feat: support gpt-4o mini (#1665)

* feat: support gpt-4o mini * feat: fix gpt-4o mini image price
songquanpeng · Jul 22, 2024 · 296ab01 · 296ab01
1 parent 5f03c85
commit 296ab01
Show file tree

Hide file tree

Showing 3 changed files with 27 additions and 11 deletions.
diff --git a/relay/adaptor/openai/constants.go b/relay/adaptor/openai/constants.go
@@ -8,6 +8,7 @@ var ModelList = []string{
 	"gpt-4-32k", "gpt-4-32k-0314", "gpt-4-32k-0613",
 	"gpt-4-turbo-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09",
 	"gpt-4o", "gpt-4o-2024-05-13",
+	"gpt-4o-mini", "gpt-4o-mini-2024-07-18",
 	"gpt-4-vision-preview",
 	"text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large",
 	"text-curie-001", "text-babbage-001", "text-ada-001", "text-davinci-002", "text-davinci-003",

diff --git a/relay/adaptor/openai/token.go b/relay/adaptor/openai/token.go
@@ -110,7 +110,7 @@ func CountTokenMessages(messages []model.Message, model string) int {
 						if imageUrl["detail"] != nil {
 							detail = imageUrl["detail"].(string)
 						}
-						imageTokens, err := countImageTokens(url, detail)
+						imageTokens, err := countImageTokens(url, detail, model)
 						if err != nil {
 							logger.SysError("error counting image tokens: " + err.Error())
 						} else {
@@ -134,11 +134,15 @@ const (
 	lowDetailCost         = 85
 	highDetailCostPerTile = 170
 	additionalCost        = 85
+	// gpt-4o-mini cost higher than other model
+	gpt4oMiniLowDetailCost  = 2833
+	gpt4oMiniHighDetailCost = 5667
+	gpt4oMiniAdditionalCost = 2833
 )
 
 // https://platform.openai.com/docs/guides/vision/calculating-costs
 // https://github.com/openai/openai-cookbook/blob/05e3f9be4c7a2ae7ecf029a7c32065b024730ebe/examples/How_to_count_tokens_with_tiktoken.ipynb
-func countImageTokens(url string, detail string) (_ int, err error) {
+func countImageTokens(url string, detail string, model string) (_ int, err error) {
 	var fetchSize = true
 	var width, height int
 	// Reference: https://platform.openai.com/docs/guides/vision/low-or-high-fidelity-image-understanding
@@ -172,6 +176,9 @@ func countImageTokens(url string, detail string) (_ int, err error) {
 	}
 	switch detail {
 	case "low":
+		if strings.HasPrefix(model, "gpt-4o-mini") {
+			return gpt4oMiniLowDetailCost, nil
+		}
 		return lowDetailCost, nil
 	case "high":
 		if fetchSize {
@@ -191,6 +198,9 @@ func countImageTokens(url string, detail string) (_ int, err error) {
 			height = int(float64(height) * ratio)
 		}
 		numSquares := int(math.Ceil(float64(width)/512) * math.Ceil(float64(height)/512))
+		if strings.HasPrefix(model, "gpt-4o-mini") {
+			return numSquares*gpt4oMiniHighDetailCost + gpt4oMiniAdditionalCost, nil
+		}
 		result := numSquares*highDetailCostPerTile + additionalCost
 		return result, nil
 	default:

diff --git a/relay/billing/ratio/model.go b/relay/billing/ratio/model.go
@@ -28,15 +28,17 @@ var ModelRatio = map[string]float64{
 	"gpt-4-32k":               30,
 	"gpt-4-32k-0314":          30,
 	"gpt-4-32k-0613":          30,
-	"gpt-4-1106-preview":      5,    // $0.01 / 1K tokens
-	"gpt-4-0125-preview":      5,    // $0.01 / 1K tokens
-	"gpt-4-turbo-preview":     5,    // $0.01 / 1K tokens
-	"gpt-4-turbo":             5,    // $0.01 / 1K tokens
-	"gpt-4-turbo-2024-04-09":  5,    // $0.01 / 1K tokens
-	"gpt-4o":                  2.5,  // $0.005 / 1K tokens
-	"gpt-4o-2024-05-13":       2.5,  // $0.005 / 1K tokens
-	"gpt-4-vision-preview":    5,    // $0.01 / 1K tokens
-	"gpt-3.5-turbo":           0.25, // $0.0005 / 1K tokens
+	"gpt-4-1106-preview":      5,     // $0.01 / 1K tokens
+	"gpt-4-0125-preview":      5,     // $0.01 / 1K tokens
+	"gpt-4-turbo-preview":     5,     // $0.01 / 1K tokens
+	"gpt-4-turbo":             5,     // $0.01 / 1K tokens
+	"gpt-4-turbo-2024-04-09":  5,     // $0.01 / 1K tokens
+	"gpt-4o":                  2.5,   // $0.005 / 1K tokens
+	"gpt-4o-2024-05-13":       2.5,   // $0.005 / 1K tokens
+	"gpt-4o-mini":             0.075, // $0.00015 / 1K tokens
+	"gpt-4o-mini-2024-07-18":  0.075, // $0.00015 / 1K tokens
+	"gpt-4-vision-preview":    5,     // $0.01 / 1K tokens
+	"gpt-3.5-turbo":           0.25,  // $0.0005 / 1K tokens
 	"gpt-3.5-turbo-0301":      0.75,
 	"gpt-3.5-turbo-0613":      0.75,
 	"gpt-3.5-turbo-16k":       1.5, // $0.003 / 1K tokens
@@ -308,6 +310,9 @@ func GetCompletionRatio(name string, channelType int) float64 {
 		return 4.0 / 3.0
 	}
 	if strings.HasPrefix(name, "gpt-4") {
+		if strings.HasPrefix(name, "gpt-4o-mini") {
+			return 4
+		}
 		if strings.HasPrefix(name, "gpt-4-turbo") ||
 			strings.HasPrefix(name, "gpt-4o") ||
 			strings.HasSuffix(name, "preview") {