model_ratio.go 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611
  1. package ratio_setting
  2. import (
  3. "encoding/json"
  4. "one-api/common"
  5. "one-api/setting/operation_setting"
  6. "strings"
  7. "sync"
  8. )
  9. // from songquanpeng/one-api
  10. const (
  11. USD2RMB = 7.3 // 暂定 1 USD = 7.3 RMB
  12. USD = 500 // $0.002 = 1 -> $1 = 500
  13. RMB = USD / USD2RMB
  14. )
  15. // modelRatio
  16. // https://platform.openai.com/docs/models/model-endpoint-compatibility
  17. // https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
  18. // https://openai.com/pricing
  19. // TODO: when a new api is enabled, check the pricing here
  20. // 1 === $0.002 / 1K tokens
  21. // 1 === ¥0.014 / 1k tokens
  22. var defaultModelRatio = map[string]float64{
  23. //"midjourney": 50,
  24. "gpt-4-gizmo-*": 15,
  25. "gpt-4o-gizmo-*": 2.5,
  26. "gpt-4-all": 15,
  27. "gpt-4o-all": 15,
  28. "gpt-4": 15,
  29. //"gpt-4-0314": 15, //deprecated
  30. "gpt-4-0613": 15,
  31. "gpt-4-32k": 30,
  32. //"gpt-4-32k-0314": 30, //deprecated
  33. "gpt-4-32k-0613": 30,
  34. "gpt-4-1106-preview": 5, // $10 / 1M tokens
  35. "gpt-4-0125-preview": 5, // $10 / 1M tokens
  36. "gpt-4-turbo-preview": 5, // $10 / 1M tokens
  37. "gpt-4-vision-preview": 5, // $10 / 1M tokens
  38. "gpt-4-1106-vision-preview": 5, // $10 / 1M tokens
  39. "chatgpt-4o-latest": 2.5, // $5 / 1M tokens
  40. "gpt-4o": 1.25, // $2.5 / 1M tokens
  41. "gpt-4o-audio-preview": 1.25, // $2.5 / 1M tokens
  42. "gpt-4o-audio-preview-2024-10-01": 1.25, // $2.5 / 1M tokens
  43. "gpt-4o-2024-05-13": 2.5, // $5 / 1M tokens
  44. "gpt-4o-2024-08-06": 1.25, // $2.5 / 1M tokens
  45. "gpt-4o-2024-11-20": 1.25, // $2.5 / 1M tokens
  46. "gpt-4o-realtime-preview": 2.5,
  47. "gpt-4o-realtime-preview-2024-10-01": 2.5,
  48. "gpt-4o-realtime-preview-2024-12-17": 2.5,
  49. "gpt-4o-mini-realtime-preview": 0.3,
  50. "gpt-4o-mini-realtime-preview-2024-12-17": 0.3,
  51. "gpt-image-1": 2.5,
  52. "o1": 7.5,
  53. "o1-2024-12-17": 7.5,
  54. "o1-preview": 7.5,
  55. "o1-preview-2024-09-12": 7.5,
  56. "o1-mini": 0.55,
  57. "o1-mini-2024-09-12": 0.55,
  58. "o3-mini": 0.55,
  59. "o3-mini-2025-01-31": 0.55,
  60. "o3-mini-high": 0.55,
  61. "o3-mini-2025-01-31-high": 0.55,
  62. "o3-mini-low": 0.55,
  63. "o3-mini-2025-01-31-low": 0.55,
  64. "o3-mini-medium": 0.55,
  65. "o3-mini-2025-01-31-medium": 0.55,
  66. "gpt-4o-mini": 0.075,
  67. "gpt-4o-mini-2024-07-18": 0.075,
  68. "gpt-4-turbo": 5, // $0.01 / 1K tokens
  69. "gpt-4-turbo-2024-04-09": 5, // $0.01 / 1K tokens
  70. "gpt-4.5-preview": 37.5,
  71. "gpt-4.5-preview-2025-02-27": 37.5,
  72. //"gpt-3.5-turbo-0301": 0.75, //deprecated
  73. "gpt-3.5-turbo": 0.25,
  74. "gpt-3.5-turbo-0613": 0.75,
  75. "gpt-3.5-turbo-16k": 1.5, // $0.003 / 1K tokens
  76. "gpt-3.5-turbo-16k-0613": 1.5,
  77. "gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
  78. "gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
  79. "gpt-3.5-turbo-0125": 0.25,
  80. "babbage-002": 0.2, // $0.0004 / 1K tokens
  81. "davinci-002": 1, // $0.002 / 1K tokens
  82. "text-ada-001": 0.2,
  83. "text-babbage-001": 0.25,
  84. "text-curie-001": 1,
  85. //"text-davinci-002": 10,
  86. //"text-davinci-003": 10,
  87. "text-davinci-edit-001": 10,
  88. "code-davinci-edit-001": 10,
  89. "whisper-1": 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
  90. "tts-1": 7.5, // 1k characters -> $0.015
  91. "tts-1-1106": 7.5, // 1k characters -> $0.015
  92. "tts-1-hd": 15, // 1k characters -> $0.03
  93. "tts-1-hd-1106": 15, // 1k characters -> $0.03
  94. "davinci": 10,
  95. "curie": 10,
  96. "babbage": 10,
  97. "ada": 10,
  98. "text-embedding-3-small": 0.01,
  99. "text-embedding-3-large": 0.065,
  100. "text-embedding-ada-002": 0.05,
  101. "text-search-ada-doc-001": 10,
  102. "text-moderation-stable": 0.1,
  103. "text-moderation-latest": 0.1,
  104. "claude-instant-1": 0.4, // $0.8 / 1M tokens
  105. "claude-2.0": 4, // $8 / 1M tokens
  106. "claude-2.1": 4, // $8 / 1M tokens
  107. "claude-3-haiku-20240307": 0.125, // $0.25 / 1M tokens
  108. "claude-3-5-haiku-20241022": 0.5, // $1 / 1M tokens
  109. "claude-3-sonnet-20240229": 1.5, // $3 / 1M tokens
  110. "claude-3-5-sonnet-20240620": 1.5,
  111. "claude-3-5-sonnet-20241022": 1.5,
  112. "claude-3-7-sonnet-20250219": 1.5,
  113. "claude-3-7-sonnet-20250219-thinking": 1.5,
  114. "claude-sonnet-4-20250514": 1.5,
  115. "claude-3-opus-20240229": 7.5, // $15 / 1M tokens
  116. "claude-opus-4-20250514": 7.5,
  117. "ERNIE-4.0-8K": 0.120 * RMB,
  118. "ERNIE-3.5-8K": 0.012 * RMB,
  119. "ERNIE-3.5-8K-0205": 0.024 * RMB,
  120. "ERNIE-3.5-8K-1222": 0.012 * RMB,
  121. "ERNIE-Bot-8K": 0.024 * RMB,
  122. "ERNIE-3.5-4K-0205": 0.012 * RMB,
  123. "ERNIE-Speed-8K": 0.004 * RMB,
  124. "ERNIE-Speed-128K": 0.004 * RMB,
  125. "ERNIE-Lite-8K-0922": 0.008 * RMB,
  126. "ERNIE-Lite-8K-0308": 0.003 * RMB,
  127. "ERNIE-Tiny-8K": 0.001 * RMB,
  128. "BLOOMZ-7B": 0.004 * RMB,
  129. "Embedding-V1": 0.002 * RMB,
  130. "bge-large-zh": 0.002 * RMB,
  131. "bge-large-en": 0.002 * RMB,
  132. "tao-8k": 0.002 * RMB,
  133. "PaLM-2": 1,
  134. "gemini-1.5-pro-latest": 1.25, // $3.5 / 1M tokens
  135. "gemini-1.5-flash-latest": 0.075,
  136. "gemini-2.0-flash": 0.05,
  137. "gemini-2.5-pro-exp-03-25": 0.625,
  138. "gemini-2.5-pro-preview-03-25": 0.625,
  139. "gemini-2.5-flash-preview-04-17": 0.075,
  140. "gemini-2.5-flash-preview-04-17-thinking": 0.075,
  141. "gemini-2.5-flash-preview-04-17-nothinking": 0.075,
  142. "gemini-2.5-flash-preview-05-20": 0.075,
  143. "gemini-2.5-flash-preview-05-20-thinking": 0.075,
  144. "gemini-2.5-flash-preview-05-20-nothinking": 0.075,
  145. "gemini-2.5-flash-thinking-*": 0.075, // 用于为后续所有2.5 flash thinking budget 模型设置默认倍率
  146. "gemini-2.5-pro-thinking-*": 0.625, // 用于为后续所有2.5 pro thinking budget 模型设置默认倍率
  147. "text-embedding-004": 0.001,
  148. "chatglm_turbo": 0.3572, // ¥0.005 / 1k tokens
  149. "chatglm_pro": 0.7143, // ¥0.01 / 1k tokens
  150. "chatglm_std": 0.3572, // ¥0.005 / 1k tokens
  151. "chatglm_lite": 0.1429, // ¥0.002 / 1k tokens
  152. "glm-4": 7.143, // ¥0.1 / 1k tokens
  153. "glm-4v": 0.05 * RMB, // ¥0.05 / 1k tokens
  154. "glm-4-alltools": 0.1 * RMB, // ¥0.1 / 1k tokens
  155. "glm-3-turbo": 0.3572,
  156. "glm-4-plus": 0.05 * RMB,
  157. "glm-4-0520": 0.1 * RMB,
  158. "glm-4-air": 0.001 * RMB,
  159. "glm-4-airx": 0.01 * RMB,
  160. "glm-4-long": 0.001 * RMB,
  161. "glm-4-flash": 0,
  162. "glm-4v-plus": 0.01 * RMB,
  163. "qwen-turbo": 0.8572, // ¥0.012 / 1k tokens
  164. "qwen-plus": 10, // ¥0.14 / 1k tokens
  165. "text-embedding-v1": 0.05, // ¥0.0007 / 1k tokens
  166. "SparkDesk-v1.1": 1.2858, // ¥0.018 / 1k tokens
  167. "SparkDesk-v2.1": 1.2858, // ¥0.018 / 1k tokens
  168. "SparkDesk-v3.1": 1.2858, // ¥0.018 / 1k tokens
  169. "SparkDesk-v3.5": 1.2858, // ¥0.018 / 1k tokens
  170. "SparkDesk-v4.0": 1.2858,
  171. "360GPT_S2_V9": 0.8572, // ¥0.012 / 1k tokens
  172. "360gpt-turbo": 0.0858, // ¥0.0012 / 1k tokens
  173. "360gpt-turbo-responsibility-8k": 0.8572, // ¥0.012 / 1k tokens
  174. "360gpt-pro": 0.8572, // ¥0.012 / 1k tokens
  175. "360gpt2-pro": 0.8572, // ¥0.012 / 1k tokens
  176. "embedding-bert-512-v1": 0.0715, // ¥0.001 / 1k tokens
  177. "embedding_s1_v1": 0.0715, // ¥0.001 / 1k tokens
  178. "semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
  179. "hunyuan": 7.143, // ¥0.1 / 1k tokens // https://cloud.tencent.com/document/product/1729/97731#e0e6be58-60c8-469f-bdeb-6c264ce3b4d0
  180. // https://platform.lingyiwanwu.com/docs#-计费单元
  181. // 已经按照 7.2 来换算美元价格
  182. "yi-34b-chat-0205": 0.18,
  183. "yi-34b-chat-200k": 0.864,
  184. "yi-vl-plus": 0.432,
  185. "yi-large": 20.0 / 1000 * RMB,
  186. "yi-medium": 2.5 / 1000 * RMB,
  187. "yi-vision": 6.0 / 1000 * RMB,
  188. "yi-medium-200k": 12.0 / 1000 * RMB,
  189. "yi-spark": 1.0 / 1000 * RMB,
  190. "yi-large-rag": 25.0 / 1000 * RMB,
  191. "yi-large-turbo": 12.0 / 1000 * RMB,
  192. "yi-large-preview": 20.0 / 1000 * RMB,
  193. "yi-large-rag-preview": 25.0 / 1000 * RMB,
  194. "command": 0.5,
  195. "command-nightly": 0.5,
  196. "command-light": 0.5,
  197. "command-light-nightly": 0.5,
  198. "command-r": 0.25,
  199. "command-r-plus": 1.5,
  200. "command-r-08-2024": 0.075,
  201. "command-r-plus-08-2024": 1.25,
  202. "deepseek-chat": 0.27 / 2,
  203. "deepseek-coder": 0.27 / 2,
  204. "deepseek-reasoner": 0.55 / 2, // 0.55 / 1k tokens
  205. // Perplexity online 模型对搜索额外收费,有需要应自行调整,此处不计入搜索费用
  206. "llama-3-sonar-small-32k-chat": 0.2 / 1000 * USD,
  207. "llama-3-sonar-small-32k-online": 0.2 / 1000 * USD,
  208. "llama-3-sonar-large-32k-chat": 1 / 1000 * USD,
  209. "llama-3-sonar-large-32k-online": 1 / 1000 * USD,
  210. // grok
  211. "grok-3-beta": 1.5,
  212. "grok-3-mini-beta": 0.15,
  213. "grok-2": 1,
  214. "grok-2-vision": 1,
  215. "grok-beta": 2.5,
  216. "grok-vision-beta": 2.5,
  217. "grok-3-fast-beta": 2.5,
  218. "grok-3-mini-fast-beta": 0.3,
  219. }
  220. var defaultModelPrice = map[string]float64{
  221. "suno_music": 0.1,
  222. "suno_lyrics": 0.01,
  223. "dall-e-3": 0.04,
  224. "imagen-3.0-generate-002": 0.03,
  225. "gpt-4-gizmo-*": 0.1,
  226. "mj_imagine": 0.1,
  227. "mj_variation": 0.1,
  228. "mj_reroll": 0.1,
  229. "mj_blend": 0.1,
  230. "mj_modal": 0.1,
  231. "mj_zoom": 0.1,
  232. "mj_shorten": 0.1,
  233. "mj_high_variation": 0.1,
  234. "mj_low_variation": 0.1,
  235. "mj_pan": 0.1,
  236. "mj_inpaint": 0,
  237. "mj_custom_zoom": 0,
  238. "mj_describe": 0.05,
  239. "mj_upscale": 0.05,
  240. "swap_face": 0.05,
  241. "mj_upload": 0.05,
  242. }
  243. var (
  244. modelPriceMap map[string]float64 = nil
  245. modelPriceMapMutex = sync.RWMutex{}
  246. )
  247. var (
  248. modelRatioMap map[string]float64 = nil
  249. modelRatioMapMutex = sync.RWMutex{}
  250. )
  251. var (
  252. CompletionRatio map[string]float64 = nil
  253. CompletionRatioMutex = sync.RWMutex{}
  254. )
  255. var defaultCompletionRatio = map[string]float64{
  256. "gpt-4-gizmo-*": 2,
  257. "gpt-4o-gizmo-*": 3,
  258. "gpt-4-all": 2,
  259. "gpt-image-1": 8,
  260. }
  261. // InitRatioSettings initializes all model related settings maps
  262. func InitRatioSettings() {
  263. // Initialize modelPriceMap
  264. modelPriceMapMutex.Lock()
  265. modelPriceMap = defaultModelPrice
  266. modelPriceMapMutex.Unlock()
  267. // Initialize modelRatioMap
  268. modelRatioMapMutex.Lock()
  269. modelRatioMap = defaultModelRatio
  270. modelRatioMapMutex.Unlock()
  271. // Initialize CompletionRatio
  272. CompletionRatioMutex.Lock()
  273. CompletionRatio = defaultCompletionRatio
  274. CompletionRatioMutex.Unlock()
  275. // Initialize cacheRatioMap
  276. cacheRatioMapMutex.Lock()
  277. cacheRatioMap = defaultCacheRatio
  278. cacheRatioMapMutex.Unlock()
  279. // initialize imageRatioMap
  280. imageRatioMapMutex.Lock()
  281. imageRatioMap = defaultImageRatio
  282. imageRatioMapMutex.Unlock()
  283. }
  284. func GetModelPriceMap() map[string]float64 {
  285. modelPriceMapMutex.RLock()
  286. defer modelPriceMapMutex.RUnlock()
  287. return modelPriceMap
  288. }
  289. func ModelPrice2JSONString() string {
  290. modelPriceMapMutex.RLock()
  291. defer modelPriceMapMutex.RUnlock()
  292. jsonBytes, err := json.Marshal(modelPriceMap)
  293. if err != nil {
  294. common.SysError("error marshalling model price: " + err.Error())
  295. }
  296. return string(jsonBytes)
  297. }
  298. func UpdateModelPriceByJSONString(jsonStr string) error {
  299. modelPriceMapMutex.Lock()
  300. defer modelPriceMapMutex.Unlock()
  301. modelPriceMap = make(map[string]float64)
  302. return json.Unmarshal([]byte(jsonStr), &modelPriceMap)
  303. }
  304. // GetModelPrice 返回模型的价格,如果模型不存在则返回-1,false
  305. func GetModelPrice(name string, printErr bool) (float64, bool) {
  306. modelPriceMapMutex.RLock()
  307. defer modelPriceMapMutex.RUnlock()
  308. if strings.HasPrefix(name, "gpt-4-gizmo") {
  309. name = "gpt-4-gizmo-*"
  310. }
  311. if strings.HasPrefix(name, "gpt-4o-gizmo") {
  312. name = "gpt-4o-gizmo-*"
  313. }
  314. price, ok := modelPriceMap[name]
  315. if !ok {
  316. if printErr {
  317. common.SysError("model price not found: " + name)
  318. }
  319. return -1, false
  320. }
  321. return price, true
  322. }
  323. func UpdateModelRatioByJSONString(jsonStr string) error {
  324. modelRatioMapMutex.Lock()
  325. defer modelRatioMapMutex.Unlock()
  326. modelRatioMap = make(map[string]float64)
  327. return json.Unmarshal([]byte(jsonStr), &modelRatioMap)
  328. }
  329. // 处理带有思考预算的模型名称,方便统一定价
  330. func handleThinkingBudgetModel(name, prefix, wildcard string) string {
  331. if strings.HasPrefix(name, prefix) && strings.Contains(name, "-thinking-") {
  332. return wildcard
  333. }
  334. return name
  335. }
  336. func GetModelRatio(name string) (float64, bool) {
  337. modelRatioMapMutex.RLock()
  338. defer modelRatioMapMutex.RUnlock()
  339. name = handleThinkingBudgetModel(name, "gemini-2.5-flash", "gemini-2.5-flash-thinking-*")
  340. name = handleThinkingBudgetModel(name, "gemini-2.5-pro", "gemini-2.5-pro-thinking-*")
  341. if strings.HasPrefix(name, "gpt-4-gizmo") {
  342. name = "gpt-4-gizmo-*"
  343. }
  344. ratio, ok := modelRatioMap[name]
  345. if !ok {
  346. return 37.5, operation_setting.SelfUseModeEnabled
  347. }
  348. return ratio, true
  349. }
  350. func DefaultModelRatio2JSONString() string {
  351. jsonBytes, err := json.Marshal(defaultModelRatio)
  352. if err != nil {
  353. common.SysError("error marshalling model ratio: " + err.Error())
  354. }
  355. return string(jsonBytes)
  356. }
  357. func GetDefaultModelRatioMap() map[string]float64 {
  358. return defaultModelRatio
  359. }
  360. func GetCompletionRatioMap() map[string]float64 {
  361. CompletionRatioMutex.RLock()
  362. defer CompletionRatioMutex.RUnlock()
  363. return CompletionRatio
  364. }
  365. func CompletionRatio2JSONString() string {
  366. CompletionRatioMutex.RLock()
  367. defer CompletionRatioMutex.RUnlock()
  368. jsonBytes, err := json.Marshal(CompletionRatio)
  369. if err != nil {
  370. common.SysError("error marshalling completion ratio: " + err.Error())
  371. }
  372. return string(jsonBytes)
  373. }
  374. func UpdateCompletionRatioByJSONString(jsonStr string) error {
  375. CompletionRatioMutex.Lock()
  376. defer CompletionRatioMutex.Unlock()
  377. CompletionRatio = make(map[string]float64)
  378. return json.Unmarshal([]byte(jsonStr), &CompletionRatio)
  379. }
  380. func GetCompletionRatio(name string) float64 {
  381. CompletionRatioMutex.RLock()
  382. defer CompletionRatioMutex.RUnlock()
  383. if strings.Contains(name, "/") {
  384. if ratio, ok := CompletionRatio[name]; ok {
  385. return ratio
  386. }
  387. }
  388. hardCodedRatio, contain := getHardcodedCompletionModelRatio(name)
  389. if contain {
  390. return hardCodedRatio
  391. }
  392. if ratio, ok := CompletionRatio[name]; ok {
  393. return ratio
  394. }
  395. return hardCodedRatio
  396. }
  397. func getHardcodedCompletionModelRatio(name string) (float64, bool) {
  398. lowercaseName := strings.ToLower(name)
  399. if strings.HasPrefix(name, "gpt-4-gizmo") {
  400. name = "gpt-4-gizmo-*"
  401. }
  402. if strings.HasPrefix(name, "gpt-4o-gizmo") {
  403. name = "gpt-4o-gizmo-*"
  404. }
  405. if strings.HasPrefix(name, "gpt-4") && !strings.HasSuffix(name, "-all") && !strings.HasSuffix(name, "-gizmo-*") {
  406. if strings.HasPrefix(name, "gpt-4o") {
  407. if name == "gpt-4o-2024-05-13" {
  408. return 3, true
  409. }
  410. return 4, true
  411. }
  412. // gpt-4.5-preview匹配
  413. if strings.HasPrefix(name, "gpt-4.5-preview") {
  414. return 2, true
  415. }
  416. if strings.HasPrefix(name, "gpt-4-turbo") || strings.HasSuffix(name, "gpt-4-1106") || strings.HasSuffix(name, "gpt-4-1105") {
  417. return 3, true
  418. }
  419. // 没有特殊标记的 gpt-4 模型默认倍率为 2
  420. return 2, false
  421. }
  422. if strings.HasPrefix(name, "o1") || strings.HasPrefix(name, "o3") {
  423. return 4, true
  424. }
  425. if name == "chatgpt-4o-latest" {
  426. return 3, true
  427. }
  428. if strings.Contains(name, "claude-3") {
  429. return 5, true
  430. } else if strings.Contains(name, "claude-sonnet-4") || strings.Contains(name, "claude-opus-4") {
  431. return 5, true
  432. } else if strings.Contains(name, "claude-instant-1") || strings.Contains(name, "claude-2") {
  433. return 3, true
  434. }
  435. if strings.HasPrefix(name, "gpt-3.5") {
  436. if name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
  437. // https://openai.com/blog/new-embedding-models-and-api-updates
  438. // Updated GPT-3.5 Turbo model and lower pricing
  439. return 3, true
  440. }
  441. if strings.HasSuffix(name, "1106") {
  442. return 2, true
  443. }
  444. return 4.0 / 3.0, true
  445. }
  446. if strings.HasPrefix(name, "mistral-") {
  447. return 3, true
  448. }
  449. if strings.HasPrefix(name, "gemini-") {
  450. if strings.HasPrefix(name, "gemini-1.5") {
  451. return 4, true
  452. } else if strings.HasPrefix(name, "gemini-2.0") {
  453. return 4, true
  454. } else if strings.HasPrefix(name, "gemini-2.5-pro") { // 移除preview来增加兼容性,这里假设正式版的倍率和preview一致
  455. return 8, true
  456. } else if strings.HasPrefix(name, "gemini-2.5-flash") { // 同上
  457. if strings.HasSuffix(name, "-nothinking") {
  458. return 4, false
  459. } else {
  460. return 3.5 / 0.6, false
  461. }
  462. }
  463. return 4, false
  464. }
  465. if strings.HasPrefix(name, "command") {
  466. switch name {
  467. case "command-r":
  468. return 3, true
  469. case "command-r-plus":
  470. return 5, true
  471. case "command-r-08-2024":
  472. return 4, true
  473. case "command-r-plus-08-2024":
  474. return 4, true
  475. default:
  476. return 4, false
  477. }
  478. }
  479. // hint 只给官方上4倍率,由于开源模型供应商自行定价,不对其进行补全倍率进行强制对齐
  480. if lowercaseName == "deepseek-chat" || lowercaseName == "deepseek-reasoner" {
  481. return 4, true
  482. }
  483. if strings.HasPrefix(name, "ERNIE-Speed-") {
  484. return 2, true
  485. } else if strings.HasPrefix(name, "ERNIE-Lite-") {
  486. return 2, true
  487. } else if strings.HasPrefix(name, "ERNIE-Character") {
  488. return 2, true
  489. } else if strings.HasPrefix(name, "ERNIE-Functions") {
  490. return 2, true
  491. }
  492. switch name {
  493. case "llama2-70b-4096":
  494. return 0.8 / 0.64, true
  495. case "llama3-8b-8192":
  496. return 2, true
  497. case "llama3-70b-8192":
  498. return 0.79 / 0.59, true
  499. }
  500. return 1, false
  501. }
  502. func GetAudioRatio(name string) float64 {
  503. if strings.Contains(name, "-realtime") {
  504. if strings.HasSuffix(name, "gpt-4o-realtime-preview") {
  505. return 8
  506. } else if strings.Contains(name, "gpt-4o-mini-realtime-preview") {
  507. return 10 / 0.6
  508. } else {
  509. return 20
  510. }
  511. }
  512. if strings.Contains(name, "-audio") {
  513. if strings.HasPrefix(name, "gpt-4o-audio-preview") {
  514. return 40 / 2.5
  515. } else if strings.HasPrefix(name, "gpt-4o-mini-audio-preview") {
  516. return 10 / 0.15
  517. } else {
  518. return 40
  519. }
  520. }
  521. return 20
  522. }
  523. func GetAudioCompletionRatio(name string) float64 {
  524. if strings.HasPrefix(name, "gpt-4o-realtime") {
  525. return 2
  526. } else if strings.HasPrefix(name, "gpt-4o-mini-realtime") {
  527. return 2
  528. }
  529. return 2
  530. }
  531. func ModelRatio2JSONString() string {
  532. modelRatioMapMutex.RLock()
  533. defer modelRatioMapMutex.RUnlock()
  534. jsonBytes, err := json.Marshal(modelRatioMap)
  535. if err != nil {
  536. common.SysError("error marshalling model ratio: " + err.Error())
  537. }
  538. return string(jsonBytes)
  539. }
  540. var defaultImageRatio = map[string]float64{
  541. "gpt-image-1": 2,
  542. }
  543. var imageRatioMap map[string]float64
  544. var imageRatioMapMutex sync.RWMutex
  545. func ImageRatio2JSONString() string {
  546. imageRatioMapMutex.RLock()
  547. defer imageRatioMapMutex.RUnlock()
  548. jsonBytes, err := json.Marshal(imageRatioMap)
  549. if err != nil {
  550. common.SysError("error marshalling cache ratio: " + err.Error())
  551. }
  552. return string(jsonBytes)
  553. }
  554. func UpdateImageRatioByJSONString(jsonStr string) error {
  555. imageRatioMapMutex.Lock()
  556. defer imageRatioMapMutex.Unlock()
  557. imageRatioMap = make(map[string]float64)
  558. return json.Unmarshal([]byte(jsonStr), &imageRatioMap)
  559. }
  560. func GetImageRatio(name string) (float64, bool) {
  561. imageRatioMapMutex.RLock()
  562. defer imageRatioMapMutex.RUnlock()
  563. ratio, ok := imageRatioMap[name]
  564. if !ok {
  565. return 1, false // Default to 1 if not found
  566. }
  567. return ratio, true
  568. }