Spaces:
Sleeping
Sleeping
| import os | |
| import json | |
| import requests | |
| from sentence_transformers import SentenceTransformer, util | |
| from dotenv import load_dotenv | |
| # Load environment variables | |
| load_dotenv() | |
| # API configurations | |
| HF_API_TOKEN = os.getenv("HF_API_TOKEN") | |
| if not HF_API_TOKEN: | |
| raise ValueError("Hugging Face API Token is missing from .env file") | |
| # DeepSeek API endpoint | |
| DEEPSEEK_API = "https://api-inference.huggingface.co/models/deepseek-ai/deepseek-llm-7b" | |
| headers = {"Authorization": f"Bearer {HF_API_TOKEN}"} | |
| # Load models | |
| similarity_model = SentenceTransformer('paraphrase-MiniLM-L6-v2') | |
| # Load dataset | |
| with open("ds.json") as f: | |
| dataset = json.load(f) | |
| # Precompute embeddings | |
| dataset_questions = [item.get("Question", "").lower().strip() for item in dataset] | |
| dataset_answers = [item.get("Answer", "") for item in dataset] | |
| dataset_embeddings = similarity_model.encode(dataset_questions, convert_to_tensor=True) | |
| def query_deepseek(prompt): | |
| payload = { | |
| "inputs": prompt, | |
| "parameters": { | |
| "max_new_tokens": 500, | |
| "temperature": 0.7, | |
| "do_sample": True | |
| } | |
| } | |
| try: | |
| response = requests.post(DEEPSEEK_API, headers=headers, json=payload) | |
| response.raise_for_status() | |
| result = response.json() | |
| return result[0].get("generated_text", "").strip() if isinstance(result, list) and result else "" | |
| except Exception as e: | |
| print(f"DeepSeek API error: {e}") | |
| return "" | |
| def get_best_answer(user_input): | |
| # Find best match from dataset | |
| user_embedding = similarity_model.encode(user_input.lower().strip(), convert_to_tensor=True) | |
| similarities = util.pytorch_cos_sim(user_embedding, dataset_embeddings)[0] | |
| best_match_idx = similarities.argmax().item() | |
| best_score = similarities[best_match_idx].item() | |
| if best_score >= 0.65: # Good match found | |
| original_answer = dataset_answers[best_match_idx] | |
| prompt = f"""Rephrase this university answer to be more helpful while keeping key information: | |
| Question: {user_input} | |
| Original Answer: {original_answer} | |
| Improved Answer:""" | |
| else: # No good match | |
| prompt = f"""As a university assistant, provide a helpful response to: | |
| Question: {user_input} | |
| Answer:""" | |
| deepseek_response = query_deepseek(prompt) | |
| if deepseek_response: | |
| for marker in ["Improved Answer:", "Answer:"]: | |
| if marker in deepseek_response: | |
| return deepseek_response.split(marker)[-1].strip() | |
| return deepseek_response | |
| return dataset_answers[best_match_idx] if best_score >= 0.65 else """I couldn't find specific information. | |
| Please visit the UOE website: https://ue.edu.pk/""" |