Update app.py

app.py

@@ -67,7 +67,9 @@ class S3DirectStream:
 
     def load_model_from_s3(self, model_name):
         try:
-            model_prefix = model_name.lower()
+            profile, model = model_name.split("/", 1) if "/" in model_name else ("", model_name)
+
+            model_prefix = f"{profile}/{model}".lower()
             model_files = self.get_model_file_parts(model_prefix)
 
             if not model_files:
@@ -89,14 +91,22 @@ class S3DirectStream:
             model = AutoModelForCausalLM.from_pretrained(f"s3://{self.bucket_name}/{model_prefix}", config=config_json, from_tf=False)
             return model
 
+        except HTTPException as e:
+            raise e
         except Exception as e:
-            raise HTTPException(status_code=500, detail=f"Error al cargar el modelo desde S3: {e}")
+            try:
+                logger.error(f"Error al cargar el modelo desde S3, intentando desde Hugging Face: {e}")
+                model = AutoModelForCausalLM.from_pretrained(model_name)
+                return model
+            except Exception as hf_error:
+                raise HTTPException(status_code=500, detail=f"Error al cargar el modelo desde Hugging Face: {hf_error}")
 
     def load_tokenizer_from_s3(self, model_name):
         try:
-            tokenizer_stream = self.stream_from_s3(f"{model_name}/tokenizer.json")
+            profile, model = model_name.split("/", 1) if "/" in model_name else ("", model_name)
+            tokenizer_stream = self.stream_from_s3(f"{profile}/{model}/tokenizer.json")
             tokenizer_data = tokenizer_stream.read().decode("utf-8")
-            tokenizer = AutoTokenizer.from_pretrained(f"s3://{self.bucket_name}/{model_name}")
+            tokenizer = AutoTokenizer.from_pretrained(f"s3://{self.bucket_name}/{profile}/{model}")
             return tokenizer
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Error al cargar el tokenizer desde S3: {e}")
@@ -107,7 +117,8 @@ class S3DirectStream:
 
             for file in tqdm(files_to_download, desc="Subiendo archivos a S3"):
                 file_name = os.path.basename(file)
-                s3_key = f"{model_name}/{file_name}"
+                profile, model = model_name.split("/", 1) if "/" in model_name else ("", model_name)
+                s3_key = f"{profile}/{model}/{file_name}"
                 if not self.file_exists_in_s3(s3_key):
                     self.upload_file_to_s3(file, s3_key)
 
@@ -116,12 +127,23 @@ class S3DirectStream:
 
     def upload_file_to_s3(self, file_path, s3_key):
         try:
-            with open(file_path, 'rb') as data:
-                self.s3_client.put_object(Bucket=self.bucket_name, Key=s3_key, Body=data)
+            self.create_s3_folders(s3_key)
+            s3_client.put_object(Bucket=self.bucket_name, Key=s3_key, Body=open(file_path, 'rb'))
             os.remove(file_path)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Error al subir archivo a S3: {e}")
 
+    def create_s3_folders(self, s3_key):
+        try:
+            folder_keys = s3_key.split('/')
+            for i in range(1, len(folder_keys)):
+                folder_key = '/'.join(folder_keys[:i]) + '/'
+                if not self.file_exists_in_s3(folder_key):
+                    self.s3_client.put_object(Bucket=self.bucket_name, Key=folder_key, Body='')
+
+        except Exception as e:
+            raise HTTPException(status_code=500, detail=f"Error al crear carpetas en S3: {e}")
+
     def file_exists_in_s3(self, s3_key):
         try:
             self.s3_client.head_object(Bucket=self.bucket_name, Key=s3_key)