Incorporar model_ai y configuración base para servicios de IA

Author: eduranm

config/settings/base.py

@@ -24,6 +24,7 @@
 ROOT_DIR = Path(__file__).resolve(strict=True).parent.parent.parent
 # core/
 APPS_DIR = ROOT_DIR / "core"
+LLAMA_MODEL_DIR = ROOT_DIR / "model_ai/download"
 
 env = environ.Env()
 READ_DOT_ENV_FILE = env.bool("DJANGO_READ_DOT_ENV_FILE", default=False)
@@ -80,6 +81,7 @@
     "tracker",
     "reference",
     "xml_manager",
+    "model_ai",
 ]
 
 INSTALLED_APPS = DJANGO_APPS + THIRD_PARTY_APPS + LOCAL_APPS + WAGTAIL
@@ -214,7 +216,7 @@
 
 # Wagtail settings
 
-WAGTAIL_SITE_NAME = "markapi"
+WAGTAIL_SITE_NAME = "SciELO XML Tools"
 
 # Search
 # https://docs.wagtail.org/en/stable/topics/search/backends.html
@@ -295,6 +297,8 @@
 }
 
 # LLAMA
-LLAMA_ENABLED = env.bool("LLAMA_ENABLED", default=False)
-LLAMA_MODEL_DIR = ROOT_DIR / "llama3/llama-3.2"
-MODEL_LLAMA = "llama-3.2-3b-instruct-q4_k_m.gguf"
+LLAMA_ENABLED = env.bool("LLAMA_ENABLED", default=True)
+MODEL_LLAMA = "llama-3.2-3b-instruct-q4_k_m.gguf"
+
+#Aumento en el límite de campos
+DATA_UPLOAD_MAX_NUMBER_FIELDS = 10000

llama3/__init__.py model_ai/__init__.pyllama3/__init__.py renamed to model_ai/__init__.py

@@ -0,0 +1,6 @@
+from django.apps import AppConfig
+
+
+class ModelIAConfig(AppConfig):
+    default_auto_field = "django.db.models.BigAutoField"
+    name = "model_ai"

model_ai/apps.py

@@ -0,0 +1,8 @@
+class LlamaDisabledError(Exception):
+    pass
+
+class LlamaModelNotFoundError(FileNotFoundError):
+    pass
+
+class LlamaNotInstalledError(ImportError):
+    pass

model_ai/exceptions.py

@@ -0,0 +1,136 @@
+# Standard library imports
+import logging
+import os
+
+from config.settings.base import (
+    LLAMA_ENABLED,
+    LLAMA_MODEL_DIR,
+)
+
+# Third-party imports
+import google.generativeai as genai
+
+# Local application imports
+from model_ai import messages
+from model_ai.exceptions import (
+    LlamaDisabledError,
+    LlamaModelNotFoundError,
+    LlamaNotInstalledError,
+)
+from model_ai.models import LlamaModel
+
+
+class LlamaService:
+  # Singleton pattern to cache the LLaMA model instance
+  _cached_llm = None
+
+  def __init__(self, messages=None, response_format=None, max_tokens=4000, temperature=0.1, top_p=0.1, mode='chat', nthreads=2):
+    self.messages = messages
+    self.response_format = response_format
+    self.max_tokens = max_tokens
+    self.temperature = temperature
+    self.top_p = top_p
+    self.mode = mode
+
+    model_ai = LlamaModel.objects.first()
+
+    # Try to use Gemini if configured
+    if not model_ai and model_ai.api_key_gemini:
+
+      if not LLAMA_ENABLED:
+        raise LlamaDisabledError("LLaMA is disabled in settings.")
+      
+      if LlamaService._cached_llm is None:
+        try:
+          from llama_cpp import Llama
+        except ImportError as e:
+          raise LlamaNotInstalledError("The 'llama-cpp-python' package is not installed. Please use the llama-activated Docker image (Dockerfile.llama).") from e
+
+        model_ai = LlamaModel.objects.first()
+        if not model_ai:
+          raise LlamaModelNotFoundError("No LLaMA model configured in the database. Please add a LLaMA model entry.")
+      
+        model_path = os.path.join(LLAMA_MODEL_DIR, model_ai.name_file)
+        if not os.path.isfile(model_path):
+          raise LlamaModelNotFoundError(f"LLaMA model file not found at {model_path}. Please ensure the model is downloaded and the path is correct.")
+
+        try:
+          LlamaService._cached_llm = Llama(model_path=model_path, n_ctx=max_tokens, n_threads=nthreads)
+        except Exception as e:
+          raise RuntimeError(f"Failed to initialize LLaMA model: {e}") from e
+        
+      self.llm = LlamaService._cached_llm
+
+  def run(self, user_input):
+    if self.mode == 'chat':
+      return self._run_as_chat(user_input)
+    elif self.mode == 'prompt':
+      return self._run_as_content_generation(user_input)
+
+  def _run_as_chat(self, user_input):
+    """ Run LLaMA in chat mode."""
+    input = self.messages.copy()
+    input.append({
+      'role': 'user',
+      'content': user_input
+    })
+    return self.llm.create_chat_completion(
+      messages=input, 
+      response_format=self.response_format, 
+      max_tokens=self.max_tokens, 
+      temperature=self.temperature, 
+      top_p=self.top_p
+    )
+  
+  def _run_as_content_generation(self, user_input):
+    """ Run LLaMA in completion mode."""
+    model_ai = LlamaModel.objects.first()
+
+    # Try to use Gemini if configured
+    if model_ai and model_ai.api_key_gemini:
+
+      # Setup Gemini API key
+      genai.configure(api_key=model_ai.api_key_gemini)
+
+      # Fetch the Gemini model
+      # FIXME: Hardcoded model name
+      model = genai.GenerativeModel('models/gemini-3.1-flash-lite-preview')
+
+      # Generate content using Gemini
+      return model.generate_content(user_input).text
+
+    # Gemini not configured, fallback to LLaMA
+    else:
+      return self.llm(
+        user_input, 
+        max_tokens=self.max_tokens, 
+        temperature=self.temperature, 
+        stop=["\n\n"]
+      )
+  
+class LlamaInputSettings:
+    @staticmethod
+    def get_first_metadata(text):
+        logging.debug(messages.ALL_FIRST_BLOCK.format(text=text))
+        return messages.ALL_FIRST_BLOCK.format(text=text)
+
+    @staticmethod
+    def get_doi_and_section():
+        return messages.DOI_AND_SECTION_MESSAGES, messages.DOI_AND_SECTION_FORMAT
+
+    @staticmethod
+    def get_titles():
+        return messages.TITLE_MESSAGES, messages.TITLE_RESPONSE_FORMAT
+
+    @staticmethod
+    def get_author_config():
+        return messages.AUTHOR_MESSAGES, messages.AUTHOR_RESPONSE_FORMAT
+
+    @staticmethod
+    def get_affiliations():
+        return messages.AFFILIATION_MESSAGES, messages.AFFILIATION_RESPONSE_FORMAT
+
+    @staticmethod
+    def get_reference():
+        return messages.REFERENCE_MESSAGES, messages.REFERENCE_RESPONSE_FORMAT
+