Ecosystem/ai-gateway
5
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: support Mistral's models now (close #1051)

Browse Source
This commit is contained in:
JustSong 2024-03-03 21:46:45 +08:00
parent b35f3523d3
commit 9d8967f7d3
9 changed files with 61 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
README.md
View File

@ -67,6 +67,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
+ [x] [OpenAI ChatGPT 系列模型](https://platform.openai.com/docs/guides/gpt/chat-completions-api)（支持 [Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference)）
+ [x] [Anthropic Claude 系列模型](https://anthropic.com)
+ [x] [Google PaLM2/Gemini 系列模型](https://developers.generativeai.google)
+ [x] [Mistral 系列模型](https://mistral.ai/)
+ [x] [百度文心一言系列模型](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
+ [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
+ [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)

2
common/constants.go
View File

@ -66,6 +66,7 @@ const (
ChannelTypeMoonshot = 25
ChannelTypeBaichuan = 26
ChannelTypeMinimax = 27
ChannelTypeMistral = 28
)
var ChannelBaseURLs = []string{
@ -97,6 +98,7 @@ var ChannelBaseURLs = []string{
"https://api.moonshot.cn", // 25
"https://api.baichuan-ai.com", // 26
"https://api.minimax.chat", // 27
"https://api.mistral.ai", // 28
}
const (

23
common/model-ratio.go
View File

@ -17,7 +17,6 @@ const (
// https://platform.openai.com/docs/models/model-endpoint-compatibility
// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
// https://openai.com/pricing
// TODO: when a new api is enabled, check the pricing here
// 1 === $0.002 / 1K tokens
// 1 === ￥0.014 / 1k tokens
var ModelRatio = map[string]float64{
@ -116,15 +115,29 @@ var ModelRatio = map[string]float64{
"abab6-chat": 0.1 * RMB,
"abab5.5-chat": 0.015 * RMB,
"abab5.5s-chat": 0.005 * RMB,
// https://docs.mistral.ai/platform/pricing/
"open-mistral-7b": 0.25 / 1000 * USD,
"open-mixtral-8x7b": 0.7 / 1000 * USD,
"mistral-small-latest": 2.0 / 1000 * USD,
"mistral-medium-latest": 2.7 / 1000 * USD,
"mistral-large-latest": 8.0 / 1000 * USD,
"mistral-embed": 0.1 / 1000 * USD,
}
var CompletionRatio = map[string]float64{}
var DefaultModelRatio map[string]float64
var DefaultCompletionRatio map[string]float64
func init() {
DefaultModelRatio = make(map[string]float64)
for k, v := range ModelRatio {
DefaultModelRatio[k] = v
}
DefaultCompletionRatio = make(map[string]float64)
for k, v := range CompletionRatio {
DefaultCompletionRatio[k] = v
}
}
func ModelRatio2JSONString() string {
@ -155,8 +168,6 @@ func GetModelRatio(name string) float64 {
return ratio
}
var CompletionRatio = map[string]float64{}
func CompletionRatio2JSONString() string {
jsonBytes, err := json.Marshal(CompletionRatio)
if err != nil {
@ -174,6 +185,9 @@ func GetCompletionRatio(name string) float64 {
if ratio, ok := CompletionRatio[name]; ok {
return ratio
}
if ratio, ok := DefaultCompletionRatio[name]; ok {
return ratio
}
if strings.HasPrefix(name, "gpt-3.5") {
if strings.HasSuffix(name, "0125") {
// https://openai.com/blog/new-embedding-models-and-api-updates
@ -206,5 +220,8 @@ func GetCompletionRatio(name string) float64 {
if strings.HasPrefix(name, "claude-2") {
return 2.965517
}
if strings.HasPrefix(name, "mistral-") {
return 3
}
return 1
}

12
controller/model.go
View File

@ -6,6 +6,7 @@ import (
"github.com/songquanpeng/one-api/relay/channel/ai360"
"github.com/songquanpeng/one-api/relay/channel/baichuan"
"github.com/songquanpeng/one-api/relay/channel/minimax"
"github.com/songquanpeng/one-api/relay/channel/mistral"
"github.com/songquanpeng/one-api/relay/channel/moonshot"
"github.com/songquanpeng/one-api/relay/constant"
"github.com/songquanpeng/one-api/relay/helper"
@ -122,6 +123,17 @@ func init() {
Parent: nil,
})
}
for _, modelName := range mistral.ModelList {
openAIModels = append(openAIModels, OpenAIModels{
Id: modelName,
Object: "model",
Created: 1626777600,
OwnedBy: "mistralai",
Permission: permission,
Root: modelName,
Parent: nil,
})
}
openAIModelsMap = make(map[string]OpenAIModels)
for _, model := range openAIModels {
openAIModelsMap[model.Id] = model

10
relay/channel/mistral/constants.go Normal file
View File

@ -0,0 +1,10 @@
package mistral
var ModelList = []string{
"open-mistral-7b",
"open-mixtral-8x7b",
"mistral-small-latest",
"mistral-medium-latest",
"mistral-large-latest",
"mistral-embed",
}

5
relay/channel/openai/adaptor.go
View File

@ -9,6 +9,7 @@ import (
"github.com/songquanpeng/one-api/relay/channel/ai360"
"github.com/songquanpeng/one-api/relay/channel/baichuan"
"github.com/songquanpeng/one-api/relay/channel/minimax"
"github.com/songquanpeng/one-api/relay/channel/mistral"
"github.com/songquanpeng/one-api/relay/channel/moonshot"
"github.com/songquanpeng/one-api/relay/model"
"github.com/songquanpeng/one-api/relay/util"
@ -94,6 +95,8 @@ func (a *Adaptor) GetModelList() []string {
return baichuan.ModelList
case common.ChannelTypeMinimax:
return minimax.ModelList
case common.ChannelTypeMistral:
return mistral.ModelList
default:
return ModelList
}
@ -111,6 +114,8 @@ func (a *Adaptor) GetChannelName() string {
return "baichuan"
case common.ChannelTypeMinimax:
return "minimax"
case common.ChannelTypeMistral:
return "mistralai"
default:
return "openai"
}

10
relay/controller/helper.go
View File

@ -172,10 +172,8 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *util.R
if err != nil {
logger.Error(ctx, "error update user quota cache: "+err.Error())
}
if quota != 0 {
logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
model.UpdateChannelUsedQuota(meta.ChannelId, quota)
}
logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
model.UpdateChannelUsedQuota(meta.ChannelId, quota)
}

6
web/berry/src/constants/ChannelConstants.js
View File

@ -29,6 +29,12 @@ export const CHANNEL_OPTIONS = {
value: 24,
color: 'orange'
},
28: {
key: 28,
text: 'Mistral AI',
value: 28,
color: 'orange'
},
15: {
key: 15,
text: '百度文心千帆',

1
web/default/src/constants/channel.constants.js
View File

@ -4,6 +4,7 @@ export const CHANNEL_OPTIONS = [
{ key: 3, text: 'Azure OpenAI', value: 3, color: 'olive' },
{ key: 11, text: 'Google PaLM2', value: 11, color: 'orange' },
{ key: 24, text: 'Google Gemini', value: 24, color: 'orange' },
{ key: 28, text: 'Mistral AI', value: 28, color: 'orange' },
{ key: 15, text: '百度文心千帆', value: 15, color: 'blue' },
{ key: 17, text: '阿里通义千问', value: 17, color: 'orange' },
{ key: 18, text: '讯飞星火认知', value: 18, color: 'blue' },