Restore top_k config option in VLM app

Upstream vLLM bug seems to have been fixed and vLLM no longer
returns a HTTP 400 when top_k parameter is included.

Author: sd109

charts/azimuth-image-analysis/azimuth-ui.schema.yaml

@@ -28,7 +28,5 @@ sortOrder:
   - /azimuth-llm/ui/appSettings/llm_params/frequency_penalty
   - /azimuth-llm/ui/appSettings/llm_params/presence_penalty
   - /azimuth-llm/ui/appSettings/llm_params/top_p
-  # vLLM responds with HTTP 400 BadRequest when top_k is
-  # passed to a vision model (but ollama accepts it)
-  # - /azimuth-llm/ui/appSettings/llm_params/top_k
+  - /azimuth-llm/ui/appSettings/llm_params/top_k
   - /azimuth-llm/api/modelMaxContextLength

charts/azimuth-image-analysis/values.schema.json

@@ -89,6 +89,13 @@
                                             "exclusiveMinimum": 0,
                                             "maximum": 1
                                         },
+                                        "top_k": {
+                                            "type": "integer",
+                                            "title": "LLM Top K",
+                                            "description": "The [top k](https://docs.vllm.ai/en/stable/dev/sampling_params.html) value to use when generating LLM responses (must be an integer).",
+                                            "default": -1,
+                                            "minimum": -1
+                                        },
                                         "presence_penalty": {
                                             "type": "number",
                                             "title": "LLM Presence Penalty",