new model

Dockerfile.txt

@@ -1,3 +1,5 @@
+#dockerfile.txt
+
 FROM python:3.10
 
 # Prevent Python from writing .pyc files / enable unbuffered logs

app.py

@@ -14,7 +14,14 @@ from tensorflow import keras
 from PIL import Image
 
 import tenacity
-from fake_useragent import UserAgent
+try:
+    from fake_useragent import UserAgent
+    user_agent = UserAgent().random
+except Exception:
+    user_agent = "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "\
+                 "(KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36"
+header_template = {"User-Agent": user_agent}
+
 
 # --- LangChain v0.2 family ---
 from langchain_text_splitters import RecursiveCharacterTextSplitter
@@ -41,7 +48,7 @@ from pydantic import BaseModel, Field  # <-- switched to Pydantic v2
 from huggingface_hub import snapshot_download
 
 # === LLM endpoint moderno (compatible con huggingface_hub>=0.23) ===
-from langchain_huggingface import HuggingFaceEndpoint  # Opción 1
+from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint
 
 # Theming + URL list
 import theme
@@ -97,8 +104,6 @@ image_gradio_app = gr.Interface(
 # ============================================
 # 2) KNOWLEDGE LOADING (RAG: loader + splitter)
 # ============================================
-user_agent = UserAgent().random
-header_template = {"User-Agent": user_agent}
 
 @tenacity.retry(wait=tenacity.wait_fixed(3), stop=tenacity.stop_after_attempt(3), reraise=True)
 def load_url(url: str):
@@ -179,26 +184,25 @@ qa_prompt = ChatPromptTemplate.from_template(
 )
 
 
+
 # =============================
-# 4) LLM — Hugging Face Inference API (Mixtral)
+# 4) LLM — Hugging Face Inference API (Llama 3 chat)
 # =============================
-# Requiere el secreto HUGGINGFACEHUB_API_TOKEN en el Space
-llm = HuggingFaceEndpoint(
+endpoint = HuggingFaceEndpoint(
     repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
-    task="conversational",                         # ok para Mixtral
-    # estos parámetros VAN afuera de model_kwargs:
+    task="conversational",
     max_new_tokens=2000,
     temperature=0.1,
     top_k=30,
     repetition_penalty=1.03,
     return_full_text=False,
-    # token del Space
     huggingfacehub_api_token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
-    timeout=120,                                    # opcional
-    # deja model_kwargs vacío o úsalo solo para args que no tengan campo dedicado
-    model_kwargs={},                                 
+    timeout=120,
+    model_kwargs={},
 )
 
+# Usa el wrapper de modelo de **chat**
+llm = ChatHuggingFace(client=endpoint)
 
 # ===========================================
 # 5) Chain (memory + robust JSON extraction)

requirements.txt

@@ -1,3 +1,4 @@
+# requirements.txt
 # Core
 transformers>=4.42.0
 torch