rename toolchain/ --> llama_toolchain/

llama_toolchain/cli/inference/configure.py

@@ -0,0 +1,82 @@
+import argparse
+import os
+import textwrap
+
+from pathlib import Path
+
+from llama_toolchain.cli.subcommand import Subcommand
+from llama_toolchain.utils import DEFAULT_DUMP_DIR
+
+
+CONFIGS_BASE_DIR = os.path.join(DEFAULT_DUMP_DIR, "configs")
+
+
+class InferenceConfigure(Subcommand):
+    """Llama cli for configuring llama toolchain configs"""
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "configure",
+            prog="llama inference configure",
+            description="Configure llama toolchain inference configs",
+            epilog=textwrap.dedent(
+                """
+                Example:
+                    llama inference configure
+                """
+            ),
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_inference_configure_cmd)
+
+    def _add_arguments(self):
+        pass
+
+    def read_user_inputs(self):
+        checkpoint_dir = input("Enter the checkpoint directory for the model (e.g., ~/.llama/checkpoints/Meta-Llama-3-8B/): ")
+        model_parallel_size = input("Enter model parallel size (e.g., 1 for 8B / 8 for 70B and 405B): ")
+
+        return checkpoint_dir, model_parallel_size
+
+    def write_output_yaml(
+        self,
+        checkpoint_dir,
+        model_parallel_size,
+        yaml_output_path
+    ):
+        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())
+
+        print(f"YAML configuration has been written to {yaml_output_path}")
+
+    def _run_inference_configure_cmd(self, args: argparse.Namespace) -> None:
+        checkpoint_dir, model_parallel_size = self.read_user_inputs()
+        checkpoint_dir = os.path.expanduser(checkpoint_dir)
+
+        if not (
+            checkpoint_dir.endswith("original") or
+            checkpoint_dir.endswith("original/")
+        ):
+            checkpoint_dir = os.path.join(checkpoint_dir, "original")
+
+        os.makedirs(CONFIGS_BASE_DIR, exist_ok=True)
+        yaml_output_path = Path(CONFIGS_BASE_DIR) / "inference.yaml"
+
+        self.write_output_yaml(
+            checkpoint_dir,
+            model_parallel_size,
+            yaml_output_path,
+        )

llama_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

llama_toolchain/cli/inference/inference.py

@@ -0,0 +1,30 @@
+import argparse
+import textwrap
+
+from llama_toolchain.cli.inference.configure import InferenceConfigure
+from llama_toolchain.cli.inference.start import InferenceStart
+from llama_toolchain.cli.subcommand import Subcommand
+
+
+class InferenceParser(Subcommand):
+    """Llama cli for inference apis"""
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "inference",
+            prog="llama inference",
+            description="Run inference on a llama model",
+            epilog=textwrap.dedent(
+                """
+                Example:
+                    llama inference start <options>
+                """
+            ),
+        )
+
+        subparsers = self.parser.add_subparsers(title="inference_subcommands")
+
+        # Add sub-commandsa
+        InferenceStart.create(subparsers)
+        InferenceConfigure.create(subparsers)

llama_toolchain/cli/inference/start.py

@@ -0,0 +1,54 @@
+import argparse
+import textwrap
+
+from llama_toolchain.cli.subcommand import Subcommand
+
+from llama_toolchain.inference.server import main as inference_server_init
+
+
+class InferenceStart(Subcommand):
+    """Llama Inference cli for starting inference server"""
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "start",
+            prog="llama inference start",
+            description="Start an inference server",
+            epilog=textwrap.dedent(
+                """
+                Example:
+                    llama inference start <options>
+                """
+            ),
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_inference_start_cmd)
+
+    def _add_arguments(self):
+        self.parser.add_argument(
+            "--port",
+            type=int,
+            help="Port to run the server on. Defaults to 5000",
+            default=5000,
+        )
+        self.parser.add_argument(
+            "--disable-ipv6",
+            action="store_true",
+            help="Disable IPv6 support",
+            default=False,
+        )
+        self.parser.add_argument(
+            "--config",
+            type=str,
+            help="Path to config file",
+            default="inference"
+        )
+
+    def _run_inference_start_cmd(self, args: argparse.Namespace) -> None:
+        inference_server_init(
+            config_path=args.config,
+            port=args.port,
+            disable_ipv6=args.disable_ipv6,
+        )