use default_config file to configure inference

requirements.txt

@@ -1,5 +1,6 @@
 accelerate
 black==24.4.2
+blobfile
 codeshield
 fairscale
 fastapi

toolchain/cli/inference/configure.py

@@ -46,17 +46,17 @@ class InferenceConfigure(Subcommand):
         model_parallel_size,
         yaml_output_path
     ):
-        yaml_content = textwrap.dedent(f"""
-            inference_config:
-                impl_type: "inline"
-                inline_config:
-                    checkpoint_type: "pytorch"
-                    checkpoint_dir: {checkpoint_dir}/
-                    tokenizer_path: {checkpoint_dir}/tokenizer.model
-                    model_parallel_size: {model_parallel_size}
-                    max_seq_len: 2048
-                    max_batch_size: 1
-            """)
+        current_dir = os.path.dirname(os.path.abspath(__file__))
+        default_conf_path = os.path.join(current_dir, "default_configuration.yaml")
+
+        with open(default_conf_path, "r") as f:
+            yaml_content = f.read()
+
+        yaml_content = yaml_content.format(
+            checkpoint_dir=checkpoint_dir,
+            model_parallel_size=model_parallel_size,
+        )
+
         with open(yaml_output_path, 'w') as yaml_file:
             yaml_file.write(yaml_content.strip())
 

toolchain/cli/inference/default_configuration.yaml

@@ -0,0 +1,9 @@
+inference_config:
+  impl_type: "inline"
+  inline_config:
+    checkpoint_type: "pytorch"
+    checkpoint_dir: {checkpoint_dir}/
+    tokenizer_path: {checkpoint_dir}/tokenizer.model
+    model_parallel_size: {model_parallel_size}
+    max_seq_len: 2048
+    max_batch_size: 1