create openai deployments with default capacity as per model list

Azure · Aug 26, 2023 · b6b137e · b6b137e
1 parent 5e7d295
commit b6b137e
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 6 deletions.
diff --git a/src/common/details/azcli/AzCli.cs b/src/common/details/azcli/AzCli.cs
@@ -63,6 +63,7 @@ public struct CognitiveServicesModelInfo
             public string Name { get; set; }
             public string Format { get; set; }
             public string Version { get; set; }
+            public string DefaultCapacity { get; set; }
         }
 
         public struct CognitiveSearchResourceInfo
@@ -245,7 +246,7 @@ public static async Task<ProcessResponse<CognitiveServicesModelInfo[]>> ListCogn
             var cmdPart = "cognitiveservices model list";
             var subPart = subscriptionId != null ? $"--subscription {subscriptionId}" : "";
 
-            var process = await ProcessHelpers.ParseShellCommandJson<JArray>("az", $"{cmdPart} {subPart} -l {regionLocation} --query \"[].{{Name:model.name,Format:model.format,Version:model.version}}\"", GetAzureHttpUserAgentEnvironment());
+            var process = await ProcessHelpers.ParseShellCommandJson<JArray>("az", $"{cmdPart} {subPart} -l {regionLocation} --query \"[].{{Name:model.name,Format:model.format,Version:model.version,DefaultCapacity:model.skus[0].capacity.default}}\"", GetAzureHttpUserAgentEnvironment());
 
             var x = new ProcessResponse<CognitiveServicesModelInfo[]>();
             x.StdOutput = process.StdOutput;
@@ -260,6 +261,7 @@ public static async Task<ProcessResponse<CognitiveServicesModelInfo[]>> ListCogn
                 x.Payload[i].Name = model["Name"].Value<string>();
                 x.Payload[i].Format = model["Format"].Value<string>();
                 x.Payload[i].Version = model["Version"].Value<string>();
+                x.Payload[i].DefaultCapacity = model["DefaultCapacity"].Value<string>();
                 i++;
             }
 
@@ -313,12 +315,12 @@ public static async Task<ProcessResponse<ResourceGroupInfo>> CreateResourceGroup
             return x;
         }
 
-        public static async Task<ProcessResponse<CognitiveServicesDeploymentInfo>> CreateCognitiveServicesDeployment(string subscriptionId, string group, string resourceName, string deploymentName, string modelName, string modelVersion, string modelFormat)
+        public static async Task<ProcessResponse<CognitiveServicesDeploymentInfo>> CreateCognitiveServicesDeployment(string subscriptionId, string group, string resourceName, string deploymentName, string modelName, string modelVersion, string modelFormat, string scaleCapacity)
         {
             var cmdPart = "cognitiveservices account deployment create";
             var subPart = subscriptionId != null ? $"--subscription {subscriptionId}" : "";
 
-            var process = await ProcessHelpers.ParseShellCommandJson<JObject>("az", $"{cmdPart} {subPart} -g {group} -n {resourceName} --deployment-name {deploymentName} --model-name {modelName} --model-version {modelVersion} --model-format {modelFormat} --sku-capacity 1 --sku-name \"Standard\"", GetAzureHttpUserAgentEnvironment());
+            var process = await ProcessHelpers.ParseShellCommandJson<JObject>("az", $"{cmdPart} {subPart} -g {group} -n {resourceName} --deployment-name {deploymentName} --model-name {modelName} --model-version {modelVersion} --model-format {modelFormat} --sku-capacity {scaleCapacity} --sku-name \"Standard\"", GetAzureHttpUserAgentEnvironment());
 
             var x = new ProcessResponse<CognitiveServicesDeploymentInfo>();
             x.StdOutput = process.StdOutput;

diff --git a/src/common/details/azcli/AzCliConsoleGui_AiDeploymentPicker.cs b/src/common/details/azcli/AzCliConsoleGui_AiDeploymentPicker.cs
@@ -106,15 +106,22 @@ public class AiResourceDeploymentPicker
                 Console.Write("\rModel: ");
                 var choices = models.Payload.Select(x => x.Name + " (version " + x.Version + ")").ToArray();
 
-                var select = Array.FindIndex(choices, x => x.StartsWith("gpt"));
+                var scanFor = deploymentExtra.ToLower() switch {
+                    "chat" => "gpt",
+                    "embeddings" => "embedding",
+                    _ => deploymentExtra.ToLower()
+                };
+                var select = Math.Max(0, Array.FindIndex(choices, x => x.Contains(scanFor)));
+
                 var index = ListBoxPicker.PickIndexOf(choices, 60, 30, normal, selected, select);
                 if (index < 0) return (null, null);
 
                 var modelName = models.Payload[index].Name;
                 Console.WriteLine($"\rModel: {modelName}");
 
-                var modelVersion = modelName.Contains("gpt") ? "0613" : "2";
                 var modelFormat = "OpenAI";
+                var modelVersion = models.Payload[index].Version;
+                var scaleCapacity = models.Payload[index].DefaultCapacity;
 
                 Console.Write("\rName: ");
                 choices = new string[] {
@@ -134,7 +141,7 @@ public class AiResourceDeploymentPicker
                 if (pick != choices.Length - 1) Console.WriteLine($"\rName: {deploymentName}");
 
                 Console.Write("*** CREATING ***");
-                var response = await AzCli.CreateCognitiveServicesDeployment(subscriptionId, resource.Group, resource.Name, deploymentName, modelName, modelVersion, modelFormat);
+                var response = await AzCli.CreateCognitiveServicesDeployment(subscriptionId, resource.Group, resource.Name, deploymentName, modelName, modelVersion, modelFormat, scaleCapacity);
 
                 Console.Write("\r");
                 if (string.IsNullOrEmpty(response.StdOutput) && !string.IsNullOrEmpty(response.StdError))