mirror of
https://github.com/cline/cline.git
synced 2025-06-03 03:59:07 +00:00
f4df887fcd
* feat: Add DeepSeek-R1 (deepseek-reasoner) support - Add new deepseek-reasoner model with proper pricing info - Fix temperature parameter being sent to unsupported deepseek-reasoner model - Improve model selection logic in DeepSeekHandler - Update CHANGELOG with new features and fixes - Bump version to 3.1.11 * style: apply prettier formatting to deepseek provider and api definitions
69 lines
2.5 KiB
TypeScript
69 lines
2.5 KiB
TypeScript
import { Anthropic } from "@anthropic-ai/sdk"
|
|
import OpenAI from "openai"
|
|
import { ApiHandler } from "../"
|
|
import { ApiHandlerOptions, DeepSeekModelId, ModelInfo, deepSeekDefaultModelId, deepSeekModels } from "../../shared/api"
|
|
import { convertToOpenAiMessages } from "../transform/openai-format"
|
|
import { ApiStream } from "../transform/stream"
|
|
|
|
export class DeepSeekHandler implements ApiHandler {
|
|
private options: ApiHandlerOptions
|
|
private client: OpenAI
|
|
|
|
constructor(options: ApiHandlerOptions) {
|
|
this.options = options
|
|
this.client = new OpenAI({
|
|
baseURL: "https://api.deepseek.com/v1",
|
|
apiKey: this.options.deepSeekApiKey,
|
|
})
|
|
}
|
|
|
|
async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
|
|
const model = this.getModel()
|
|
const stream = await this.client.chat.completions.create({
|
|
model: model.id,
|
|
max_completion_tokens: model.info.maxTokens,
|
|
messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
|
|
stream: true,
|
|
stream_options: { include_usage: true },
|
|
// Only set temperature for non-reasoner models
|
|
...(model.id === "deepseek-reasoner" ? {} : { temperature: 0 }),
|
|
})
|
|
|
|
for await (const chunk of stream) {
|
|
const delta = chunk.choices[0]?.delta
|
|
if (delta?.content) {
|
|
yield {
|
|
type: "text",
|
|
text: delta.content,
|
|
}
|
|
}
|
|
|
|
if (chunk.usage) {
|
|
yield {
|
|
type: "usage",
|
|
inputTokens: chunk.usage.prompt_tokens || 0, // (deepseek reports total input AND cache reads/writes, see context caching: https://api-docs.deepseek.com/guides/kv_cache) where the input tokens is the sum of the cache hits/misses, while anthropic reports them as separate tokens. This is important to know for 1) context management truncation algorithm, and 2) cost calculation (NOTE: we report both input and cache stats but for now set input price to 0 since all the cost calculation will be done using cache hits/misses)
|
|
outputTokens: chunk.usage.completion_tokens || 0,
|
|
// @ts-ignore-next-line
|
|
cacheReadTokens: chunk.usage.prompt_cache_hit_tokens || 0,
|
|
// @ts-ignore-next-line
|
|
cacheWriteTokens: chunk.usage.prompt_cache_miss_tokens || 0,
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
getModel(): { id: DeepSeekModelId; info: ModelInfo } {
|
|
const modelId = this.options.apiModelId
|
|
if (!modelId || !(modelId in deepSeekModels)) {
|
|
return {
|
|
id: deepSeekDefaultModelId,
|
|
info: deepSeekModels[deepSeekDefaultModelId],
|
|
}
|
|
}
|
|
return {
|
|
id: modelId as DeepSeekModelId,
|
|
info: deepSeekModels[modelId as DeepSeekModelId],
|
|
}
|
|
}
|
|
}
|