feat: support gemini-2.0-flash (#2055)

* feat: support gemini-2.0-flash - Enhance model support by adding new entries and refining checks for system instruction compatibility. - Update logging display behavior and adjust default quotas for better user experience. - Revamp pricing structures in the billing system to reflect current model values and deprecate outdated entries. - Streamline code by replacing hardcoded values with configurations for maintainability. * feat: add new Gemini 2.0 flash models to adapter and billing ratio * fix: update GetRequestURL to support gemini-1.5 model in versioning
2025-11-04 15:53:42 +08:00 · 2025-02-12 00:34:25 +08:00
parent 6916debf66
commit fef7ae048b
7 changed files with 76 additions and 31 deletions
--- a/relay/adaptor/gemini/adaptor.go
+++ b/relay/adaptor/gemini/adaptor.go
@@ -5,9 +5,10 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"strings"

 	"github.com/gin-gonic/gin"
-
+	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
 	channelhelper "github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
@@ -20,17 +21,12 @@ type Adaptor struct {
 }

 func (a *Adaptor) Init(meta *meta.Meta) {
-
 }

 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
-	var defaultVersion string
-	switch meta.ActualModelName {
-	case "gemini-2.0-flash-exp",
-		"gemini-2.0-flash-thinking-exp",
-		"gemini-2.0-flash-thinking-exp-01-21":
-		defaultVersion = "v1beta"
-	default:
+	defaultVersion := config.GeminiVersion
+	if strings.Contains(meta.ActualModelName, "gemini-2.0") ||
+		strings.Contains(meta.ActualModelName, "gemini-1.5") {
 		defaultVersion = "v1beta"
 	}

--- a/relay/adaptor/gemini/constants.go
+++ b/relay/adaptor/gemini/constants.go
@@ -4,8 +4,38 @@ package gemini

 var ModelList = []string{
 	"gemini-pro", "gemini-1.0-pro",
-	"gemini-1.5-flash", "gemini-1.5-pro",
+	// "gemma-2-2b-it", "gemma-2-9b-it", "gemma-2-27b-it",
+	"gemini-1.5-flash", "gemini-1.5-flash-8b",
+	"gemini-1.5-pro", "gemini-1.5-pro-experimental",
 	"text-embedding-004", "aqa",
-	"gemini-2.0-flash-exp",
-	"gemini-2.0-flash-thinking-exp", "gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-flash", "gemini-2.0-flash-exp",
+	"gemini-2.0-flash-lite-preview-02-05",
+	"gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-pro-exp-02-05",
+}
+
+// ModelsSupportSystemInstruction is the list of models that support system instruction.
+//
+// https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/system-instructions
+var ModelsSupportSystemInstruction = []string{
+	// "gemini-1.0-pro-002",
+	// "gemini-1.5-flash", "gemini-1.5-flash-001", "gemini-1.5-flash-002",
+	// "gemini-1.5-flash-8b",
+	// "gemini-1.5-pro", "gemini-1.5-pro-001", "gemini-1.5-pro-002",
+	// "gemini-1.5-pro-experimental",
+	"gemini-2.0-flash", "gemini-2.0-flash-exp",
+	"gemini-2.0-flash-thinking-exp-01-21",
+}
+
+// IsModelSupportSystemInstruction check if the model support system instruction.
+//
+// Because the main version of Go is 1.20, slice.Contains cannot be used
+func IsModelSupportSystemInstruction(model string) bool {
+	for _, m := range ModelsSupportSystemInstruction {
+		if m == model {
+			return true
+		}
+	}
+
+	return false
 }
--- a/relay/adaptor/gemini/main.go
+++ b/relay/adaptor/gemini/main.go
@@ -132,9 +132,16 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 		}
 		// Converting system prompt to prompt from user for the same reason
 		if content.Role == "system" {
-			content.Role = "user"
 			shouldAddDummyModelMessage = true
+			if IsModelSupportSystemInstruction(textRequest.Model) {
+				geminiRequest.SystemInstruction = &content
+				geminiRequest.SystemInstruction.Role = ""
+				continue
+			} else {
+				content.Role = "user"
+			}
 		}
+
 		geminiRequest.Contents = append(geminiRequest.Contents, content)

 		// If a system message is the last message, we need to add a dummy model message to make gemini happy
--- a/relay/adaptor/gemini/model.go
+++ b/relay/adaptor/gemini/model.go
@@ -1,10 +1,11 @@
 package gemini

 type ChatRequest struct {
-	Contents         []ChatContent        `json:"contents"`
-	SafetySettings   []ChatSafetySettings `json:"safety_settings,omitempty"`
-	GenerationConfig ChatGenerationConfig `json:"generation_config,omitempty"`
-	Tools            []ChatTools          `json:"tools,omitempty"`
+	Contents          []ChatContent        `json:"contents"`
+	SafetySettings    []ChatSafetySettings `json:"safety_settings,omitempty"`
+	GenerationConfig  ChatGenerationConfig `json:"generation_config,omitempty"`
+	Tools             []ChatTools          `json:"tools,omitempty"`
+	SystemInstruction *ChatContent         `json:"system_instruction,omitempty"`
 }

 type EmbeddingRequest struct {
--- a/relay/adaptor/vertexai/gemini/adapter.go
+++ b/relay/adaptor/vertexai/gemini/adapter.go
@@ -16,10 +16,12 @@ import (

 var ModelList = []string{
 	"gemini-pro", "gemini-pro-vision",
-	"gemini-1.5-pro-001", "gemini-1.5-flash-001",
-	"gemini-1.5-pro-002", "gemini-1.5-flash-002",
-	"gemini-2.0-flash-exp",
-	"gemini-2.0-flash-thinking-exp", "gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-exp-1206",
+	"gemini-1.5-pro-001", "gemini-1.5-pro-002",
+	"gemini-1.5-flash-001", "gemini-1.5-flash-002",
+	"gemini-2.0-flash-exp", "gemini-2.0-flash-001",
+	"gemini-2.0-flash-lite-preview-02-05",
+	"gemini-2.0-flash-thinking-exp-01-21",
 }

 type Adaptor struct {
--- a/relay/billing/ratio/model.go
+++ b/relay/billing/ratio/model.go
@@ -115,15 +115,24 @@ var ModelRatio = map[string]float64{
 	"bge-large-en":       0.002 * RMB,
 	"tao-8k":             0.002 * RMB,
 	// https://ai.google.dev/pricing
-	"gemini-pro":                          1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
-	"gemini-1.0-pro":                      1,
-	"gemini-1.5-pro":                      1,
-	"gemini-1.5-pro-001":                  1,
-	"gemini-1.5-flash":                    1,
-	"gemini-1.5-flash-001":                1,
-	"gemini-2.0-flash-exp":                1,
-	"gemini-2.0-flash-thinking-exp":       1,
-	"gemini-2.0-flash-thinking-exp-01-21": 1,
+	// https://cloud.google.com/vertex-ai/generative-ai/pricing
+	// "gemma-2-2b-it":                       0,
+	// "gemma-2-9b-it":                       0,
+	// "gemma-2-27b-it":                      0,
+	"gemini-pro":                          0.25 * MILLI_USD, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	"gemini-1.0-pro":                      0.125 * MILLI_USD,
+	"gemini-1.5-pro":                      1.25 * MILLI_USD,
+	"gemini-1.5-pro-001":                  1.25 * MILLI_USD,
+	"gemini-1.5-pro-experimental":         1.25 * MILLI_USD,
+	"gemini-1.5-flash":                    0.075 * MILLI_USD,
+	"gemini-1.5-flash-001":                0.075 * MILLI_USD,
+	"gemini-1.5-flash-8b":                 0.0375 * MILLI_USD,
+	"gemini-2.0-flash-exp":                0.075 * MILLI_USD,
+	"gemini-2.0-flash":                    0.15 * MILLI_USD,
+	"gemini-2.0-flash-001":                0.15 * MILLI_USD,
+	"gemini-2.0-flash-lite-preview-02-05": 0.075 * MILLI_USD,
+	"gemini-2.0-flash-thinking-exp-01-21": 0.075 * MILLI_USD,
+	"gemini-2.0-pro-exp-02-05":            1.25 * MILLI_USD,
 	"aqa":                                 1,
 	// https://open.bigmodel.cn/pricing
 	"glm-zero-preview": 0.01 * RMB,
--- a/web/default/src/helpers/utils.js
+++ b/web/default/src/helpers/utils.js
@@ -224,4 +224,4 @@ export function getChannelModels(type) {
    return channelModels[type];
  }
  return [];
-}
+}