Update model.py

1/model.py

@@ -2,6 +2,7 @@ import triton_python_backend_utils as pb_utils
 from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
 import numpy as np
 import json
+import os
 
 class TritonPythonModel:
     def initialize(self, args):
@@ -12,9 +13,27 @@ class TritonPythonModel:
         """
         self.logger = pb_utils.Logger
 
+        current_file_path = os.path.abspath(__file__)
+        self.logger.log_info(f"current_file_path: {current_file_path}")
+
+
         self.model_name = args["model_name"]
         model_repository = args["model_repository"]
         model_path = f"{model_repository}/{self.model_name}"
+        #model_path = "/cheetah/input/model/gemma-3-1b-it/gemma-3-1b-it"
+
+        input_model_path = model_path
+
+        if os.path.exists(input_model_path):
+            file_list = os.listdir(input_model_path)
+            self.logger.log_info(f"'{input_model_path}' 디렉토리의 파일 목록:")
+        for file_name in file_list:
+            self.logger.log_info(file_name)
+        else:
+            self.logger.log_info(f"'{input_model_path}' 디렉토리가 존재하지 않습니다.")
+
+        self.logger.log_info(f"model_repository: {model_repository}")
+        self.logger.log_info(f"model_path: {model_path}")
 
         self.model_config = json.loads(args["model_config"])
 

config.pbtxt

@@ -1,6 +1,7 @@
+# Triton backend to use
 name: "gemma-3-1b-it"
-max_batch_size: 0
 backend: "python"
+max_batch_size: 0
 
 # Triton should expect as input a single string
 # input of variable length named 'text_input'
@@ -71,6 +72,7 @@ output [
   }
 ]
 
+
 parameters: [
   {
     key: "enable_inference_trace",
@@ -84,3 +86,4 @@ instance_group [
     count: 1
   }
 ]
+

gemma-3-1b-it/sub/test.txt

@@ -1,3 +1,6 @@
 test.txt
 test.txt
 test.txt
+test.txt
+test.txt
+