xinyin025/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-11-12 03:13:41 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

feat: support gpt-4o-audio

Browse Source
This commit is contained in:
Laisky.Cai
2025-01-14 06:38:07 +00:00
parent c6c8053ccc
commit 2fc6caaae5
14 changed files with 401 additions and 198 deletions
Download Patch File Download Diff File Expand all files Collapse all files

235
relay/billing/ratio/model.go
View File

@@ -3,6 +3,7 @@ package ratio
import (
"encoding/json"
"fmt"
"math"
"strings"
"github.com/songquanpeng/one-api/common/logger"
@@ -22,65 +23,71 @@ const (
// 1 === ￥0.014 / 1k tokens
var ModelRatio = map[string]float64{
// https://openai.com/pricing
"gpt-4": 15,
"gpt-4-0314": 15,
"gpt-4-0613": 15,
"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
"chatgpt-4o-latest": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-05-13": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-08-06": 1.25, // $0.0025 / 1K tokens
"gpt-4o-2024-11-20": 1.25, // $0.0025 / 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
"gpt-3.5-turbo-16k-0613": 1.5,
"gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
"gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
"gpt-3.5-turbo-0125": 0.25, // $0.0005 / 1K tokens
"o1": 7.5, // $15.00 / 1M input tokens
"o1-2024-12-17": 7.5,
"o1-preview": 7.5, // $15.00 / 1M input tokens
"o1-preview-2024-09-12": 7.5,
"o1-mini": 1.5, // $3.00 / 1M input tokens
"o1-mini-2024-09-12": 1.5,
"davinci-002": 1, // $0.002 / 1K tokens
"babbage-002": 0.2, // $0.0004 / 1K tokens
"text-ada-001": 0.2,
"text-babbage-001": 0.25,
"text-curie-001": 1,
"text-davinci-002": 10,
"text-davinci-003": 10,
"text-davinci-edit-001": 10,
"code-davinci-edit-001": 10,
"whisper-1": 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
"tts-1": 7.5, // $0.015 / 1K characters
"tts-1-1106": 7.5,
"tts-1-hd": 15, // $0.030 / 1K characters
"tts-1-hd-1106": 15,
"davinci": 10,
"curie": 10,
"babbage": 10,
"ada": 10,
"text-embedding-ada-002": 0.05,
"text-embedding-3-small": 0.01,
"text-embedding-3-large": 0.065,
"text-search-ada-doc-001": 10,
"text-moderation-stable": 0.1,
"text-moderation-latest": 0.1,
"dall-e-2": 0.02 * USD, // $0.016 - $0.020 / image
"dall-e-3": 0.04 * USD, // $0.040 - $0.120 / image
"gpt-4": 15,
"gpt-4-0314": 15,
"gpt-4-0613": 15,
"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
"chatgpt-4o-latest": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-05-13": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-08-06": 1.25, // $0.0025 / 1K tokens
"gpt-4o-2024-11-20": 1.25, // $0.0025 / 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
// Audio billing will mix text and audio tokens, the unit price is different.
// Here records the cost of text, the cost multiplier of audio
// relative to text is in AudioRatio
"gpt-4o-audio-preview": 1.25, // $0.0025 / 1K tokens
"gpt-4o-audio-preview-2024-12-17": 1.25, // $0.0025 / 1K tokens
"gpt-4o-audio-preview-2024-10-01": 1.25, // $0.0025 / 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
"gpt-3.5-turbo-16k-0613": 1.5,
"gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
"gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
"gpt-3.5-turbo-0125": 0.25, // $0.0005 / 1K tokens
"o1": 7.5, // $15.00 / 1M input tokens
"o1-2024-12-17": 7.5,
"o1-preview": 7.5, // $15.00 / 1M input tokens
"o1-preview-2024-09-12": 7.5,
"o1-mini": 1.5, // $3.00 / 1M input tokens
"o1-mini-2024-09-12": 1.5,
"davinci-002": 1, // $0.002 / 1K tokens
"babbage-002": 0.2, // $0.0004 / 1K tokens
"text-ada-001": 0.2,
"text-babbage-001": 0.25,
"text-curie-001": 1,
"text-davinci-002": 10,
"text-davinci-003": 10,
"text-davinci-edit-001": 10,
"code-davinci-edit-001": 10,
"whisper-1": 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
"tts-1": 7.5, // $0.015 / 1K characters
"tts-1-1106": 7.5,
"tts-1-hd": 15, // $0.030 / 1K characters
"tts-1-hd-1106": 15,
"davinci": 10,
"curie": 10,
"babbage": 10,
"ada": 10,
"text-embedding-ada-002": 0.05,
"text-embedding-3-small": 0.01,
"text-embedding-3-large": 0.065,
"text-search-ada-doc-001": 10,
"text-moderation-stable": 0.1,
"text-moderation-latest": 0.1,
"dall-e-2": 0.02 * USD, // $0.016 - $0.020 / image
"dall-e-3": 0.04 * USD, // $0.040 - $0.120 / image
// https://www.anthropic.com/api#pricing
"claude-instant-1.2": 0.8 / 1000 * USD,
"claude-2.0": 8.0 / 1000 * USD,
@@ -254,7 +261,6 @@ var ModelRatio = map[string]float64{
"llama3-groq-70b-8192-tool-use-preview": 0.89 / 1000000 * USD,
"llama3-groq-8b-8192-tool-use-preview": 0.19 / 1000000 * USD,
"mixtral-8x7b-32768": 0.24 / 1000000 * USD,
// https://platform.lingyiwanwu.com/docs#-计费单元
"yi-34b-chat-0205": 2.5 / 1000 * RMB,
"yi-34b-chat-200k": 12.0 / 1000 * RMB,
@@ -333,6 +339,68 @@ var ModelRatio = map[string]float64{
"mistralai/mixtral-8x7b-instruct-v0.1": 0.300 * USD,
}
// AudioRatio represents the price ratio between audio tokens and text tokens
var AudioRatio = map[string]float64{
"gpt-4o-audio-preview": 16,
"gpt-4o-audio-preview-2024-12-17": 16,
"gpt-4o-audio-preview-2024-10-01": 40,
}
// GetAudioPromptRatio returns the audio prompt ratio for the given model.
func GetAudioPromptRatio(actualModelName string) float64 {
var v float64
if ratio, ok := AudioRatio[actualModelName]; ok {
v = ratio
} else {
v = 16
}
return v
}
// AudioCompletionRatio is the completion ratio for audio models.
var AudioCompletionRatio = map[string]float64{
"whisper-1": 0,
"gpt-4o-audio-preview": 2,
"gpt-4o-audio-preview-2024-12-17": 2,
"gpt-4o-audio-preview-2024-10-01": 2,
}
// GetAudioCompletionRatio returns the completion ratio for audio models.
func GetAudioCompletionRatio(actualModelName string) float64 {
var v float64
if ratio, ok := AudioCompletionRatio[actualModelName]; ok {
v = ratio
} else {
v = 2
}
return v
}
// AudioTokensPerSecond is the number of audio tokens per second for each model.
var AudioPromptTokensPerSecond = map[string]float64{
// $0.006 / minute -> $0.002 / 20 seconds -> $0.002 / 1K tokens
"whisper-1": 1000 / 20,
// gpt-4o-audio series processes 10 tokens per second
"gpt-4o-audio-preview": 10,
"gpt-4o-audio-preview-2024-12-17": 10,
"gpt-4o-audio-preview-2024-10-01": 10,
}
// GetAudioPromptTokensPerSecond returns the number of audio tokens per second
// for the given model.
func GetAudioPromptTokensPerSecond(actualModelName string) int {
var v float64
if tokensPerSecond, ok := AudioPromptTokensPerSecond[actualModelName]; ok {
v = tokensPerSecond
} else {
v = 10
}
return int(math.Ceil(v))
}
var CompletionRatio = map[string]float64{
// aws llama3
"llama3-8b-8192(33)": 0.0006 / 0.0003,
@@ -397,19 +465,21 @@ func GetModelRatio(name string, channelType int) float64 {
if strings.HasPrefix(name, "command-") && strings.HasSuffix(name, "-internet") {
name = strings.TrimSuffix(name, "-internet")
}
model := fmt.Sprintf("%s(%d)", name, channelType)
if ratio, ok := ModelRatio[model]; ok {
return ratio
}
if ratio, ok := DefaultModelRatio[model]; ok {
return ratio
}
if ratio, ok := ModelRatio[name]; ok {
return ratio
}
if ratio, ok := DefaultModelRatio[name]; ok {
return ratio
for _, targetName := range []string{model, name} {
for _, ratioMap := range []map[string]float64{
ModelRatio,
DefaultModelRatio,
AudioRatio,
} {
if ratio, ok := ratioMap[targetName]; ok {
return ratio
}
}
}
logger.SysError("model ratio not found: " + name)
return 30
}
@@ -432,18 +502,19 @@ func GetCompletionRatio(name string, channelType int) float64 {
name = strings.TrimSuffix(name, "-internet")
}
model := fmt.Sprintf("%s(%d)", name, channelType)
if ratio, ok := CompletionRatio[model]; ok {
return ratio
}
if ratio, ok := DefaultCompletionRatio[model]; ok {
return ratio
}
if ratio, ok := CompletionRatio[name]; ok {
return ratio
}
if ratio, ok := DefaultCompletionRatio[name]; ok {
return ratio
for _, targetName := range []string{model, name} {
for _, ratioMap := range []map[string]float64{
CompletionRatio,
DefaultCompletionRatio,
AudioCompletionRatio,
} {
if ratio, ok := ratioMap[targetName]; ok {
return ratio
}
}
}
if strings.HasPrefix(name, "gpt-3.5") {
if name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
// https://openai.com/blog/new-embedding-models-and-api-updates