Merge pull request #117 from Abirdcfly/more

feat: add baichuan-7b chatglm-6b in dashscope as llm

examples/dashscope/main.go

@@ -53,9 +53,9 @@ func main() {
 			panic(err)
 		}
 	}
-	for _, model := range []dashscope.Model{dashscope.LLAMA27BCHATV2, dashscope.LLAMA213BCHATV2} {
+	for _, model := range []dashscope.Model{dashscope.LLAMA27BCHATV2, dashscope.LLAMA213BCHATV2, dashscope.BAICHUAN7BV1, dashscope.CHATGLM6BV2} {
 		klog.V(0).Infof("\nChat with %s\n", model)
-		resp, err := sampleChatWithLlama2(apiKey, model)
+		resp, err := sampleChatWithOthers(apiKey, model)
 		if err != nil {
 			panic(err)
 		}
@@ -107,9 +107,12 @@ func sampleChat(apiKey string, model dashscope.Model) (llms.Response, error) {
 	return client.Call(params.Marshal())
 }
 
-func sampleChatWithLlama2(apiKey string, model dashscope.Model) (llms.Response, error) {
+func sampleChatWithOthers(apiKey string, model dashscope.Model) (llms.Response, error) {
 	client := dashscope.NewDashScope(apiKey, false)
-	params := dashscope.DefaultModelParams()
+	params := dashscope.DefaultModelParamsSimpleChat()
+	if model == dashscope.CHATGLM6BV2 {
+		params.Input.History = &[]string{}
+	}
 	params.Model = model
 	params.Input.Prompt = samplePrompt
 	return client.Call(params.Marshal())

pkg/llms/dashscope/api.go

@@ -40,6 +40,8 @@ const (
 	// LLaMa2 系列大语言模型由 Meta 开发并公开发布，其规模从 70 亿到 700 亿参数不等。在灵积上提供的 llama2-7b-chat-v2 和 llama2-13b-chat-v2，分别为 7B 和 13B 规模的 LLaMa2 模型，针对对话场景微调优化后的版本。
 	LLAMA27BCHATV2   Model = "llama2-7b-chat-v2"
 	LLAMA213BCHATV2  Model = "llama2-13b-chat-v2"
+	BAICHUAN7BV1     Model = "baichuan-7b-v1"          // baichuan-7B 是由百川智能开发的一个开源的大规模预训练模型。基于 Transformer 结构，在大约 1.2 万亿 tokens 上训练的 70 亿参数模型，支持中英双语，上下文窗口长度为 4096。在标准的中文和英文权威 benchmark（C-EVAL/MMLU）上均取得同尺寸最好的效果。
+	CHATGLM6BV2      Model = "chatglm-6b-v2"           // ChatGLM2 模型是由智谱 AI 出品的大规模语言模型，它在灵积平台上的模型名称为 "chatglm-6b-v2".
 	EmbeddingV1      Model = "text-embedding-v1"       // 通用文本向量 同步调用
 	EmbeddingAsyncV1 Model = "text-embedding-async-v1" // 通用文本向量 批处理调用
 )
@@ -68,7 +70,7 @@ func (z *DashScope) Call(data []byte) (llms.Response, error) {
 	if err := params.Unmarshal(data); err != nil {
 		return nil, err
 	}
-	return do(context.TODO(), DashScopeChatURL, z.apiKey, data, z.sse, false)
+	return do(context.TODO(), DashScopeChatURL, z.apiKey, data, z.sse, false, params.Model)
 }
 
 func (z *DashScope) Validate() (llms.Response, error) {

pkg/llms/dashscope/http_client.go

@@ -21,6 +21,8 @@ import (
 	"context"
 	"encoding/json"
 	"net/http"
+
+	"github.com/kubeagi/arcadia/pkg/llms"
 )
 
 func setHeaders(req *http.Request, token string, sse, async bool) {
@@ -39,8 +41,8 @@ func setHeaders(req *http.Request, token string, sse, async bool) {
 	req.Header.Set("Authorization", "Bearer "+token)
 }
 
-func parseHTTPResponse(resp *http.Response) (data *Response, err error) {
-	if err = json.NewDecoder(resp.Body).Decode(&data); err != nil {
+func parseHTTPResponse(resp *http.Response, data llms.Response) (llms.Response, error) {
+	if err := json.NewDecoder(resp.Body).Decode(&data); err != nil {
 		return nil, err
 	}
 	return data, nil
@@ -61,11 +63,17 @@ func req(ctx context.Context, apiURL, token string, data []byte, sse, async bool
 
 	return http.DefaultClient.Do(req)
 }
-func do(ctx context.Context, apiURL, token string, data []byte, sse, async bool) (*Response, error) {
+func do(ctx context.Context, apiURL, token string, data []byte, sse, async bool, model Model) (llms.Response, error) {
 	resp, err := req(ctx, apiURL, token, data, sse, async)
 	if err != nil {
 		return nil, err
 	}
 	defer resp.Body.Close()
-	return parseHTTPResponse(resp)
+	var respData llms.Response
+	if model == CHATGLM6BV2 {
+		respData = &ResponseChatGLB6B{}
+	} else {
+		respData = &Response{}
+	}
+	return parseHTTPResponse(resp, respData)
 }

pkg/llms/dashscope/params.go

@@ -42,14 +42,15 @@ var _ llms.ModelParams = (*ModelParams)(nil)
 type ModelParams struct {
 	Model      Model      `json:"model"`
 	Input      Input      `json:"input"`
-	Parameters Parameters `json:"parameters"`
+	Parameters Parameters `json:"parameters,omitempty"`
 }
 
 // +kubebuilder:object:generate=true
 
 type Input struct {
 	Messages []Message `json:"messages,omitempty"`
 	Prompt   string    `json:"prompt,omitempty"`
+	History  *[]string `json:"history,omitempty"`
 }
 
 type Parameters struct {
@@ -80,6 +81,14 @@ func DefaultModelParams() ModelParams {
 		},
 	}
 }
+func DefaultModelParamsSimpleChat() ModelParams {
+	return ModelParams{
+		Model: QWEN14BChat,
+		Input: Input{
+			Prompt: "",
+		},
+	}
+}
 
 func (params *ModelParams) Marshal() []byte {
 	data, err := json.Marshal(params)

pkg/llms/dashscope/response.go

@@ -17,12 +17,14 @@ limitations under the License.
 package dashscope
 
 import (
+	"bytes"
 	"encoding/json"
 
 	"github.com/kubeagi/arcadia/pkg/llms"
 )
 
 var _ llms.Response = (*Response)(nil)
+var _ llms.Response = (*ResponseChatGLB6B)(nil)
 
 type CommonResponse struct {
 	// https://help.aliyun.com/zh/dashscope/response-status-codes
@@ -40,6 +42,7 @@ type Response struct {
 type Output struct {
 	Choices []Choice `json:"choices,omitempty"`
 	Text    string   `json:"text,omitempty"`
+	History []string `json:"history,omitempty"`
 }
 
 type FinishReason string
@@ -77,5 +80,43 @@ func (response *Response) Bytes() []byte {
 }
 
 func (response *Response) String() string {
-	return string(response.Bytes())
+	if response.Output.Text != "" {
+		return response.Output.Text
+	}
+	buf := &bytes.Buffer{}
+	for _, c := range response.Output.Choices {
+		buf.WriteString(c.Message.Content)
+	}
+	return buf.String()
+}
+
+type ResponseChatGLB6B struct {
+	CommonResponse
+	Output struct {
+		Text struct {
+			Response string `json:"response,omitempty"`
+		} `json:"text,omitempty"`
+		History []string `json:"history,omitempty"`
+	} `json:"output"`
+	Usage Usage `json:"usage"`
+}
+
+func (r *ResponseChatGLB6B) Type() llms.LLMType {
+	return llms.DashScope
+}
+
+func (r *ResponseChatGLB6B) String() string {
+	return r.Output.Text.Response
+}
+
+func (r *ResponseChatGLB6B) Bytes() []byte {
+	bytes, err := json.Marshal(r)
+	if err != nil {
+		return []byte{}
+	}
+	return bytes
+}
+
+func (r *ResponseChatGLB6B) Unmarshal(bytes []byte) error {
+	return json.Unmarshal(bytes, r)
 }