|
| 1 | +import type { CancellationToken } from 'vscode'; |
| 2 | +import type { Response } from '@env/fetch'; |
| 3 | +import { mistralProviderDescriptor as provider } from '../../constants.ai'; |
| 4 | +import { AIError, AIErrorReason } from '../../errors'; |
| 5 | +import type { AIActionType, AIModel } from './models/model'; |
| 6 | +import { OpenAICompatibleProviderBase } from './openAICompatibleProviderBase'; |
| 7 | + |
| 8 | +type MistralModel = AIModel<typeof provider.id>; |
| 9 | +const models: MistralModel[] = [ |
| 10 | + { |
| 11 | + id: 'mistral-medium-latest', |
| 12 | + name: 'Mistral Medium', |
| 13 | + maxTokens: { input: 131072, output: 4096 }, |
| 14 | + provider: provider, |
| 15 | + }, |
| 16 | + { |
| 17 | + id: 'mistral-medium-2505', |
| 18 | + name: 'Mistral Medium', |
| 19 | + maxTokens: { input: 131072, output: 4096 }, |
| 20 | + provider: provider, |
| 21 | + hidden: true, |
| 22 | + }, |
| 23 | + { |
| 24 | + id: 'codestral-latest', |
| 25 | + name: 'Codestral', |
| 26 | + maxTokens: { input: 262144, output: 4096 }, |
| 27 | + provider: provider, |
| 28 | + default: true, |
| 29 | + }, |
| 30 | + { |
| 31 | + id: 'codestral-2501', |
| 32 | + name: 'Codestral', |
| 33 | + maxTokens: { input: 262144, output: 4096 }, |
| 34 | + provider: provider, |
| 35 | + hidden: true, |
| 36 | + }, |
| 37 | + { |
| 38 | + id: 'mistral-large-latest', |
| 39 | + name: 'Mistral Large', |
| 40 | + maxTokens: { input: 131072, output: 4096 }, |
| 41 | + provider: provider, |
| 42 | + }, |
| 43 | + { |
| 44 | + id: 'mistral-large-2411', |
| 45 | + name: 'Mistral Large', |
| 46 | + maxTokens: { input: 131072, output: 4096 }, |
| 47 | + provider: provider, |
| 48 | + hidden: true, |
| 49 | + }, |
| 50 | + { |
| 51 | + id: 'devstral-small-latest', |
| 52 | + name: 'Devstral Small', |
| 53 | + maxTokens: { input: 131072, output: 4096 }, |
| 54 | + provider: provider, |
| 55 | + }, |
| 56 | + { |
| 57 | + id: 'devstral-small-2505', |
| 58 | + name: 'Devstral Small', |
| 59 | + maxTokens: { input: 131072, output: 4096 }, |
| 60 | + provider: provider, |
| 61 | + hidden: true, |
| 62 | + }, |
| 63 | + { |
| 64 | + id: 'mistral-small-latest', |
| 65 | + name: 'Mistral Small', |
| 66 | + maxTokens: { input: 131072, output: 4096 }, |
| 67 | + provider: provider, |
| 68 | + }, |
| 69 | + { |
| 70 | + id: 'mistral-small-2503', |
| 71 | + name: 'Mistral Small', |
| 72 | + maxTokens: { input: 131072, output: 4096 }, |
| 73 | + provider: provider, |
| 74 | + hidden: true, |
| 75 | + }, |
| 76 | +]; |
| 77 | + |
| 78 | +export class MistralProvider extends OpenAICompatibleProviderBase<typeof provider.id> { |
| 79 | + readonly id = provider.id; |
| 80 | + readonly name = provider.name; |
| 81 | + protected readonly descriptor = provider; |
| 82 | + protected readonly config = { |
| 83 | + keyUrl: 'https://console.mistral.ai/api-keys', |
| 84 | + keyValidator: /^[a-zA-Z0-9]{32}$/, |
| 85 | + }; |
| 86 | + |
| 87 | + getModels(): Promise<readonly AIModel<typeof provider.id>[]> { |
| 88 | + return Promise.resolve(models); |
| 89 | + } |
| 90 | + |
| 91 | + protected getUrl(_model: AIModel<typeof provider.id>): string { |
| 92 | + return 'https://api.mistral.ai/v1/chat/completions'; |
| 93 | + } |
| 94 | + |
| 95 | + protected override fetchCore<TAction extends AIActionType>( |
| 96 | + action: TAction, |
| 97 | + model: AIModel<typeof provider.id>, |
| 98 | + apiKey: string, |
| 99 | + request: object, |
| 100 | + cancellation: CancellationToken | undefined, |
| 101 | + ): Promise<Response> { |
| 102 | + if ('max_completion_tokens' in request) { |
| 103 | + const { max_completion_tokens: max, ...rest } = request; |
| 104 | + request = max ? { max_tokens: max, ...rest } : rest; |
| 105 | + } |
| 106 | + return super.fetchCore(action, model, apiKey, request, cancellation); |
| 107 | + } |
| 108 | + |
| 109 | + protected override async handleFetchFailure<TAction extends AIActionType>( |
| 110 | + rsp: Response, |
| 111 | + action: TAction, |
| 112 | + model: AIModel<typeof provider.id>, |
| 113 | + retries: number, |
| 114 | + maxInputTokens: number, |
| 115 | + ): Promise<{ retry: true; maxInputTokens: number }> { |
| 116 | + if (rsp.status !== 404 && rsp.status !== 429) { |
| 117 | + let json; |
| 118 | + try { |
| 119 | + json = (await rsp.json()) as MistralError | undefined; |
| 120 | + } catch {} |
| 121 | + |
| 122 | + debugger; |
| 123 | + |
| 124 | + const message = `${json?.type}: ${json?.message?.detail |
| 125 | + ?.map(d => `${d.msg} (${d.type} @ ${d.loc.join(',')})`) |
| 126 | + .join(', ')}`; |
| 127 | + |
| 128 | + if (json?.type === 'invalid_request_error') { |
| 129 | + if (message?.includes('prompt is too long')) { |
| 130 | + if (retries < 2) { |
| 131 | + return { retry: true, maxInputTokens: maxInputTokens - 200 * (retries || 1) }; |
| 132 | + } |
| 133 | + |
| 134 | + throw new AIError( |
| 135 | + AIErrorReason.RequestTooLarge, |
| 136 | + new Error(`(${this.name}) ${rsp.status}: ${message || rsp.statusText}`), |
| 137 | + ); |
| 138 | + } |
| 139 | + } |
| 140 | + |
| 141 | + throw new Error(`(${this.name}) ${rsp.status}: ${message || rsp.statusText}`); |
| 142 | + } |
| 143 | + |
| 144 | + return super.handleFetchFailure(rsp, action, model, retries, maxInputTokens); |
| 145 | + } |
| 146 | +} |
| 147 | + |
| 148 | +interface MistralError { |
| 149 | + object: 'error'; |
| 150 | + type: 'invalid_request_error' | string; |
| 151 | + message: { |
| 152 | + detail: { |
| 153 | + type: string; |
| 154 | + msg: string; |
| 155 | + loc: string[]; |
| 156 | + input: number; |
| 157 | + }[]; |
| 158 | + }; |
| 159 | + code: unknown | null; |
| 160 | + param: unknown | null; |
| 161 | +} |
0 commit comments