Add support for converting to OCI artifacts

Makefile

@@ -7,7 +7,7 @@ SHAREDIR ?= ${PREFIX}/share
 PYTHON ?= $(shell command -v python3 python|head -n1)
 DESTDIR ?= /
 PATH := $(PATH):$(HOME)/.local/bin
-MYPIP ?= pip
+MYPIP ?= uv pip
 IMAGE ?= ramalama
 PROJECT_DIR:=$(shell dirname $(realpath $(lastword $(MAKEFILE_LIST))))
 EXCLUDE_DIRS := .venv venv .tox build

docs/ramalama-convert.1.md

@@ -39,14 +39,15 @@ Image to use when converting to GGUF format (when then `--gguf` option has been
 executable and available in the `PATH`. The script is available from the `llama.cpp` GitHub repo. Defaults to the current
 `quay.io/ramalama/ramalama-rag` image.
 
-#### **--type**=*raw* | *car*
+#### **--type**="artifact" | *raw* | *car*
 
-type of OCI Model Image to convert.
+Convert the MODEL to the specified OCI Object
 
-| Type | Description                                                   |
-| ---- | ------------------------------------------------------------- |
-| car  | Includes base image with the model stored in a /models subdir |
-| raw  | Only the model and a link file model.file to it stored at /   |
+| Type     | Description                                                   |
+| -------- | ------------------------------------------------------------- |
+| artifact | Store AI Models as artifacts                                  |
+| car      | Traditional OCI image including base image with the model stored in a /models subdir |
+| raw      | Traditional OCI image including only the model and a link file `model.file` pointed at it stored at /   |
 
 ## EXAMPLE
 

docs/ramalama.conf

@@ -32,6 +32,14 @@
 #
 #carimage = "registry.access.redhat.com/ubi10-micro:latest"
 
+# Convert the MODEL to the specified OCI Object
+# Options: artifact, car, raw
+#
+# artifact: Store AI Models as artifacts
+# car:      Traditional OCI image including base image with the model stored in a /models subdir
+# raw:      Traditional OCI image including only the model and a link file `model.file` pointed at it stored at /
+#convert_type = "raw"
+
 # Run RamaLama in the default container.
 #
 #container = true

docs/ramalama.conf.5.md

@@ -84,6 +84,18 @@ Min chunk size to attempt reusing from the cache via KV shifting
 Run RamaLama in the default container.
 RAMALAMA_IN_CONTAINER environment variable overrides this field.
 
+**convert_type**="raw"
+
+Convert the MODEL to the specified OCI Object
+Options: artifact, car, raw
+
+| Type     | Description                                                   |
+| -------- | ------------------------------------------------------------- |
+| artifact | Store AI Models as artifacts                                  |
+| car      | Traditional OCI image including base image with the model stored in a /models subdir |
+| raw      | Traditional OCI image including only the model and a link file `model.file` pointed at it stored at /   |
+
+
 **ctx_size**=0
 
 Size of the prompt context (0 = loaded from model)

ramalama/cli.py

@@ -734,11 +734,12 @@ def convert_parser(subparsers):
     )
     parser.add_argument(
         "--type",
-        default="raw",
-        choices=["car", "raw"],
+        default=CONFIG.convert_type,
+        choices=["artifact", "car", "raw"],
         help="""\
 type of OCI Model Image to push.
 
+Model "artifact" stores the AI Model as an OCI Artifact.
 Model "car" includes base image with the model stored in a /models subdir.
 Model "raw" contains the model and a link file model.file to it stored at /.""",
     )
@@ -775,11 +776,12 @@ def push_parser(subparsers):
     add_network_argument(parser)
     parser.add_argument(
         "--type",
-        default="raw",
-        choices=["car", "raw"],
+        default=CONFIG.convert_type,
+        choices=["artifact", "car", "raw"],
         help="""\
 type of OCI Model Image to push.
 
+Model "artifact" stores the AI Model as an OCI Artifact.
 Model "car" includes base image with the model stored in a /models subdir.
 Model "raw" contains the model and a link file model.file to it stored at /.""",
     )
@@ -794,19 +796,25 @@ def push_parser(subparsers):
     parser.set_defaults(func=push_cli)
 
 
-def _get_source_model(args):
+def _get_source_model(args, transport=None):
     src = shortnames.resolve(args.SOURCE)
-    smodel = New(src, args)
+    smodel = New(src, args, transport=transport)
     if smodel.type == "OCI":
+        if not args.TARGET:
+            return smodel
         raise ValueError(f"converting from an OCI based image {src} is not supported")
     if not smodel.exists() and not args.dryrun:
         smodel.pull(args)
     return smodel
 
 
 def push_cli(args):
-    source_model = _get_source_model(args)
     target = args.SOURCE
+    transport = None
+    if not args.TARGET:
+        transport = "oci"
+    source_model = _get_source_model(args, transport=transport)
+
     if args.TARGET:
         target = shortnames.resolve(args.TARGET)
     target_model = New(target, args)
@@ -1189,9 +1197,14 @@ def serve_cli(args):
         model.ensure_model_exists(args)
     except KeyError as e:
         try:
+            if "://" in args.MODEL:
+                raise e
             args.quiet = True
             model = TransportFactory(args.MODEL, args, ignore_stderr=True).create_oci()
             model.ensure_model_exists(args)
+            # Since this is a OCI model, prepend oci://
+            args.MODEL = f"oci://{args.MODEL}"
+
         except Exception:
             raise e
 
@@ -1432,27 +1445,42 @@ def rm_parser(subparsers):
     parser.set_defaults(func=rm_cli)
 
 
+def _rm_oci_model(model, args) -> bool:
+    # attempt to remove as a container image
+    try:
+        m = TransportFactory(model, args, ignore_stderr=True).create_oci()
+        return m.remove(args)
+    except Exception:
+        return False
+
+
 def _rm_model(models, args):
+    exceptions = []
     for model in models:
         model = shortnames.resolve(model)
 
         try:
             m = New(model, args)
-            m.remove(args)
-        except KeyError as e:
+            if m.remove(args):
+                continue
+            # Failed to remove and might be OCI so attempt to remove OCI
+            if args.ignore:
+                _rm_oci_model(model, args)
+                continue
+        except (KeyError, subprocess.CalledProcessError) as e:
             for prefix in MODEL_TYPES:
                 if model.startswith(prefix + "://"):
                     if not args.ignore:
                         raise e
-            try:
-                # attempt to remove as a container image
-                m = TransportFactory(model, args, ignore_stderr=True).create_oci()
-                m.remove(args)
-                return
-            except Exception:
-                pass
-            if not args.ignore:
-                raise e
+            # attempt to remove as a container image
+            if _rm_oci_model(model, args) or args.ignore:
+                continue
+            exceptions.append(e)
+
+    if len(exceptions) > 0:
+        for exception in exceptions[1:]:
+            perror("Error: " + str(exception).strip("'\""))
+        raise exceptions[0]
 
 
 def rm_cli(args):
@@ -1512,7 +1540,7 @@ def inspect_parser(subparsers):
 def inspect_cli(args):
     args.pull = "never"
     model = New(args.MODEL, args)
-    model.inspect(args.all, args.get == "all", args.get, args.json, args.dryrun)
+    print(model.inspect(args.all, args.get == "all", args.get, args.json, args.dryrun))
 
 
 def main() -> None:
@@ -1544,9 +1572,11 @@ def eprint(e, exit_code):
         args.func(args)
     except urllib.error.HTTPError as e:
         eprint(f"pulling {e.geturl()} failed: {e}", errno.EINVAL)
+    except FileNotFoundError as e:
+        eprint(e, errno.ENOENT)
     except HelpException:
         parser.print_help()
-    except (ConnectionError, IndexError, KeyError, ValueError, NoRefFileFound) as e:
+    except (IsADirectoryError, ConnectionError, IndexError, KeyError, ValueError, NoRefFileFound) as e:
         eprint(e, errno.EINVAL)
     except NotImplementedError as e:
         eprint(e, errno.ENOSYS)

ramalama/common.py

@@ -283,7 +283,7 @@ def verify_checksum(filename: str) -> bool:
 
 
 def genname():
-    return "ramalama_" + "".join(random.choices(string.ascii_letters + string.digits, k=10))
+    return "ramalama-" + "".join(random.choices(string.ascii_letters + string.digits, k=10))
 
 
 def engine_version(engine: SUPPORTED_ENGINES) -> str:

ramalama/config.py

@@ -221,6 +221,7 @@ class BaseConfig:
     carimage: str = "registry.access.redhat.com/ubi10-micro:latest"
     container: bool = None  # type: ignore
     ctx_size: int = 0
+    convert_type: Literal["artifact", "car", "raw"] = "raw"
     default_image: str = DEFAULT_IMAGE
     default_rag_image: str = DEFAULT_RAG_IMAGE
     dryrun: bool = False

ramalama/kube.py

@@ -2,7 +2,7 @@
 import platform
 from typing import Optional, Tuple
 
-from ramalama.common import MNT_DIR, RAG_DIR, genname, get_accel_env_vars
+from ramalama.common import MNT_DIR, RAG_DIR, get_accel_env_vars
 from ramalama.file import PlainFile
 from ramalama.path_utils import normalize_host_path_for_container
 from ramalama.version import version
@@ -17,6 +17,7 @@ def __init__(
         mmproj_paths: Optional[Tuple[str, str]],
         args,
         exec_args,
+        artifact: bool,
     ):
         self.src_model_path, self.dest_model_path = model_paths
         self.src_chat_template_path, self.dest_chat_template_path = (
@@ -29,27 +30,30 @@ def __init__(
         if getattr(args, "name", None):
             self.name = args.name
         else:
-            self.name = genname()
+            self.name = "ramalama"
 
         self.args = args
         self.exec_args = exec_args
         self.image = args.image
+        self.artifact = artifact
 
     def _gen_volumes(self):
         mounts = """\
         volumeMounts:"""
 
         volumes = """
       volumes:"""
-
         if os.path.exists(self.src_model_path):
             m, v = self._gen_path_volume()
             mounts += m
             volumes += v
         else:
+            subPath = ""
+            if not self.artifact:
+                subPath = """
+          subPath: /models"""
             mounts += f"""
-        - mountPath: {MNT_DIR}
-          subPath: /models
+        - mountPath: {MNT_DIR}{subPath}
           name: model"""
             volumes += self._gen_oci_volume()
 
@@ -104,7 +108,7 @@ def _gen_path_volume(self):
     def _gen_oci_volume(self):
         return f"""
       - image:
-          reference: {self.ai_image}
+          reference: {self.src_model_path}
           pullPolicy: IfNotPresent
         name: model"""
 
@@ -176,7 +180,7 @@ def __gen_env_vars():
         for k, v in env_vars.items():
             env_spec += f"""
         - name: {k}
-          value: {v}"""
+          value: \"{v}\""""
 
         return env_spec
 
@@ -191,7 +195,7 @@ def generate(self) -> PlainFile:
 # it into Kubernetes.
 #
 # Created with ramalama-{_version}
-apiVersion: v1
+apiVersion: apps/v1
 kind: Deployment
 metadata:
   name: {self.name}