Add runtime flag so we can alternatively serve via vllm

Also added 'ls' alias for 'list'. Port logic is buggy, just support
--port for now. Pass 'args' to pull function, instead of "args.store".

Signed-off-by: Eric Curtin <[email protected]>

Author: ericcurtin

ramalama/cli.py

@@ -33,6 +33,10 @@ def init_cli():
     parser.add_argument("--dryrun",
                         action='store_true',
                         help="show container runtime command without executing it")
+    parser.add_argument("--runtime",
+                        default="llama.cpp",
+                        choices=["llama.cpp", "vllm"],
+                        help="specify the runtime to use (default: llama.cpp). Valid options are 'llama.cpp' and 'vllm'.")
     parser.add_argument("--nocontainer",
                         default=not use_container(),
                         action='store_true',
@@ -163,16 +167,20 @@ def list_files_by_modification():
                   reverse=True)
 
 
-def list_parser(subparsers):
-    parser = subparsers.add_parser(
-        'list', help='List all downloaded AI Models')
+def add_list_parser(subparsers, name, func):
+    parser = subparsers.add_parser(name, help='List all downloaded AI Models')
     parser.add_argument("-n", "--noheading", dest="noheading",
                         action='store_true',
                         help="do not display heading")
     parser.add_argument("--json", dest="json",
                         action='store_true',
                         help="print using json")
-    parser.set_defaults(func=list_cli)
+    parser.set_defaults(func=func)
+
+
+def list_parser(subparsers):
+    add_list_parser(subparsers, 'list', list_cli)
+    add_list_parser(subparsers, 'ls', list_cli)
 
 
 def list_cli(args):
@@ -247,15 +255,9 @@ def run_cli(args):
 
 
 def serve_parser(subparsers):
-    port = "8080"
-    host = os.getenv('RAMALAMA_HOST', port)
-    split = host.rsplit(':', 1)
-    if len(split) > 1:
-        port = split[1]
-
     parser = subparsers.add_parser(
-        'serve', help='Serve RESTAPI on specified AI Model')
-    parser.add_argument("--port", default=port,
+        'serve', help='Serve REST API on specified AI Model')
+    parser.add_argument("--port", default="8080",
                         help="port for AI Model server to listen on")
     parser.add_argument('model')         # positional argument
     parser.set_defaults(func=serve_cli)
@@ -299,9 +301,6 @@ def run_container(args):
             wd = target
             break
 
-    port = "8080"
-    host = os.getenv('RAMALAMA_HOST', port)
-
     conman_args = [conman, "run",
                    "--rm",
                    "-it",
@@ -310,9 +309,8 @@ def run_container(args):
                    f"-v{home}:{home}",
                    "-v/tmp:/tmp",
                    f"-v{sys.argv[0]}:/usr/bin/ramalama:ro",
-                   "-e", "RAMALAMA_HOST",
                    "-e", "RAMALAMA_TRANSPORT",
-                   "-p", f"{host}:{port}",
+                   "-p", f"{args.port}:{args.port}",
                    f"-v{wd}:/usr/share/ramalama/ramalama:ro"]
     if os.path.exists("/dev/dri"):
         conman_args += ["--device", "/dev/dri"]

ramalama/model.py

@@ -31,14 +31,18 @@ def push(self, args):
             f"ramalama push for {self.type} not implemented")
 
     def run(self, args):
-        symlink_path = self.pull(args.store)
+        symlink_path = self.pull(args)
         exec_cmd(["llama-cli", "-m",
                   symlink_path, "--log-disable", "-cnv", "-p", "You are a helpful assistant"])
 
     def serve(self, args):
-        symlink_path = self.pull(args.store)
+        symlink_path = self.pull(args)
 
-        if args.port:
-            args.port = os.getenv("RAMALAMA_HOST")
+        # Check the runtime argument and execute the corresponding command
+        if args.runtime == "vllm":
+            exec_cmd(["vllm", "serve", "--port", args.port, symlink_path])
+        elif args.runtime == "llama.cpp":
+            exec_cmd(["llama-server", "--port", args.port, "-m", symlink_path])
+        else:
+            raise ValueError(f"Unsupported runtime: {args.runtime}")
 
-        exec_cmd(["llama-server", "--port", args.port, "-m", symlink_path])