Spaces:
Running
Running
Update rag7.py
Browse files
rag7.py
CHANGED
|
@@ -6,19 +6,17 @@ import os
|
|
| 6 |
from sentence_transformers import SentenceTransformer
|
| 7 |
from huggingface_hub import InferenceClient
|
| 8 |
|
| 9 |
-
# === CONFIGURACIÓN ===
|
| 10 |
MODEL_NAME = "openai/gpt-oss-20b"
|
| 11 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 12 |
|
| 13 |
if not HF_TOKEN:
|
| 14 |
raise RuntimeError("❌ HF_TOKEN no encontrado en rag7.py")
|
| 15 |
|
| 16 |
-
# === CARGAR FAISS RAG7 ===
|
| 17 |
index_path = "nlp_index.faiss"
|
| 18 |
docs_path = "nlp_docs.pkl"
|
| 19 |
|
| 20 |
if not os.path.exists(index_path) or not os.path.exists(docs_path):
|
| 21 |
-
raise FileNotFoundError("❌ Faltan archivos de RAG7
|
| 22 |
|
| 23 |
index = faiss.read_index(index_path)
|
| 24 |
with open(docs_path, "rb") as f:
|
|
@@ -28,7 +26,8 @@ with open(docs_path, "rb") as f:
|
|
| 28 |
|
| 29 |
embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
|
| 30 |
|
| 31 |
-
def retrieve_context(query: str, k: int =
|
|
|
|
| 32 |
try:
|
| 33 |
emb = embedding_model.encode([query], convert_to_numpy=True).astype('float32')
|
| 34 |
emb = emb / np.linalg.norm(emb)
|
|
@@ -39,6 +38,10 @@ def retrieve_context(query: str, k: int = 2) -> str:
|
|
| 39 |
|
| 40 |
def generate_practical_guide(message: str, temperature=0.7, top_p=0.95, max_tokens=2048) -> str:
|
| 41 |
context = retrieve_context(message)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
if context:
|
| 43 |
full_prompt = (
|
| 44 |
f"Responde usando únicamente la siguiente información de contexto. Si el contexto no responde la pregunta, usa tu conocimiento general pero sé honesto sobre sus límites.\n\n"
|
|
@@ -50,7 +53,7 @@ def generate_practical_guide(message: str, temperature=0.7, top_p=0.95, max_toke
|
|
| 50 |
|
| 51 |
client = InferenceClient(token=HF_TOKEN, model=MODEL_NAME, timeout=60)
|
| 52 |
messages = [
|
| 53 |
-
{"role": "system", "content":
|
| 54 |
{"role": "user", "content": full_prompt}
|
| 55 |
]
|
| 56 |
|
|
|
|
| 6 |
from sentence_transformers import SentenceTransformer
|
| 7 |
from huggingface_hub import InferenceClient
|
| 8 |
|
|
|
|
| 9 |
MODEL_NAME = "openai/gpt-oss-20b"
|
| 10 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 11 |
|
| 12 |
if not HF_TOKEN:
|
| 13 |
raise RuntimeError("❌ HF_TOKEN no encontrado en rag7.py")
|
| 14 |
|
|
|
|
| 15 |
index_path = "nlp_index.faiss"
|
| 16 |
docs_path = "nlp_docs.pkl"
|
| 17 |
|
| 18 |
if not os.path.exists(index_path) or not os.path.exists(docs_path):
|
| 19 |
+
raise FileNotFoundError("❌ Faltan archivos de RAG7")
|
| 20 |
|
| 21 |
index = faiss.read_index(index_path)
|
| 22 |
with open(docs_path, "rb") as f:
|
|
|
|
| 26 |
|
| 27 |
embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
|
| 28 |
|
| 29 |
+
def retrieve_context(query: str, k: int = 3) -> str:
|
| 30 |
+
"""Recupera los 3 fragmentos más relevantes (aumentado de 2 a 3 para más contexto)"""
|
| 31 |
try:
|
| 32 |
emb = embedding_model.encode([query], convert_to_numpy=True).astype('float32')
|
| 33 |
emb = emb / np.linalg.norm(emb)
|
|
|
|
| 38 |
|
| 39 |
def generate_practical_guide(message: str, temperature=0.7, top_p=0.95, max_tokens=2048) -> str:
|
| 40 |
context = retrieve_context(message)
|
| 41 |
+
|
| 42 |
+
# ✅ Prompt del sistema idéntico al de tu Space público
|
| 43 |
+
system_message = "Eres un asistente experto en desarrollo humano. Responde con claridad, profundidad y empatía, citando conceptos de los libros si es relevante."
|
| 44 |
+
|
| 45 |
if context:
|
| 46 |
full_prompt = (
|
| 47 |
f"Responde usando únicamente la siguiente información de contexto. Si el contexto no responde la pregunta, usa tu conocimiento general pero sé honesto sobre sus límites.\n\n"
|
|
|
|
| 53 |
|
| 54 |
client = InferenceClient(token=HF_TOKEN, model=MODEL_NAME, timeout=60)
|
| 55 |
messages = [
|
| 56 |
+
{"role": "system", "content": system_message},
|
| 57 |
{"role": "user", "content": full_prompt}
|
| 58 |
]
|
| 59 |
|