@lobehub/chat
Version:
Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.
110 lines (105 loc) • 3.69 kB
text/typescript
import { AIChatModelCard } from '@/types/aiModel';
// https://docs.wair.ac.cn/maas/jiage.html
const taichuChatModels: AIChatModelCard[] = [
{
abilities: {
reasoning: true,
},
contextWindowTokens: 32_768,
description:
'taichu_o1是新一代推理大模型,通过多模态交互和强化学习实现类人思维链,支持复杂决策推演,在保持高精度输出的同时展现可模型推理的思维路径,适用于策略分析与深度思考等场景。',
displayName: 'Taichu O1',
enabled: true,
id: 'taichu_o1',
type: 'chat',
},
{
abilities: {
functionCall: true,
},
contextWindowTokens: 32_768,
description: '基于海量高质数据训练,具有更强的文本理解、内容创作、对话问答等能力',
displayName: 'Taichu 2.0',
enabled: true,
id: 'taichu_llm',
pricing: {
currency: 'CNY',
input: 2,
output: 2,
},
type: 'chat',
},
{
abilities: {
vision: true,
},
contextWindowTokens: 4096,
description: '融合了图像理解、知识迁移、逻辑归因等能力,在图文问答领域表现突出',
displayName: 'Taichu 2.0 VL',
enabled: true,
id: 'taichu_vl',
pricing: {
currency: 'CNY',
input: 5,
output: 5,
},
type: 'chat',
},
{
abilities: {
reasoning: true,
},
contextWindowTokens: 131_072,
description:
'DeepSeek-R1 是一款强化学习(RL)驱动的推理模型,解决了模型中的重复性和可读性问题。在 RL 之前,DeepSeek-R1 引入了冷启动数据,进一步优化了推理性能。它在数学、代码和推理任务中与 OpenAI-o1 表现相当,并且通过精心设计的训练方法,提升了整体效果。',
displayName: 'DeepSeek R1',
id: 'deepseek_r1',
type: 'chat',
},
{
abilities: {
reasoning: true,
},
contextWindowTokens: 131_072,
description:
'DeepSeek-R1-Distill-Qwen-14B 是基于 Qwen2.5-14B 通过知识蒸馏得到的模型。该模型使用 DeepSeek-R1 生成的 80 万个精选样本进行微调,展现出优秀的推理能力。',
displayName: 'DeepSeek R1 Distill Qwen 14B',
id: 'deepseek_r1_distill_qwen_14b',
type: 'chat',
},
{
abilities: {
reasoning: true,
},
contextWindowTokens: 131_072,
description:
'DeepSeek-R1-Distill-Qwen-32B 是基于 Qwen2.5-32B 通过知识蒸馏得到的模型。该模型使用 DeepSeek-R1 生成的 80 万个精选样本进行微调,在数学、编程和推理等多个领域展现出卓越的性能。',
displayName: 'DeepSeek R1 Distill Qwen 32B',
id: 'deepseek_r1_distill_qwen_32b',
type: 'chat',
},
{
abilities: {
reasoning: true,
},
contextWindowTokens: 131_072,
description:
'DeepSeek-R1-Distill-Llama-70B 是基于 Llama-3.3-70B-Instruct 经过蒸馏训练得到的模型。该模型是 DeepSeek-R1 系列的一部分,通过使用 DeepSeek-R1 生成的样本进行微调,在数学、编程和推理等多个领域展现出优秀的性能。',
displayName: 'DeepSeek R1 Distill Llama 70B',
id: 'deepseek_r1_distill_llama_70b',
type: 'chat',
},
{
abilities: {
reasoning: true,
},
contextWindowTokens: 32_768,
description:
'Qwen 系列中等规模的推理模型。与传统的指令调优模型相比,具备思考和推理能力的 QwQ 在下游任务中,尤其是在解决难题时,能够显著提升性能。',
displayName: 'QwQ 32B',
id: 'qwq_32b',
type: 'chat',
},
];
export const allModels = [...taichuChatModels];
export default allModels;