added mixtral-8x7b

app.py

@@ -56,15 +56,15 @@ DEMO_EXAMPLES = [
 
 # ── Función principal ─────────────────────────────────────────────────────────
 
-def corregir(htr_text: str, top_k: int, mostrar_prompt: bool):
+def corregir(htr_text: str, top_k: int, mostrar_prompt: bool, model: str):
     if not htr_text.strip():
-        return "", "", "", "", "⚠ Introduce un texto HTR para corregir."
+        return "", "", "", "", " Introduce un texto HTR para corregir."
 
     if not os.getenv("OPENAI_API_KEY"):
         return "", "", "", "", " Falta OPENAI_API_KEY en el fichero .env"
 
     try:
-        result = corrector.correct(htr_text, top_k=int(top_k))
+        result = corrector.correct(htr_text, top_k=int(top_k), model= model)
     except Exception as e:
         return "", "", "", "", f" Error al llamar a la API: {e}"
 
@@ -235,6 +235,14 @@ with gr.Blocks(
                             minimum=1, maximum=10, value=5, step=1,
                             label="Documents retrieved (k)",
                         )
+                        model_selector = gr.Dropdown(
+                            label="Modelo LLM",
+                            choices=[
+                                "llama-3.3-70b-versatile",
+                                "mixtral-8x7b-32768",
+                            ],
+                            value="llama-3.3-70b-versatile",
+                        )
                         show_prompt = gr.Checkbox(label="Show RAG prompt", value=False)
                     btn_corregir = gr.Button("✦ Correct with RAG", variant="primary")
 
@@ -263,7 +271,7 @@ with gr.Blocks(
 
             btn_corregir.click(
                 fn=corregir,
-                inputs=[htr_input, top_k_slider, show_prompt],
+                inputs=[htr_input, top_k_slider, show_prompt, model_selector],
                 outputs=[corrected_out, docs_out, analysis_out, diff_out, status_out, prompt_out],
             )
 

rag_corrector.py

@@ -53,7 +53,7 @@ class RAGCorrector:
         self.client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"), base_url=os.getenv("OPENAI_BASE_URL", "https://api.x.ai/v1"),)
     # ── API pública ──────────────────────────────────────────────────────────
 
-    def correct(self, htr_text: str, top_k: int = TOP_K) -> Dict:
+    def correct(self, htr_text: str, top_k: int = TOP_K, model: str = None) -> Dict:
         """
         Corrige un texto HTR usando RAG.
 
@@ -71,7 +71,7 @@ class RAGCorrector:
 
         prompt = self._build_prompt(htr_text, retrieved, htr_errors, grafia_warns)
 
-        corrected = self._call_llm(prompt)
+        corrected = self._call_llm(prompt, model=model or MODEL)
 
         return {
             "corrected":    corrected,
@@ -79,7 +79,7 @@ class RAGCorrector:
             "retrieved":    retrieved,
             "htr_errors":   htr_errors,
             "grafia_warns": grafia_warns,
-            "model":        MODEL,
+            "model":        model or MODEL,
         }
 
     # ── Detección de patrones ────────────────────────────────────────────────
@@ -159,9 +159,9 @@ class RAGCorrector:
 
     # ── Llamada al LLM ───────────────────────────────────────────────────────
 
-    def _call_llm(self, user_prompt: str) -> str:
+    def _call_llm(self, user_prompt: str, model: str = MODEL) -> str:
         response = self.client.chat.completions.create(
-            model=MODEL,
+            model=model, # usa el modelo que llega, no el de .env
             temperature=0.1,        # baja temperatura: reproducibilidad
             max_tokens=1024,
             messages=[