Merge pull request #572 from Abirdcfly/maxtoken

feat: add maxTokens in apiserver
kubeagi · Jan 17, 2024 · 8ff68ab · 8ff68ab
2 parents c046903 + 59c59c9
commit 8ff68ab
Show file tree

Hide file tree

Showing 13 changed files with 105 additions and 2 deletions.
diff --git a/api/app-node/chain/v1alpha1/llmchain_types.go b/api/app-node/chain/v1alpha1/llmchain_types.go
@@ -38,6 +38,7 @@ type CommonChainConfig struct {
 	// Usually this value is just empty
 	Model string `json:"model,omitempty"`
 	// MaxTokens is the maximum number of tokens to generate to use in a llm call.
+	// +kubebuilder:default=1024
 	MaxTokens int `json:"maxTokens,omitempty"`
 	// Temperature is the temperature for sampling to use in a llm call, between 0 and 1.
 	//+kubebuilder:validation:Minimum=0

diff --git a/apiserver/graph/generated/generated.go b/apiserver/graph/generated/generated.go
diff --git a/apiserver/graph/generated/models_gen.go b/apiserver/graph/generated/models_gen.go
diff --git a/apiserver/graph/schema/application.gql b/apiserver/graph/schema/application.gql
@@ -57,6 +57,7 @@ mutation updateApplicationConfig($input: UpdateApplicationConfigInput!){
             llm
             temperature
             maxLength
+            maxTokens
             conversionWindowSize
             knowledgebase
             scoreThreshold
@@ -91,6 +92,7 @@ query getApplication($name: String!, $namespace: String!){
             llm
             temperature
             maxLength
+            maxTokens
             conversionWindowSize
             knowledgebase
             scoreThreshold

diff --git a/apiserver/graph/schema/application.graphqls b/apiserver/graph/schema/application.graphqls
@@ -49,6 +49,11 @@ type Application {
     """
     maxLength: Int
 
+    """
+    maxTokens 最大输出token
+    """
+    maxTokens: Int
+
     """
     conversionWindowSize 对话轮次
     """
@@ -277,6 +282,11 @@ input UpdateApplicationConfigInput {
     """
     maxLength: Int
 
+    """
+    maxTokens 最大输出token
+    """
+    maxTokens: Int
+
     """
     conversionWindowSize 对话轮次
     """

diff --git a/apiserver/pkg/application/application.go b/apiserver/pkg/application/application.go
@@ -48,6 +48,7 @@ func addDefaultValue(gApp *generated.Application, app *v1alpha1.Application) {
 	gApp.ScoreThreshold = pointer.Float64(0.3)
 	gApp.Temperature = pointer.Float64(0.7)
 	gApp.MaxLength = pointer.Int(1024)
+	gApp.MaxTokens = pointer.Int(1024)
 	gApp.ConversionWindowSize = pointer.Int(5)
 }
 
@@ -85,6 +86,7 @@ func cr2app(prompt *apiprompt.Prompt, chainConfig *apichain.CommonChainConfig, r
 		gApp.Model = pointer.String(chainConfig.Model)
 		gApp.Temperature = pointer.Float64(chainConfig.Temperature)
 		gApp.MaxLength = pointer.Int(chainConfig.MaxLength)
+		gApp.MaxTokens = pointer.Int(chainConfig.MaxTokens)
 		gApp.ConversionWindowSize = pointer.Int(chainConfig.Memory.ConversionWindowSize)
 	}
 	for _, node := range app.Spec.Nodes {
@@ -367,13 +369,15 @@ func UpdateApplicationConfig(ctx context.Context, c dynamic.Interface, input gen
 					},
 					Model:       pointer.StringDeref(input.Model, ""),
 					MaxLength:   pointer.IntDeref(input.MaxLength, 0),
+					MaxTokens:   pointer.IntDeref(input.MaxTokens, 0),
 					Temperature: pointer.Float64Deref(input.Temperature, 0),
 				},
 			},
 		}
 		if err = createOrUpdateResource(ctx, c, common.SchemaOf(&common.ArcadiaAPIGroup, strings.ToLower(chainKind)), input.Namespace, input.Name, func() {
 			qachain.Spec.Model = pointer.StringDeref(input.Model, qachain.Spec.Model)
 			qachain.Spec.MaxLength = pointer.IntDeref(input.MaxLength, qachain.Spec.MaxLength)
+			qachain.Spec.MaxTokens = pointer.IntDeref(input.MaxTokens, qachain.Spec.MaxTokens)
 			qachain.Spec.Temperature = pointer.Float64Deref(input.Temperature, qachain.Spec.Temperature)
 			qachain.Spec.Memory.ConversionWindowSize = pointer.IntDeref(input.ConversionWindowSize, qachain.Spec.Memory.ConversionWindowSize)
 		}, qachain); err != nil {
@@ -401,13 +405,15 @@ func UpdateApplicationConfig(ctx context.Context, c dynamic.Interface, input gen
 					},
 					Model:       pointer.StringDeref(input.Model, ""),
 					MaxLength:   pointer.IntDeref(input.MaxLength, 0),
+					MaxTokens:   pointer.IntDeref(input.MaxTokens, 0),
 					Temperature: pointer.Float64Deref(input.Temperature, 0),
 				},
 			},
 		}
 		if err = createOrUpdateResource(ctx, c, common.SchemaOf(&common.ArcadiaAPIGroup, strings.ToLower(chainKind)), input.Namespace, input.Name, func() {
 			llmchain.Spec.Model = pointer.StringDeref(input.Model, llmchain.Spec.Model)
 			llmchain.Spec.MaxLength = pointer.IntDeref(input.MaxLength, llmchain.Spec.MaxLength)
+			llmchain.Spec.MaxTokens = pointer.IntDeref(input.MaxTokens, llmchain.Spec.MaxTokens)
 			llmchain.Spec.Temperature = pointer.Float64Deref(input.Temperature, llmchain.Spec.Temperature)
 			llmchain.Spec.Memory.ConversionWindowSize = pointer.IntDeref(input.ConversionWindowSize, llmchain.Spec.Memory.ConversionWindowSize)
 		}, llmchain); err != nil {

diff --git a/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_apichains.yaml b/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_apichains.yaml
@@ -56,6 +56,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer

diff --git a/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_llmchains.yaml b/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_llmchains.yaml
@@ -52,6 +52,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer

diff --git a/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_retrievalqachains.yaml b/config/crd/bases/chain.arcadia.kubeagi.k8s.com.cn_retrievalqachains.yaml
@@ -55,6 +55,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer

diff --git a/deploy/charts/arcadia/Chart.yaml b/deploy/charts/arcadia/Chart.yaml
@@ -2,7 +2,7 @@ apiVersion: v2
 name: arcadia
 description: A Helm chart(KubeBB Component) for KubeAGI Arcadia
 type: application
-version: 0.2.13
+version: 0.2.14
 appVersion: "0.1.0"
 
 keywords:

diff --git a/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_apichains.yaml b/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_apichains.yaml
@@ -56,6 +56,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer

diff --git a/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_llmchains.yaml b/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_llmchains.yaml
@@ -52,6 +52,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer

diff --git a/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_retrievalqachains.yaml b/deploy/charts/arcadia/crds/chain.arcadia.kubeagi.k8s.com.cn_retrievalqachains.yaml
@@ -55,6 +55,7 @@ spec:
                 minimum: 10
                 type: integer
               maxTokens:
+                default: 1024
                 description: MaxTokens is the maximum number of tokens to generate
                   to use in a llm call.
                 type: integer