koishi-plugin-chatluna-qwen-adapter
Version:
qwen adapter for chatluna
603 lines (593 loc) • 20.8 kB
JavaScript
var __defProp = Object.defineProperty;
var __getOwnPropNames = Object.getOwnPropertyNames;
var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
var __commonJS = (cb, mod) => function __require() {
return mod || (0, cb[__getOwnPropNames(cb)[0]])((mod = { exports: {} }).exports, mod), mod.exports;
};
// src/locales/zh-CN.schema.yml
var require_zh_CN_schema = __commonJS({
"src/locales/zh-CN.schema.yml"(exports, module) {
module.exports = { $inner: [{}, { $desc: "请求选项", apiKeys: { $inner: ["阿里云百炼的 API Key", "API 请求地址（可选）", "是否启用此配置"], $desc: "阿里云百炼的 API Key 列表。" }, additionalModels: { $desc: "额外模型列表。", $inner: { model: "模型名称", modelType: "模型类型", contextSize: "模型上下文大小", modelCapabilities: { $desc: "模型支持的能力", $inner: ["工具调用", "图片视觉输入"] } } } }, { $desc: "模型配置", maxContextRatio: "最大上下文使用比例（0~1），控制可用的模型上下文窗口大小的最大百分比。例如 0.35 表示最多使用模型上下文的 35%。", temperature: "回复的随机性程度，数值越高，回复越随机。", enableSearch: "是否启用模型自带夸克搜索功能。" }] };
}
});
// src/locales/en-US.schema.yml
var require_en_US_schema = __commonJS({
"src/locales/en-US.schema.yml"(exports, module) {
module.exports = { $inner: [{}, { $desc: "API Configuration", apiKeys: { $inner: ["Tongyi Qianwen API Key", "API Endpoint (optional)", "Enabled"], $desc: "Tongyi Qianwen API Keys" }, additionalModels: { $desc: "Additional models", $inner: { model: "Model name", modelType: "Model type", contextSize: "Context size", modelCapabilities: { $desc: "Model supported capabilities", $inner: ["Tool calling", "Visual image input"] } } } }, { $desc: "Model Parameters", maxContextRatio: "Maximum context usage ratio (0-1). Controls the maximum percentage of model context window available for use. For example, 0.35 means at most 35% of the model context can be used.", temperature: "Sampling temperature (higher values increase randomness)", enableSearch: "Enable built-in Quark search" }] };
}
});
// src/index.ts
import { ChatLunaPlugin } from "koishi-plugin-chatluna/services/chat";
import { Schema } from "koishi";
// src/client.ts
import { PlatformModelAndEmbeddingsClient } from "koishi-plugin-chatluna/llm-core/platform/client";
import {
ChatLunaChatModel,
ChatLunaEmbeddings
} from "koishi-plugin-chatluna/llm-core/platform/model";
import {
ModelCapabilities,
ModelType
} from "koishi-plugin-chatluna/llm-core/platform/types";
import {
ChatLunaError as ChatLunaError2,
ChatLunaErrorCode as ChatLunaErrorCode2
} from "koishi-plugin-chatluna/utils/error";
// src/requester.ts
import { ChatGenerationChunk } from "@langchain/core/outputs";
import {
ChatLunaError,
ChatLunaErrorCode
} from "koishi-plugin-chatluna/utils/error";
import { sseIterable } from "koishi-plugin-chatluna/utils/sse";
import { trackLogToLocal } from "koishi-plugin-chatluna/utils/logger";
import {
ModelRequester
} from "koishi-plugin-chatluna/llm-core/platform/api";
// src/utils.ts
import {
AIMessageChunk,
ChatMessageChunk,
FunctionMessageChunk,
HumanMessageChunk,
SystemMessageChunk,
ToolMessageChunk
} from "@langchain/core/messages";
import { zodToJsonSchema } from "zod-to-json-schema";
import { isMessageContentImageUrl } from "koishi-plugin-chatluna/utils/string";
import {
fetchImageUrl,
removeAdditionalProperties
} from "@chatluna/v1-shared-adapter";
import { isZodSchemaV3 } from "@langchain/core/utils/types";
function formatToolsToQWenTools(tools) {
if (tools.length < 1) {
return void 0;
}
return tools.map(formatToolToQWenTool);
}
__name(formatToolsToQWenTools, "formatToolsToQWenTools");
function formatToolToQWenTool(tool) {
const parameters = removeAdditionalProperties(
isZodSchemaV3(tool.schema) ? zodToJsonSchema(tool.schema, {
allowedAdditionalProperties: void 0
}) : tool.schema
);
return {
type: "function",
function: {
name: tool.name,
description: tool.description,
// any?
// eslint-disable-next-line @typescript-eslint/no-explicit-any
parameters
}
};
}
__name(formatToolToQWenTool, "formatToolToQWenTool");
async function langchainMessageToQWenMessage(messages, plugin, model) {
const result = [];
for (const rawMessage of messages) {
const role = messageTypeToQWenRole(rawMessage.getType());
const msg = {
content: rawMessage.content || null,
name: role === "assistant" || role === "tool" ? rawMessage.name : void 0,
role,
tool_call_id: rawMessage.tool_call_id
};
if (rawMessage.getType() === "ai") {
const toolCalls = rawMessage.tool_calls;
if (Array.isArray(toolCalls) && toolCalls.length > 0) {
msg.tool_calls = toolCalls.map((toolCall) => ({
id: toolCall.id,
type: "function",
function: {
name: toolCall.name,
arguments: JSON.stringify(toolCall.args)
}
}));
}
}
if (msg.tool_calls == null) {
delete msg.tool_calls;
}
if (msg.tool_call_id == null) {
delete msg.tool_call_id;
}
if (msg.tool_calls) {
for (const toolCall of msg.tool_calls) {
const tool = toolCall.function;
if (!tool.arguments) {
continue;
}
tool.arguments = JSON.stringify(JSON.parse(tool.arguments));
}
}
const images = rawMessage.additional_kwargs.images;
if ((model?.includes("qwen-vl") || model?.includes("omni") || model?.includes("qwen2.5-vl") || model?.includes("qwen2.5-omni") || model?.includes("qwen-omni") || model?.includes("qwen2-vl") || model?.includes("qvq")) && images != null) {
msg.content = [
{
type: "text",
text: rawMessage.content
}
];
for (const image of images) {
msg.content.push({
type: "image_url",
image_url: {
url: image,
detail: "low"
}
});
}
} else if (Array.isArray(msg.content) && msg.content.length > 0) {
msg.content = await Promise.all(
msg.content.map(async (content) => {
if (!isMessageContentImageUrl(content)) return content;
try {
const url = await fetchImageUrl(plugin, content);
return {
type: "image_url",
image_url: {
url,
detail: "low"
}
};
} catch {
return content;
}
})
);
}
result.push(msg);
}
return result;
}
__name(langchainMessageToQWenMessage, "langchainMessageToQWenMessage");
function messageTypeToQWenRole(type) {
switch (type) {
case "system":
return "system";
case "ai":
return "assistant";
case "human":
return "user";
case "function":
return "function";
case "tool":
return "tool";
default:
throw new Error(`Unknown message type: ${type}`);
}
}
__name(messageTypeToQWenRole, "messageTypeToQWenRole");
function convertDeltaToMessageChunk(delta, defaultRole) {
const role = ((delta.role?.length ?? 0) > 0 ? delta.role : defaultRole).toLowerCase();
const content = delta.content ?? "";
const reasoningContent = delta.reasoning_content ?? "";
const additionalKwargs = {};
if (reasoningContent.length > 0) {
additionalKwargs.reasoning_content = reasoningContent;
}
if (role === "user") {
return new HumanMessageChunk({ content });
} else if (role === "assistant") {
const toolCallChunks = [];
if (Array.isArray(delta.tool_calls)) {
for (const rawToolCall of delta.tool_calls) {
toolCallChunks.push({
name: rawToolCall.function?.name,
args: rawToolCall.function?.arguments,
id: rawToolCall.id,
index: rawToolCall.index
});
}
}
return new AIMessageChunk({
content,
tool_call_chunks: toolCallChunks,
additional_kwargs: additionalKwargs
});
} else if (role === "system") {
return new SystemMessageChunk({ content });
} else if (role === "function") {
return new FunctionMessageChunk({
content,
additional_kwargs: additionalKwargs,
name: delta.name
});
} else if (role === "tool") {
return new ToolMessageChunk({
content,
additional_kwargs: additionalKwargs,
tool_call_id: delta.tool_call_id
});
} else {
return new ChatMessageChunk({ content, role });
}
}
__name(convertDeltaToMessageChunk, "convertDeltaToMessageChunk");
// src/requester.ts
import {
createEmbeddings,
createRequestContext
} from "@chatluna/v1-shared-adapter";
import { AIMessageChunk as AIMessageChunk2 } from "@langchain/core/messages";
import { logger } from "koishi-plugin-chatluna";
var QWenRequester = class extends ModelRequester {
constructor(ctx, _configPool, _pluginConfig, _plugin) {
super(ctx, _configPool, _pluginConfig, _plugin);
this._pluginConfig = _pluginConfig;
}
static {
__name(this, "QWenRequester");
}
async *completionStreamInternal(params) {
let model = params.model;
let enabledThinking = null;
if (model.includes("thinking")) {
enabledThinking = !model.includes("-non-thinking");
model = model.replace("-non-thinking", "").replace("-thinking", "");
} else if (model.includes("default")) {
enabledThinking = true;
model = model.replace("-default", "-thinking");
}
const requestParams = {
model,
messages: await langchainMessageToQWenMessage(
params.input,
this._plugin,
model
),
tools: params.tools != null && !params.model.includes("vl") ? formatToolsToQWenTools(params.tools) : void 0,
stream: true,
top_p: params.topP,
temperature: params.temperature,
enable_search: params.model.includes("vl") ? void 0 : this._pluginConfig.enableSearch,
enabled_thinking: enabledThinking
};
try {
const response = await this.post(
"chat/completions",
requestParams,
{
signal: params.signal
}
);
let iterator;
try {
iterator = sseIterable(response);
} catch (e) {
if (e instanceof ChatLunaError && e.message.includes("data_inspection_failed")) {
throw new ChatLunaError(
ChatLunaErrorCode.API_UNSAFE_CONTENT,
e
);
}
throw e;
}
const defaultRole = "assistant";
let reasoningContent = "";
let isSetReasoningTime = false;
let reasoningTime = 0;
for await (const event of iterator) {
const chunk = event.data;
if (chunk === "[DONE]") {
return;
}
let data;
try {
data = JSON.parse(chunk);
} catch (err) {
throw new ChatLunaError(
ChatLunaErrorCode.API_REQUEST_FAILED,
new Error(
"error when calling qwen completion, Result: " + chunk
)
);
}
const choice = data.choices?.[0];
if (data.usage) {
yield new ChatGenerationChunk({
message: new AIMessageChunk2({
content: "",
response_metadata: {
tokenUsage: {
promptTokens: data.usage.prompt_tokens,
completionTokens: data.usage.completion_tokens,
totalTokens: data.usage.total_tokens
}
}
}),
text: ""
});
}
if (!choice) {
continue;
}
const delta = choice.delta;
const messageChunk = convertDeltaToMessageChunk(
delta,
defaultRole
);
if (delta.reasoning_content) {
reasoningContent = reasoningContent + delta.reasoning_content;
if (reasoningTime === 0) {
reasoningTime = Date.now();
}
}
if ((delta.reasoning_content == null || delta.reasoning_content === "") && delta.content && delta.content.length > 0 && reasoningTime > 0 && !isSetReasoningTime) {
reasoningTime = Date.now() - reasoningTime;
messageChunk.additional_kwargs.reasoning_time = reasoningTime;
isSetReasoningTime = true;
}
const generationChunk = new ChatGenerationChunk({
message: messageChunk,
text: messageChunk.content
});
yield generationChunk;
if (choice.finish_reason === "stop") {
break;
}
}
if (reasoningContent.length > 0) {
logger.debug(
"reasoningContent: " + reasoningContent + ", reasoningTime: " + reasoningTime / 1e3 + "s"
);
}
} catch (e) {
if (this.ctx.chatluna.config.isLog) {
await trackLogToLocal(
"Request",
JSON.stringify(requestParams),
this.logger
);
}
if (e instanceof ChatLunaError) {
throw e;
} else {
throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
}
}
}
async embeddings(params) {
const requestContext = createRequestContext(
this.ctx,
this._config.value,
this._pluginConfig,
this._plugin,
this
);
return await createEmbeddings(requestContext, params);
}
concatUrl(url) {
return "https://dashscope.aliyuncs.com/compatible-mode/v1/" + url;
}
get logger() {
return this.ctx.logger("chatluna-qwen-adapter");
}
};
// src/client.ts
import { supportImageInput } from "@chatluna/v1-shared-adapter";
var QWenClient = class extends PlatformModelAndEmbeddingsClient {
constructor(ctx, _config, plugin) {
super(ctx, plugin.platformConfigPool);
this._config = _config;
this.plugin = plugin;
this._requester = new QWenRequester(
ctx,
plugin.platformConfigPool,
_config,
plugin
);
}
static {
__name(this, "QWenClient");
}
platform = "qwen";
_requester;
async refreshModels() {
const rawModels = [
["qwen-turbo", 1e5],
["qwen-long", 1e6],
["qwen-plus", 131072],
["qwen-plus-character", 32768],
["qwen-max", 30720],
["qwen-max-latest", 131072],
["qwen3-max", 262144],
["qwen-plus-latest-non-thinking", 1e6],
["qwen-plus-latest-thinking", 1e6],
["qwen-turbo-latest-non-thinking", 1e6],
["qwen-turbo-latest-thinking", 1e6],
["qwen-flash", 1e6],
["qwen3-vl-plus-thinking", 262144],
["qwen3-vl-plus-non-thinking", 262144],
["qwen-vl-max", 131072],
["qwen-vl-max-latest", 131072],
["qwen-vl-plus", 131072],
["qwen-vl-plus-latest", 131072],
["qwen-vl-ocr", 34096],
["qwen-vl-ocr-latest", 34096],
["qwq-32b-preview", 30720],
["qvq-72b-preview", 30720],
["qwq-plus", 131072],
["qwq-plus-latest", 131072],
["qwen-omni-turbo", 32768],
["qwen-omni-turbo-latest", 32768],
["qwen-math-plus", 4e3],
["qwen-math-turbo", 4e3],
["qwen3-next-80b-a3b-default", 126976],
["qwen3-next-80b-a3b-instruct", 126024],
["qwen3-235b-a22b-default-2507", 131072],
["qwen3-235b-a22b-instruct-2507", 131072],
["qwen3-32b-thinking", 131072],
["qwen3-32b-non-thinking", 131072],
["qwen3-30b-a3b-thinking", 131072],
["qwen3-30b-a3b-non-thinking", 131072],
["qwen3-14b-thinking", 131072],
["qwen3-14b-non-thinking", 131072],
["qwen3-8b-thinking", 131072],
["qwen3-8b-non-thinking", 131072],
["qwen3-4b-thinking", 131072],
["qwen3-4b-non-thinking", 131072],
["qwen3-1.7b-thinking", 30720],
["qwen3-1.7b-non-thinking", 30720],
["qwen3-0.6b-thinking", 30720],
["qwen3-0.6b-non-thinking", 30720],
["qwen3-omni-flash-thinking", 65536],
["qwen3-omni-flash-non-thinking", 65536],
["qwen-omni-turbo", 32768],
["qwen-omni-latest", 32768],
["qwen3-vl-235b-a22b-default", 131072],
["qwen3-vl-235b-a22b-instruct", 131072],
["qwen2.5-vl-72b-instruct", 131072],
["qwen2.5-vl-32b-instruct", 129024],
["qwen2.5-vl-7b-instruct", 8192],
["qwen2.5-vl-3b-instruct", 8192],
["qwen-vl-v1", 8e3],
["Moonshot-Kimi-K2-Instruct", 131072],
["deepseek-r1", 131072],
["deepseek-v3", 65536],
["text-embedding-v1", 2048],
["text-embedding-v2", 2048],
["text-embedding-v3", 8192]
];
const additionalModels = this._config.additionalModels.map(
({ model, modelType, contextSize, modelCapabilities }) => ({
name: model,
type: modelType === "Embeddings 嵌入模型" ? ModelType.embeddings : ModelType.llm,
capabilities: modelCapabilities,
maxTokens: contextSize ?? 4096
})
);
return rawModels.map(([model, token]) => {
return {
name: model,
type: model.includes("embedding") ? ModelType.embeddings : ModelType.llm,
maxTokens: token,
capabilities: [
(model.includes("qwen-plus") || model.includes("qwen-max") || model.includes("qwen-turbo") || model.includes("qwen3") || model.includes("qwen2.5") || model.includes("omni") || model.includes("Kimi-K2") || model.includes("deepseek")) && ModelCapabilities.ToolCall,
supportImageInput(model) && ModelCapabilities.ImageInput
].filter(Boolean)
};
}).concat(additionalModels);
}
_createModel(model) {
const info = this._modelInfos[model];
if (info == null) {
throw new ChatLunaError2(ChatLunaErrorCode2.MODEL_NOT_FOUND);
}
if (info.type === ModelType.llm) {
const modelMaxContextSize = info.maxTokens;
return new ChatLunaChatModel({
modelInfo: info,
requester: this._requester,
model,
modelMaxContextSize,
maxTokenLimit: Math.floor(
(info.maxTokens || modelMaxContextSize || 128e3) * this._config.maxContextRatio
),
timeout: this._config.timeout,
temperature: this._config.temperature,
maxRetries: this._config.maxRetries,
llmType: "qwen",
isThinkModel: model.includes("reasoner") || model.includes("r1") || model.includes("thinking") || model.includes("qwq")
});
}
return new ChatLunaEmbeddings({
client: this._requester,
model: info.name,
batchSize: 5,
maxRetries: this._config.maxRetries
});
}
};
// src/index.ts
import { ModelCapabilities as ModelCapabilities2 } from "koishi-plugin-chatluna/llm-core/platform/types";
function apply(ctx, config) {
ctx.on("ready", async () => {
const plugin = new ChatLunaPlugin(ctx, config, "qwen");
plugin.parseConfig((config2) => {
return config2.apiKeys.filter(([apiKey, enabled]) => {
return apiKey.length > 0 && enabled;
}).map(([apiKey]) => {
return {
apiKey,
apiEndpoint: "",
platform: "qwen",
chatLimit: config2.chatTimeLimit,
timeout: config2.timeout,
maxRetries: config2.maxRetries,
concurrentMaxSize: config2.chatConcurrentMaxSize
};
});
});
plugin.registerClient(() => new QWenClient(ctx, config, plugin));
await plugin.initClient();
});
}
__name(apply, "apply");
var Config = Schema.intersect([
ChatLunaPlugin.Config,
Schema.object({
apiKeys: Schema.array(
Schema.tuple([
Schema.string().role("secret").default(""),
Schema.boolean().default(true)
])
).default([[]]).role("table"),
additionalModels: Schema.array(
Schema.object({
model: Schema.string(),
modelType: Schema.union([
"LLM 大语言模型",
"Embeddings 嵌入模型"
]).default("LLM 大语言模型"),
modelCapabilities: Schema.array(
Schema.union([
ModelCapabilities2.ToolCall,
ModelCapabilities2.ImageInput
])
).default([ModelCapabilities2.ToolCall]).role("checkbox"),
contextSize: Schema.number().default(128e3)
})
).default([]).role("table")
}),
Schema.object({
maxContextRatio: Schema.number().min(0).max(1).step(1e-4).role("slider").default(0.35),
temperature: Schema.percent().min(0).max(2).step(0.1).default(1),
enableSearch: Schema.boolean().default(true)
})
]).i18n({
"zh-CN": require_zh_CN_schema(),
"en-US": require_en_US_schema()
// eslint-disable-next-line @typescript-eslint/no-explicit-any
});
var inject = ["chatluna"];
var name = "chatluna-qwen-adapter";
export {
Config,
apply,
inject,
name
};