrom langchain_community.document_loaders import PyPDFLoader\nfrom langchain.text_splitter import CharacterTextSplitter\nfrom sentence_transformers import SentenceTransformer\nimport faiss\nfrom langchain.llms.base import LLM\nfrom langchain.prompts import PromptTemplate\nfrom langchain.chains import LLMChain\nfrom .config import api_key\n# 1. Load and split PDF into chunks\npdf_path = \"test.pdf\"\nloader = PyPDFLoader(pdf_path)\npages = loader.load()\ntext_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)\n\nchunks, metadatas = [], []\nfor page in pages:\n    page_num = page.metadata.get(\"page\")\n    for idx, text in enumerate(text_splitter.split_text(page.page_content)):\n        chunks.append(text)\n        metadatas.append({\"page\": page_num, \"chunk_index\": idx})\n\n# 2. Create embeddings and FAISS index\nembedding_model = SentenceTransformer('all-MiniLM-L6-v2')\nembeddings = embedding_model.encode(chunks, show_progress_bar=True)\ndim = embeddings.shape[1]\nindex = faiss.IndexFlatL2(dim)\nindex.add(np.array(embeddings))\n\n# 3. Retrieval function\ndef retrieve_chunks(question: str, top_k: int = 3) -> list:\n    q_emb = embedding_model.encode([question])\n    distances, indices = index.search(np.array(q_emb), k=top_k)\n    return [\n        {\n            \"text\": chunks[i],\n            \"page\": metadatas[i][\"page\"],\n            \"chunk_index\": metadatas[i][\"chunk_index\"],\n            \"score\": float(dist)\n        }\n        for dist, i in zip(distances[0], indices[0])\n    ]\n\n# 4. Prompt template \ncustom_template = \"\"\"\nAnswer the following question ONLY using the provided context below.\nIf the answer is not found in the context, say 'Not found in context.'\nCite all supporting text as [Page X, Chunk Y].\n\nContext:\n{summaries}\n\nQuestion: {question}\nAnswer:\n\"\"\"\nprompt = PromptTemplate(\n    template=custom_template,\n    input_variables=[\"summaries\", \"question\"]\n)\n\ndef build_context(retrieved: list) -> str:\n    return \"\\n\\n\".join([\n        f\"[Page {c['page']}, Chunk {c['chunk_index']}]\\n{c['text']}\"\n        for c in retrieved\n    ])\n\n# 5. OpenRouter LLM integration با فیلد Pydantic به‌جای __init__\nclass OpenRouterLLM(LLM):\n    api_key: str\n    model_name: str = \"openrouter/cypher-alpha:free\"\n    api_url: str = \"https://openrouter.ai/api/v1/chat/completions\"\n\n    def _call(self, prompt: str, stop=None, **kwargs) -> str:\n        headers = {\n            \"Authorization\": f\"Bearer {self.api_key}\",\n            \"Content-Type\": \"application/json\"\n        }\n        payload = {\n            \"model\": self.model_name,\n            \"messages\": [{\"role\": \"user\", \"content\": prompt}]\n        }\n        resp = requests.post(self.api_url, headers=headers, json=payload, timeout=30)\n        resp.raise_for_status()\n        data = resp.json()\n        return data[\"choices\"][0][\"message\"][\"content\"]\n\n    @property\n    def _identifying_params(self):\n        return {\"model_name\": self.model_name}\n\n    @property\n    def _llm_type(self):\n        return \"openrouter\"\n\n# 6. End-to-end execution\ndef answer_question(question: str, top_k: int = 3) -> str:\n    retrieved = retrieve_chunks(question, top_k)\n    context = build_context(retrieved)\n\n    llm = OpenRouterLLM(\n        api_key=api_key\n    )\n    chain = LLMChain(llm=llm, prompt=prompt)\n    return chain.run(summaries=context, question=question)\n\n\nclass TranslateLLM(LLM):\n    api_key: str\n    model_name: str\n    api_url: str = \"https://openrouter.ai/api/v1/chat/completions\"\n\n    def _call(self, prompt: str, stop=None, **kwargs) -> str:\n        headers = {\n            \"Authorization\": f\"Bearer {self.api_key}\",\n            \"Content-Type\": \"application/json\"\n        }\n        payload = {\n            \"model\": self.model_name,\n            \"messages\": [{\"role\": \"user\", \"content\": prompt}]\n        }\n        resp = requests.post(self.api_url, headers=headers, json=payload, timeout=30)\n        resp.raise_for_status()\n        data = resp.json()\n        return data[\"choices\"][0][\"message\"][\"content\"]\n\n    @property\n    def _identifying_params(self):\n        return {\"model_name\": self.model_name}\n\n    @property\n    def _llm_type(self):\n        return \"openrouter\"\n\n\ndef translate_fa2en(text):\n    llm = TranslateLLM(api_key=api_key, model_name=\"deepseek/deepseek-r1-0528:free\")\n\n    prompt = f\"\"\"Translate the following Persian text to English. Only output the English translation, with no extra explanation.\n\n    Persian: {text}\n    English:\"\"\"\n\n    response = llm(prompt)\n    # Strip whitespace, just in case\n    return response.strip()\n\ndef translate_en2fa(text):\n    llm = TranslateLLM(api_key=api_key, model_name=\"deepseek/deepseek-r1-0528:free\")\n\n    prompt = f\"\"\"Translate the following English text to Persian. Only output the Persian translation, with no extra explanation.\n\n       English: {text}\n       Persian:\"\"\"\n\n    response = llm(prompt)\n    # Strip whitespace, just in case\n    return response.strip()\n\n\n\nif __name__ == \"__main__\":\n    while True:\n        q = input(\"سؤال را وارد کنید (یا '1' برای خروج): \").strip()\n        if q == \"1\":\n            print(\"خروج از برنامه.\")\n            break\n        q_new = translate_fa2en(q)\n        print(f\"Processing: {q_new}\")\n        pred = answer_question(q, top_k=3)\n        pred_new = translate_en2fa(pred)\n        print(\"Answer:\", pred_new)"}}}}