feat: introduce API leveling, post_training, eval to v1alpha (#3449)

# What does this PR do?

Rather than have a single `LLAMA_STACK_VERSION`, we need to have a
`_V1`, `_V1ALPHA`, and `_V1BETA` constant.

This also necessitated addition of `level` to the `WebMethod` so that
routing can be handeled properly.


For backwards compat, the `v1` routes are being kept around and marked
as `deprecated`. When used, the server will log a deprecation warning.

Deprecation log:

<img width="1224" height="134" alt="Screenshot 2025-09-25 at 2 43 36 PM"
src="https://github.com/user-attachments/assets/0cc7c245-dafc-48f0-be99-269fb9a686f9"
/>

move:
1. post_training to `v1alpha` as it is under heavy development and not
near its final state
2. eval: job scheduling is not implemented. Relies heavily on the
datasetio API which is under development missing implementations of
specific routes indicating the structure of those routes might change.
Additionally eval depends on the `inference` API which is going to be
deprecated, eval will likely need a major API surface change to conform
to using completions properly

implements leveling in #3317 

note: integration tests will fail until the SDK is regenerated with
v1alpha/inference as opposed to v1/inference

## Test Plan

existing tests should pass with newly generated schema. Conformance will
also pass as these routes are not the ones we currently test for
stability

Signed-off-by: Charlie Doern <cdoern@redhat.com>

llama_stack/core/client.py

@@ -15,7 +15,6 @@ import httpx
 from pydantic import BaseModel, parse_obj_as
 from termcolor import cprint
 
-from llama_stack.apis.version import LLAMA_STACK_API_VERSION
 from llama_stack.providers.datatypes import RemoteProviderConfig
 
 _CLIENT_CLASSES = {}
@@ -114,7 +113,24 @@ def create_api_client_class(protocol) -> type:
                     break
                 kwargs[param.name] = args[i]
 
-            url = f"{self.base_url}/{LLAMA_STACK_API_VERSION}/{webmethod.route.lstrip('/')}"
+            # Get all webmethods for this method (supports multiple decorators)
+            webmethods = getattr(method, "__webmethods__", [])
+
+            if not webmethods:
+                raise RuntimeError(f"Method {method} has no webmethod decorators")
+
+            # Choose the preferred webmethod (non-deprecated if available)
+            preferred_webmethod = None
+            for wm in webmethods:
+                if not getattr(wm, "deprecated", False):
+                    preferred_webmethod = wm
+                    break
+
+            # If no non-deprecated found, use the first one
+            if preferred_webmethod is None:
+                preferred_webmethod = webmethods[0]
+
+            url = f"{self.base_url}/{preferred_webmethod.level}/{preferred_webmethod.route.lstrip('/')}"
 
             def convert(value):
                 if isinstance(value, list):

llama_stack/core/server/routes.py

@@ -14,7 +14,6 @@ from starlette.routing import Route
 
 from llama_stack.apis.datatypes import Api, ExternalApiSpec
 from llama_stack.apis.tools import RAGToolRuntime, SpecialToolGroup
-from llama_stack.apis.version import LLAMA_STACK_API_VERSION
 from llama_stack.core.resolver import api_protocol_map
 from llama_stack.schema_utils import WebMethod
 
@@ -54,22 +53,23 @@ def get_all_api_routes(
                     protocol_methods.append((f"{tool_group.value}.{name}", method))
 
         for name, method in protocol_methods:
-            if not hasattr(method, "__webmethod__"):
+            # Get all webmethods for this method (supports multiple decorators)
+            webmethods = getattr(method, "__webmethods__", [])
+            if not webmethods:
                 continue
 
-            # The __webmethod__ attribute is dynamically added by the @webmethod decorator
-            # mypy doesn't know about this dynamic attribute, so we ignore the attr-defined error
-            webmethod = method.__webmethod__  # type: ignore[attr-defined]
-            path = f"/{LLAMA_STACK_API_VERSION}/{webmethod.route.lstrip('/')}"
-            if webmethod.method == hdrs.METH_GET:
-                http_method = hdrs.METH_GET
-            elif webmethod.method == hdrs.METH_DELETE:
-                http_method = hdrs.METH_DELETE
-            else:
-                http_method = hdrs.METH_POST
-            routes.append(
-                (Route(path=path, methods=[http_method], name=name, endpoint=None), webmethod)
-            )  # setting endpoint to None since don't use a Router object
+            # Create routes for each webmethod decorator
+            for webmethod in webmethods:
+                path = f"/{webmethod.level}/{webmethod.route.lstrip('/')}"
+                if webmethod.method == hdrs.METH_GET:
+                    http_method = hdrs.METH_GET
+                elif webmethod.method == hdrs.METH_DELETE:
+                    http_method = hdrs.METH_DELETE
+                else:
+                    http_method = hdrs.METH_POST
+                routes.append(
+                    (Route(path=path, methods=[http_method], name=name, endpoint=None), webmethod)
+                )  # setting endpoint to None since don't use a Router object
 
         apis[api] = routes
 

llama_stack/core/server/tracing.py

@@ -45,6 +45,14 @@ class TracingMiddleware:
             logger.debug(f"No matching route found for path: {path}, falling back to FastAPI")
             return await self.app(scope, receive, send)
 
+        # Log deprecation warning if route is deprecated
+        if getattr(webmethod, "deprecated", False):
+            logger.warning(
+                f"DEPRECATED ROUTE USED: {scope.get('method', 'GET')} {path} - "
+                f"This route is deprecated and may be removed in a future version. "
+                f"Please check the docs for the supported version."
+            )
+
         trace_attributes = {"__location__": "server", "raw_path": path}
 
         # Extract W3C trace context headers and store as trace attributes