Merge pull request #762 from bjwswang/release-v0.2.0

add more info to models

api/base/v1alpha1/model.go

@@ -28,7 +28,10 @@ const (
 	// LabelModelEmbedding indicates this is a embedding model
 	LabelModelEmbedding = Group + "/embedding"
 	// LabelModelLLM indicates this is a llm model
-	LabelModelLLM      = Group + "/llm"
+	LabelModelLLM = Group + "/llm"
+	// LabelModelReranking indicates this is a reranking model
+	LabelModelReranking = Group + "/reranking"
+	// LabelModelFullPath indicates the full path in storage
 	LabelModelFullPath = Group + "/full-path"
 )
 
@@ -52,6 +55,16 @@ func (model Model) IsEmbeddingModel() bool {
 	return false
 }
 
+// IsRerankingModel checks whether this model is a reranking model
+func (model Model) IsRerankingModel() bool {
+	for _, t := range strings.Split(model.Spec.Types, ",") {
+		if strings.ToLower(t) == "reranking" {
+			return true
+		}
+	}
+	return false
+}
+
 // FullPath with bucket and object path
 func (model Model) FullPath() string {
 	return fmt.Sprintf("%s/%s", model.Namespace, model.ObjectPath())

api/base/v1alpha1/model_types.go

@@ -33,6 +33,14 @@ type ModelSpec struct {
 
 	// Source define the source of the model file
 	Source *TypedObjectReference `json:"source,omitempty"`
+
+	// HuggingFaceRepo defines the huggingface repo which hosts this model
+	HuggingFaceRepo string `json:"huggingFaceRepo,omitempty"`
+	// ModelScopeRepo defines the modelscope repo which hosts this model
+	ModelScopeRepo string `json:"modelScopeRepo,omitempty"`
+
+	// MaxContextLength defines the max context length allowed in this model
+	MaxContextLength int `json:"maxContextLength,omitempty"`
 }
 
 // ModelStatus defines the observed state of Model

config/crd/bases/arcadia.kubeagi.k8s.com.cn_models.yaml

@@ -44,6 +44,18 @@ spec:
               displayName:
                 description: DisplayName defines datasource display name
                 type: string
+              huggingFaceRepo:
+                description: HuggingFaceRepo defines the huggingface repo which hosts
+                  this model
+                type: string
+              maxContextLength:
+                description: MaxContextLength defines the max context length allowed
+                  in this model
+                type: integer
+              modelScopeRepo:
+                description: ModelScopeRepo defines the modelscope repo which hosts
+                  this model
+                type: string
               source:
                 description: Source define the source of the model file
                 properties:

controllers/base/model_controller.go

@@ -150,6 +150,11 @@ func (r *ModelReconciler) Initialize(ctx context.Context, logger logr.Logger, in
 		instanceDeepCopy.Labels[arcadiav1alpha1.LabelModelLLM] = isLLMModel
 		update = true
 	}
+	isReranking := strconv.FormatBool(instanceDeepCopy.IsRerankingModel())
+	if v := instanceDeepCopy.Labels[arcadiav1alpha1.LabelModelReranking]; v != isReranking {
+		instanceDeepCopy.Labels[arcadiav1alpha1.LabelModelReranking] = isReranking
+		update = true
+	}
 
 	// Initialize annotations
 	if instanceDeepCopy.Annotations == nil {

deploy/charts/arcadia/Chart.yaml

@@ -2,7 +2,7 @@ apiVersion: v2
 name: arcadia
 description: A Helm chart(Also a KubeBB Component) for KubeAGI Arcadia
 type: application
-version: 0.2.34
+version: 0.2.35
 appVersion: "0.1.0"
 
 keywords:

deploy/charts/arcadia/crds/arcadia.kubeagi.k8s.com.cn_models.yaml

@@ -44,6 +44,18 @@ spec:
               displayName:
                 description: DisplayName defines datasource display name
                 type: string
+              huggingFaceRepo:
+                description: HuggingFaceRepo defines the huggingface repo which hosts
+                  this model
+                type: string
+              maxContextLength:
+                description: MaxContextLength defines the max context length allowed
+                  in this model
+                type: integer
+              modelScopeRepo:
+                description: ModelScopeRepo defines the modelscope repo which hosts
+                  this model
+                type: string
               source:
                 description: Source define the source of the model file
                 properties:

deploy/charts/arcadia/templates/post-models.yaml

@@ -20,11 +20,13 @@ spec:
     官网: https://www.baichuan-ai.com/
     Github: https://github.com/baichuan-inc/Baichuan2
     HuggingFace: https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat
-    ModelScope: https://modelscope.cn/models/baichuan-inc/Baichuan2-7B-Chat/summary
+    ModelScope: https://modelscope.cn/models/baichuan-inc/Baichuan2-7B-Chat
     arXiv: https://arxiv.org/abs/2309.10305
 
     百川智能成立于2023年4月10日,由前搜狗公司CEO王小川创立。公司以帮助大众轻松、普惠地获取世界知识和专业服务为使命,致力于通过语言AI的突破,构建中国最优秀的大模型底座。
   types: "llm"
+  huggingFaceRepo: baichuan-inc/Baichuan2-7B-Chat
+  modelScopeRepo: baichuan-inc/Baichuan2-7B-Chat
 ---
 apiVersion: arcadia.kubeagi.k8s.com.cn/v1alpha1
 kind: Model
@@ -56,6 +58,8 @@ spec:
 
     北京智谱华章科技有限公司由清华大学计算机系技术成果转化而来,致力于打造新一代认知智能大模型,专注于做大模型的中国创新。
   types: "llm"
+  huggingFaceRepo: THUDM/chatglm2-6b
+  modelScopeRepo: ZhipuAI/chatglm2-6b
 ---
 apiVersion: arcadia.kubeagi.k8s.com.cn/v1alpha1
 kind: Model
@@ -88,6 +92,8 @@ spec:
 
     通义为阿里云大模型的统一品牌,覆盖语言、听觉、多模态等领域,致力于实现类人智慧的通用智能。
   types: "llm"
+  huggingFaceRepo: Qwen/Qwen-7B-Chat
+  modelScopeRepo: qwen/Qwen-7B-Chat
 ---
 apiVersion: arcadia.kubeagi.k8s.com.cn/v1alpha1
 kind: Model
@@ -120,4 +126,26 @@ spec:
 
     北京智源人工智能研究院是北京大学的直属研究机构,主要从事人工智能的数理基础、机器学习、智能信息检索与挖掘、智能体系架构与芯片、自然语言处理等领域研究。
   types: "embedding"
+  huggingFaceRepo: BAAI/bge-large-zh-v1.5
+---
+apiVersion: arcadia.kubeagi.k8s.com.cn/v1alpha1
+kind: Model
+metadata:
+  name: bge-reranker-large
+  namespace: {{ .Release.Namespace }}
+  annotations:
+    "helm.sh/hook": post-install
+    "helm.sh/hook-weight": "1"
+spec:
+  displayName: "bge-reranker-large"
+  description: |
+    bge-raranker-large 是一个通用reranking,由北京智源人工智能研究院(BAAI)推出,同时支持中英文
 
+    官网: https://www.baai.ac.cn/
+    Github: https://github.com/FlagOpen/FlagEmbedding
+    HuggingFace: https://huggingface.co/BAAI/bge-reranker-large
+    arXiv: https://arxiv.org/pdf/2309.07597
+
+    北京智源人工智能研究院是北京大学的直属研究机构,主要从事人工智能的数理基础、机器学习、智能信息检索与挖掘、智能体系架构与芯片、自然语言处理等领域研究。
+  types: "reranking"
+  huggingFaceRepo: BAAI/bge-reranker-large