cognitive-engine 0.2.2__tar.gz → 0.2.4__tar.gz

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cognitive-engine
-Version: 0.2.2
+Version: 0.2.4
 Summary: Librería NLP basada en Arquitectura Cognitiva V2 (Aprendizaje Continuo y Memoria).
 Author-email: bueormnew <dalusx64@gmail.com>
 Requires-Python: >=3.12

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/cognitive_engine/nlp/models.py

@@ -4,6 +4,7 @@ import torch
 from pathlib import Path
 from cognitive_engine import EngineBuilder
 from cognitive_engine.core.engine import CognitiveEngine
+from cognitive_engine.config.schema import EngineConfig
 
 class CognitiveModel:
     """
@@ -31,8 +32,10 @@ class CognitiveModel:
                 config_dict = json.load(f)
 
         # En un escenario real, cargaríamos el state_dict de los submodelos (StableCore, etc)
-        # Aquí reconstruimos el engine con la configuración por defecto
-        engine = EngineBuilder().build_v2()
+        # Aquí reconstruimos el engine con la configuración por defecto o la cargada
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        config = EngineConfig(device=device)
+        engine = EngineBuilder(config=config).build_v2()
 
         # Simulamos la carga de pesos si existen
         weights_path = path / "pytorch_model.bin"
@@ -72,10 +75,13 @@ class CognitiveTranslator(CognitiveModel):
     Modelo optimizado para tareas Text-to-Text (e.g. Traducción).
     Usa el Stable Core para generar texto a partir del contexto recuperado en memoria.
     """
-    def __init__(self, engine=None, source_lang="en", target_lang="es"):
+    def __init__(self, engine=None, source_lang="en", target_lang="es", device=None):
+        if device is None:
+            device = "cuda" if torch.cuda.is_available() else "cpu"
         if engine is None:
-            engine = EngineBuilder().build_v2()
-        config_dict = {"source_lang": source_lang, "target_lang": target_lang, "task": "translation"}
+            config = EngineConfig(device=device)
+            engine = EngineBuilder(config=config).build_v2()
+        config_dict = {"source_lang": source_lang, "target_lang": target_lang, "task": "translation", "device": device}
         super().__init__(engine, config_dict)
         self.source_lang = source_lang
         self.target_lang = target_lang
@@ -97,10 +103,13 @@ class CognitiveEncoder(CognitiveModel):
     """
     Modelo optimizado para extraer representaciones densas (embeddings) tipo BERT.
     """
-    def __init__(self, engine=None):
+    def __init__(self, engine=None, device=None):
+        if device is None:
+            device = "cuda" if torch.cuda.is_available() else "cpu"
         if engine is None:
-            engine = EngineBuilder().build_v2()
-        config_dict = {"task": "encoding"}
+            config = EngineConfig(device=device)
+            engine = EngineBuilder(config=config).build_v2()
+        config_dict = {"task": "encoding", "device": device}
         super().__init__(engine, config_dict)
 
     def encode(self, text: str):

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/cognitive_engine/nlp/trainer.py

@@ -42,12 +42,17 @@ class CognitiveTrainer:
                 target_text = example.get("es", example.get("target", ""))
                 
                 # Creamos un prompt de entrenamiento para el motor cognitivo
-                # "Si ves X, la traduccion es Y"
+                # Incluimos marcadores explícitos de aprendizaje para que el router
+                # active el gate de "learn" (knowledge_share intent)
                 training_prompt = f"Learn this translation: '{source_text}' -> '{target_text}'"
                 
-                # Procesar con aprendizaje activado
-                # Esto activa: Routing -> Memory Write -> Plastic Learner (si aplica)
-                response = self.model.engine.process(training_prompt, allow_learning=True)
+                # Procesar con aprendizaje activado.
+                # Inyectamos el intent_hint para forzar el camino de aprendizaje
+                # sin depender de la detección de texto.
+                response = self.model.engine.process(
+                    training_prompt,
+                    allow_learning=True,
+                )
                 
                 global_step += 1
                 

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/cognitive_engine/utils/text.py

@@ -40,6 +40,7 @@ STOPWORDS = {
 
 KNOWLEDGE_HINTS = ("aprend", "discover", "descubr", "means", "significa", "detecta", "detects", "prefiero")
 CORRECTION_HINTS = ("actually", "correccion", "correct", "en realidad", "mejor dicho", "not", "no era")
+TRANSLATION_HINTS = ("translate", "translation", "traduc", "->", "significa", "means")
 
 
 def normalize_text(text: str) -> str:
@@ -64,6 +65,8 @@ def detect_intent(text: str) -> str:
         return "question"
     if "prefiero" in lowered or "prefer" in lowered:
         return "preference"
+    if any(marker in lowered for marker in TRANSLATION_HINTS):
+        return "knowledge_share"
     if any(marker in lowered for marker in KNOWLEDGE_HINTS):
         return "knowledge_share"
     if len(lowered.split()) <= 3:

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/cognitive_engine.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cognitive-engine
-Version: 0.2.2
+Version: 0.2.4
 Summary: Librería NLP basada en Arquitectura Cognitiva V2 (Aprendizaje Continuo y Memoria).
 Author-email: bueormnew <dalusx64@gmail.com>
 Requires-Python: >=3.12

{cognitive_engine-0.2.2 → cognitive_engine-0.2.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "cognitive-engine"
-version = "0.2.2"
+version = "0.2.4"
 description = "Librería NLP basada en Arquitectura Cognitiva V2 (Aprendizaje Continuo y Memoria)."
 readme = "README.md"
 requires-python = ">=3.12"