add --version to llama stack CLI & /version endpoint (#732)

# What does this PR do?

- add --version to llama stack CLI 
- add /version endpoint
- run OpenAPI generator for the new endpoint

## Test Plan

**CLI**
<img width="184" alt="image"
src="https://github.com/user-attachments/assets/3acb1d22-453e-4b79-baf6-e98e88d0671c"
/>



**endpoint**
<img width="430" alt="image"
src="https://github.com/user-attachments/assets/79cdd670-493b-40cf-8f9e-28a4ac0988ac"
/>


## Sources

Please link relevant resources if necessary.


## Before submitting

- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Ran pre-commit to handle lint / formatting issues.
- [ ] Read the [contributor
guideline](https://github.com/meta-llama/llama-stack/blob/main/CONTRIBUTING.md),
      Pull Request section?
- [ ] Updated relevant documentation.
- [ ] Wrote necessary unit or integration tests.

docs/resources/llama-stack-spec.html

@@ -2467,6 +2467,36 @@
                     "required": true
                 }
             }
+        },
+        "/alpha/version": {
+            "get": {
+                "responses": {
+                    "200": {
+                        "description": "OK",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/VersionInfo"
+                                }
+                            }
+                        }
+                    }
+                },
+                "tags": [
+                    "Inspect"
+                ],
+                "parameters": [
+                    {
+                        "name": "X-LlamaStack-ProviderData",
+                        "in": "header",
+                        "description": "JSON-encoded provider data which will be made available to the adapter servicing the API",
+                        "required": false,
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                ]
+            }
         }
     },
     "jsonSchemaDialect": "https://json-schema.org/draft/2020-12/schema",
@@ -6457,6 +6487,9 @@
                     "gradient_accumulation_steps": {
                         "type": "integer"
                     },
+                    "max_validation_steps": {
+                        "type": "integer"
+                    },
                     "data_config": {
                         "$ref": "#/components/schemas/DataConfig"
                     },
@@ -6476,6 +6509,7 @@
                     "n_epochs",
                     "max_steps_per_epoch",
                     "gradient_accumulation_steps",
+                    "max_validation_steps",
                     "data_config",
                     "optimizer_config"
                 ]
@@ -7686,6 +7720,18 @@
                 "required": [
                     "model_id"
                 ]
+            },
+            "VersionInfo": {
+                "type": "object",
+                "properties": {
+                    "version": {
+                        "type": "string"
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "version"
+                ]
             }
         },
         "responses": {}
@@ -8382,6 +8428,10 @@
             "name": "VectorMemoryBankParams",
             "description": "<SchemaDefinition schemaRef=\"#/components/schemas/VectorMemoryBankParams\" />"
         },
+        {
+            "name": "VersionInfo",
+            "description": "<SchemaDefinition schemaRef=\"#/components/schemas/VersionInfo\" />"
+        },
         {
             "name": "ViolationLevel",
             "description": "<SchemaDefinition schemaRef=\"#/components/schemas/ViolationLevel\" />"
@@ -8576,6 +8626,7 @@
                 "UserMessage",
                 "VectorMemoryBank",
                 "VectorMemoryBankParams",
+                "VersionInfo",
                 "ViolationLevel",
                 "WolframAlphaToolDefinition"
             ]

docs/resources/llama-stack-spec.yaml

@@ -3002,6 +3002,8 @@ components:
           type: integer
         max_steps_per_epoch:
           type: integer
+        max_validation_steps:
+          type: integer
         n_epochs:
           type: integer
         optimizer_config:
@@ -3010,6 +3012,7 @@ components:
       - n_epochs
       - max_steps_per_epoch
       - gradient_accumulation_steps
+      - max_validation_steps
       - data_config
       - optimizer_config
       type: object
@@ -3192,6 +3195,14 @@ components:
       - embedding_model
       - chunk_size_in_tokens
       type: object
+    VersionInfo:
+      additionalProperties: false
+      properties:
+        version:
+          type: string
+      required:
+      - version
+      type: object
     ViolationLevel:
       enum:
       - info
@@ -4731,6 +4742,25 @@ paths:
           description: OK
       tags:
       - Telemetry
+  /alpha/version:
+    get:
+      parameters:
+      - description: JSON-encoded provider data which will be made available to the
+          adapter servicing the API
+        in: header
+        name: X-LlamaStack-ProviderData
+        required: false
+        schema:
+          type: string
+      responses:
+        '200':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/VersionInfo'
+          description: OK
+      tags:
+      - Inspect
 security:
 - Default: []
 servers:
@@ -5225,6 +5255,8 @@ tags:
 - description: <SchemaDefinition schemaRef="#/components/schemas/VectorMemoryBankParams"
     />
   name: VectorMemoryBankParams
+- description: <SchemaDefinition schemaRef="#/components/schemas/VersionInfo" />
+  name: VersionInfo
 - description: <SchemaDefinition schemaRef="#/components/schemas/ViolationLevel" />
   name: ViolationLevel
 - description: <SchemaDefinition schemaRef="#/components/schemas/WolframAlphaToolDefinition"
@@ -5411,5 +5443,6 @@ x-tagGroups:
   - UserMessage
   - VectorMemoryBank
   - VectorMemoryBankParams
+  - VersionInfo
   - ViolationLevel
   - WolframAlphaToolDefinition

llama_stack/apis/inspect/inspect.py

@@ -29,6 +29,11 @@ class HealthInfo(BaseModel):
     # TODO: add a provider level status
 
 
+@json_schema_type
+class VersionInfo(BaseModel):
+    version: str
+
+
 @runtime_checkable
 class Inspect(Protocol):
     @webmethod(route="/providers/list", method="GET")
@@ -39,3 +44,6 @@ class Inspect(Protocol):
 
     @webmethod(route="/health", method="GET")
     async def health(self) -> HealthInfo: ...
+
+    @webmethod(route="/version", method="GET")
+    async def version(self) -> VersionInfo: ...

llama_stack/cli/stack/stack.py

@@ -5,6 +5,7 @@
 # the root directory of this source tree.
 
 import argparse
+from importlib.metadata import version
 
 from llama_stack.cli.subcommand import Subcommand
 
@@ -24,6 +25,12 @@ class StackParser(Subcommand):
             description="Operations for the Llama Stack / Distributions",
         )
 
+        self.parser.add_argument(
+            "--version",
+            action="version",
+            version=f"{version('llama-stack')}",
+        )
+
         subparsers = self.parser.add_subparsers(title="stack_subcommands")
 
         # Add sub-commands

llama_stack/distribution/inspect.py

@@ -4,11 +4,18 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+from importlib.metadata import version
 from typing import Dict, List
 
 from pydantic import BaseModel
 
-from llama_stack.apis.inspect import HealthInfo, Inspect, ProviderInfo, RouteInfo
+from llama_stack.apis.inspect import (
+    HealthInfo,
+    Inspect,
+    ProviderInfo,
+    RouteInfo,
+    VersionInfo,
+)
 from llama_stack.distribution.datatypes import StackRunConfig
 from llama_stack.distribution.server.endpoints import get_all_api_endpoints
 
@@ -65,3 +72,6 @@ class DistributionInspectImpl(Inspect):
 
     async def health(self) -> HealthInfo:
         return HealthInfo(status="OK")
+
+    async def version(self) -> VersionInfo:
+        return VersionInfo(version=version("llama-stack"))