ChatGPT/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-10-15 15:30:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: support replicate chat models (#1989)

Browse Source 
* feat: add Replicate adaptor and integrate into channel and API types

* feat: support llm chat on replicate
This commit is contained in:
Laisky.Cai
2024-12-21 14:41:19 +08:00
committed by GitHub
parent 36c8f4f15c
commit 305ce14fe3
25 changed files with 908 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
relay/billing/ratio/model.go
View File

@@ -211,6 +211,50 @@ var ModelRatio = map[string]float64{
"deepl-ja": 25.0 / 1000 * USD,
// https://console.x.ai/
"grok-beta": 5.0 / 1000 * USD,
// replicate charges based on the number of generated images
// https://replicate.com/pricing
"black-forest-labs/flux-1.1-pro": 0.04 * USD,
"black-forest-labs/flux-1.1-pro-ultra": 0.06 * USD,
"black-forest-labs/flux-canny-dev": 0.025 * USD,
"black-forest-labs/flux-canny-pro": 0.05 * USD,
"black-forest-labs/flux-depth-dev": 0.025 * USD,
"black-forest-labs/flux-depth-pro": 0.05 * USD,
"black-forest-labs/flux-dev": 0.025 * USD,
"black-forest-labs/flux-dev-lora": 0.032 * USD,
"black-forest-labs/flux-fill-dev": 0.04 * USD,
"black-forest-labs/flux-fill-pro": 0.05 * USD,
"black-forest-labs/flux-pro": 0.055 * USD,
"black-forest-labs/flux-redux-dev": 0.025 * USD,
"black-forest-labs/flux-redux-schnell": 0.003 * USD,
"black-forest-labs/flux-schnell": 0.003 * USD,
"black-forest-labs/flux-schnell-lora": 0.02 * USD,
"ideogram-ai/ideogram-v2": 0.08 * USD,
"ideogram-ai/ideogram-v2-turbo": 0.05 * USD,
"recraft-ai/recraft-v3": 0.04 * USD,
"recraft-ai/recraft-v3-svg": 0.08 * USD,
"stability-ai/stable-diffusion-3": 0.035 * USD,
"stability-ai/stable-diffusion-3.5-large": 0.065 * USD,
"stability-ai/stable-diffusion-3.5-large-turbo": 0.04 * USD,
"stability-ai/stable-diffusion-3.5-medium": 0.035 * USD,
// replicate chat models
"ibm-granite/granite-20b-code-instruct-8k": 0.100 * USD,
"ibm-granite/granite-3.0-2b-instruct": 0.030 * USD,
"ibm-granite/granite-3.0-8b-instruct": 0.050 * USD,
"ibm-granite/granite-8b-code-instruct-128k": 0.050 * USD,
"meta/llama-2-13b": 0.100 * USD,
"meta/llama-2-13b-chat": 0.100 * USD,
"meta/llama-2-70b": 0.650 * USD,
"meta/llama-2-70b-chat": 0.650 * USD,
"meta/llama-2-7b": 0.050 * USD,
"meta/llama-2-7b-chat": 0.050 * USD,
"meta/meta-llama-3.1-405b-instruct": 9.500 * USD,
"meta/meta-llama-3-70b": 0.650 * USD,
"meta/meta-llama-3-70b-instruct": 0.650 * USD,
"meta/meta-llama-3-8b": 0.050 * USD,
"meta/meta-llama-3-8b-instruct": 0.050 * USD,
"mistralai/mistral-7b-instruct-v0.2": 0.050 * USD,
"mistralai/mistral-7b-v0.1": 0.050 * USD,
"mistralai/mixtral-8x7b-instruct-v0.1": 0.300 * USD,
}
var CompletionRatio = map[string]float64{
@@ -362,6 +406,7 @@ func GetCompletionRatio(name string, channelType int) float64 {
if strings.HasPrefix(name, "deepseek-") {
return 2
}
switch name {
case "llama2-70b-4096":
return 0.8 / 0.64
@@ -377,6 +422,35 @@ func GetCompletionRatio(name string, channelType int) float64 {
return 5
case "grok-beta":
return 3
// Replicate Models
// https://replicate.com/pricing
case "ibm-granite/granite-20b-code-instruct-8k":
return 5
case "ibm-granite/granite-3.0-2b-instruct":
return 8.333333333333334
case "ibm-granite/granite-3.0-8b-instruct",
"ibm-granite/granite-8b-code-instruct-128k":
return 5
case "meta/llama-2-13b",
"meta/llama-2-13b-chat",
"meta/llama-2-7b",
"meta/llama-2-7b-chat",
"meta/meta-llama-3-8b",
"meta/meta-llama-3-8b-instruct":
return 5
case "meta/llama-2-70b",
"meta/llama-2-70b-chat",
"meta/meta-llama-3-70b",
"meta/meta-llama-3-70b-instruct":
return 2.750 / 0.650 // ≈4.230769
case "meta/meta-llama-3.1-405b-instruct":
return 1
case "mistralai/mistral-7b-instruct-v0.2",
"mistralai/mistral-7b-v0.1":
return 5
case "mistralai/mixtral-8x7b-instruct-v0.1":
return 1.000 / 0.300 // ≈3.333333
}
return 1
}