|
- [
- {
- "type":"NLP",
- "name":"自然语言处理",
- "list":[
- {
- "category":"DeepSeek",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/deepseek.png",
- "style":{
- "height":"20px"
- }
- },
- "models":[
- {
- "name":"DeepSeek-R1",
- "descr":"DeepSeek-R1 671B满血版。作为 DeepSeek AI 推出的第一代推理模型,DeepSeek-R1 在数学、代码和推理任务上的表现与 OpenAI-o1 相当。DeepSeek-R1 在强化学习之前引入了冷启动数据,通过大规模强化学习(RL)训练的模型,在没有监督微调(SFT)作为预处理步骤的情况下表现出了出色的推理能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-V3",
- "descr":"这是一款强大的专家混合(MoE)语言模型,总参数量达 6710 亿,每个词元激活的参数为 370 亿。为了实现高效推理和成本效益高的训练,DeepSeek-V3 采用了多头潜在注意力(MLA)和 DeepSeekMoE 架构。此外,DeepSeek-V3 开创了一种无需辅助损失的负载均衡策略,并设定了多词元预测训练目标,以实现更强的性能。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-V3",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-V3",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-V3-MindFormers",
- "descr":"DeepSeek-V3适配基于MindSpore的Mindformers的模型权重",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai-mindformers/modelmanage/model_readme_tmpl?name=DeepSeek-V3-MindFormers"
- },
- {
- "name":"DeepSeek-R1-Distill-Qwen-1.5B",
- "descr":"基于Qwen的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Qwen-1.5B",
- "experienceGpu":{
- "url":"",
- "type":"chat",
- "recommend":true,
- "title":"独立部署体验"
- },
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Qwen-1.5B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-R1-Distill-Qwen-7B",
- "descr":"基于Qwen的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Qwen-7B",
- "experienceGpu":{
- "url":"",
- "type":"chat",
- "recommend":true,
- "title":"独立部署体验"
- },
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Qwen-7B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-R1-Distill-Qwen-14B",
- "descr":"基于Qwen的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Qwen-14B",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Qwen-14B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-R1-Distill-Qwen-32B",
- "descr":"基于Qwen的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Qwen-32B",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Qwen-32B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-R1-Distill-Llama-8B",
- "descr":"基于Llama的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Llama-8B",
- "experienceGpu":{
- "url":"",
- "type":"chat",
- "recommend":true,
- "title":"独立部署体验"
- },
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Llama-8B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"DeepSeek-R1-Distill-Llama-70B",
- "descr":"基于Llama的DeepSeek-R1模型,DeepSeek-R1-Distill模型的使用方式与Qwen或Llama模型相同。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=DeepSeek-R1-Distill-Llama-70B",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=DeepSeek-R1-Distill-Llama-70B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- }
- ]
- },
- {
- "category":"Qwen",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/qwen.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"QwQ-32B",
- "descr":"QwQ 是 Qwen 系列的推理模型。与传统的指令调优模型相比,QwQ 具备思考和推理能力,在下游任务中,尤其是困难问题上,能够实现显著增强的性能。QwQ-32B 是中型推理模型,能够在推理任务上与最先进的模型(如 DeepSeek-R1、o1-mini)竞争,表现出色。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Qwen/modelmanage/model_filelist_tmpl?name=QwQ-32B",
- "experienceNpu":{
- "url":"https://openi.pcl.ac.cn/extension/deepseek/chat?name=QwQ-32B",
- "recommend":true,
- "free":true,
- "title":"共享服务体验"
- }
- },
- {
- "name":"Qwen2-7B-Instruct",
- "descr":"Qwen是阿里巴巴集团Qwen团队研发的大语言模型和大型多模态模型系列。目前,大语言模型已升级至Qwen2版本。无论是语言模型还是多模态模型,均在大规模多语言和多模态数据上进行预训练,并通过高质量数据进行后期微调以贴近人类偏好。Qwen具备自然语言理解、文本生成、视觉理解、音频理解、工具使用、角色扮演、作为AI Agent进行互动等多种能力。本模型是Qwen2-7B-Instruct。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Qwen/modelmanage/model_readme_tmpl?name=Qwen2-7B-Instruct",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Qwen2-1.5B-Instruct",
- "descr":"Qwen是阿里巴巴集团Qwen团队研发的大语言模型和大型多模态模型系列。目前,大语言模型已升级至Qwen2版本。无论是语言模型还是多模态模型,均在大规模多语言和多模态数据上进行预训练,并通过高质量数据进行后期微调以贴近人类偏好。Qwen具备自然语言理解、文本生成、视觉理解、音频理解、工具使用、角色扮演、作为AI Agent进行互动等多种能力。本模型是Qwen2-1.5B-Instruct。",
- "detail":"https://openi.pcl.ac.cn/OpenIOSSG/OpenI_LLM_Finetune_Example/modelmanage/model_readme_tmpl?name=Qwen2-1.5B-Instruct",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- }
- ]
- },
- {
- "category":"鹏城·脑海",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/pc-mind.png",
- "style":{
- "height":"20px"
- }
- },
- "models":[
- {
- "name":"PengCheng-Mind 200B",
- "descr":"PengCheng-Mind 200B 脑海大模型是基于Transformer 架构的2010亿参数自回归式语言模型。模型全流程基于“鹏城云脑Ⅱ”(中国算力网枢纽节点)的全自主安全可控国产软硬件平台进行开发和训练,采用MindSpore框架实现在大规模集群上长期稳定的多维分布式并行训练。",
- "detail":"https://openi.pcl.ac.cn/PengChengMind/PengCheng.Mind"
- },
- {
- "name":"PengCheng-Mind 7B",
- "descr":"PengCheng-Mind 7B 脑海大模型是基于中国算力网训练,可实现跨算力中心迁移的70亿参数自回归式语言模型,该模型支持NPU版本到GPU版本的转换,具有强大的语言理解和生成能力,可深入理解输入文本,自动生成连贯、合理的语句。模型可用于多种自然语言处理任务,如文本生成、机器翻译、问答系统等。PengCheng-Mind 7B在自然语言处理领域具有广阔的应用前景,有利于加速人工智能的学术研究和智能应用的开发过程。",
- "detail":"https://openi.pcl.ac.cn/PengChengMind/PengChengMind-7B",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"PengCheng-Mind 2.6B",
- "descr":"PengCheng-Mind 2.6B(原鹏城·盘古) 脑海大模型是业界首个2000亿参数以中文为核心的预训练生成语言模型。",
- "detail":"https://openi.pcl.ac.cn/PCL-Platform.Intelligence/pcl_pangu"
- }
- ]
- },
- {
- "category":"Llama",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/llama.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"Meta-Llama-3.1-8B",
- "descr":"Llama3.1-8B基础版本,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/meta-llama/modelmanage/model_readme_tmpl?name=Meta-Llama-3.1-8B"
- },
- {
- "name":"Meta-Llama-3.1-8B-Instruct",
- "descr":"Llama3.1-8B-Instruct版本,指令调优版,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/meta-llama/modelmanage/model_readme_tmpl?name=Meta-Llama-3.1-8B-Instruct",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Meta-Llama-3-8B",
- "descr":"Llama3-8B基础版本,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama3/modelmanage/model_readme_tmpl?name=Meta-Llama-3-8B"
- },
- {
- "name":"Meta-Llama-3-8B-Instruct",
- "descr":"Llama3-8B-Instruct版本,指令调优版,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama3/modelmanage/model_readme_tmpl?name=Meta-Llama-3-8B-Instruct",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Meta-Llama-3-70B",
- "descr":"Llama3-70B基础版本,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama3/modelmanage/model_readme_tmpl?name=Meta-Llama-3-70B"
- },
- {
- "name":"Meta-Llama-3-70B-Instruct",
- "descr":"Llama3-70B-Instruct版本,指令调优版,内含Huggingface格式与original原始格式。Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama3/modelmanage/model_readme_tmpl?name=Meta-Llama-3-70B-Instruct"
- },
- {
- "name":"LlamaGuard-7b",
- "descr":"LlamaGuard是一款基于Llama2-7b的安全保障模型。它可以用于对 LLM 输入 (提示分类) 和 LLM 输出 (响应分类) 进行内容分类。它以 LLM 的方式运行:根据给定的提示或响应是否安全/不安全,在其输出中生成文本进行指示,并且如果根据策略判断为不安全,还会列出违规的子类别。简单来说,Llama-Guard 可以帮助识别 LLM 输入和输出中的潜在安全风险,并提供具体的违规类别信息。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/PurpleLlama/modelmanage/model_readme_tmpl?name=LlamaGuard-7b"
- },
- {
- "name":"llama2-7b-chat-hf",
- "descr":"Meta开源大模型llama2-7b,transformers格式对话增强版;全英文预训练模型,可使用中文微调,对学术研究或商用开放",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama2/modelmanage/model_readme_tmpl?name=llama2-7b-chat-hf"
- },
- {
- "name":"llama2-7b-hf",
- "descr":"Meta开源大模型llama2-7b,transformers格式基础版;全英文预训练模型,可使用中文微调,对学术研究或商用开放",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama2/modelmanage/model_readme_tmpl?name=llama2-7b-hf"
- },
- {
- "name":"llama2-7b-chat",
- "descr":"Meta开源大模型llama2-7b,原始格式对话增强版;全英文预训练模型,可使用中文微调,对学术研究或商用开放",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama2/modelmanage/model_readme_tmpl?name=llama2-7b-chat"
- },
- {
- "name":"llama2-7b",
- "descr":"Meta开源大模型llama2-7b,原始格式基础版;全英文预训练模型,可使用中文微调,对学术研究或商用开放",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama2/modelmanage/model_readme_tmpl?name=llama2-7b"
- },
- {
- "name":"llama_model_7B",
- "descr":"Meta(原Facebook)公司开源的预训练大预言模型 LLaMA(Large Language Model Meta AI)。LLaMA模型是在大量未标记数据上训练而成,它可以使用更少的计算能力和资源来为各种任务进行微调。Meta提供了几种不同规模(7B、13B、33B 和 65B 参数)的 LLaMA 模型。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama/modelmanage/model_readme_tmpl?name=llama_model_7B"
- },
- {
- "name":"llama_model_13B",
- "descr":"Meta(原Facebook)公司开源的预训练大预言模型 LLaMA(Large Language Model Meta AI)。LLaMA模型是在大量未标记数据上训练而成,它可以使用更少的计算能力和资源来为各种任务进行微调。Meta提供了几种不同规模(7B、13B、33B 和 65B 参数)的 LLaMA 模型。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama/modelmanage/model_readme_tmpl?name=llama_model_13B"
- },
- {
- "name":"llama_model_30B",
- "descr":"Meta(原Facebook)公司开源的预训练大预言模型 LLaMA(Large Language Model Meta AI)。LLaMA模型是在大量未标记数据上训练而成,它可以使用更少的计算能力和资源来为各种任务进行微调。Meta提供了几种不同规模(7B、13B、33B 和 65B 参数)的 LLaMA 模型。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama/modelmanage/model_readme_tmpl?name=llama_model_30B"
- },
- {
- "name":"llama_model_65B",
- "descr":"Meta(原Facebook)公司开源的预训练大预言模型 LLaMA(Large Language Model Meta AI)。LLaMA模型是在大量未标记数据上训练而成,它可以使用更少的计算能力和资源来为各种任务进行微调。Meta提供了几种不同规模(7B、13B、33B 和 65B 参数)的 LLaMA 模型。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/llama/modelmanage/model_readme_tmpl?name=llama_model_65B"
- }
- ]
- },
- {
- "category":"Yi",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/yi.png",
- "style":{
- "height":"32px"
- }
- },
- "models":[
- {
- "name":"Yi-1.5-6B-Chat",
- "descr":"Yi-1.5的60亿参数对话版。Yi-1.5是Yi的升级版本。它在Yi的基础上,使用包含5000亿高质量语料的语料库进行持续预训练,并在300万个多样化的微调样本上进行微调。与Yi相比,Yi-1.5在编程、数学、推理和指令遵循能力方面表现更强,同时仍然保持了出色的语言理解、常识推理和阅读理解能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/01-ai/modelmanage/model_readme_tmpl?name=Yi-1.5-6B-Chat",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Yi-1.5-9B-Chat",
- "descr":"Yi-1.5的90亿参数对话版。Yi-1.5是Yi的升级版本。它在Yi的基础上,使用包含5000亿高质量语料的语料库进行持续预训练,并在300万个多样化的微调样本上进行微调。与Yi相比,Yi-1.5在编程、数学、推理和指令遵循能力方面表现更强,同时仍然保持了出色的语言理解、常识推理和阅读理解能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/01-ai/modelmanage/model_readme_tmpl?name=Yi-1.5-9B-Chat",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Yi-1.5-34B-Chat",
- "descr":"Yi-1.5的340亿参数对话版。Yi-1.5是Yi的升级版本。它在Yi的基础上,使用包含5000亿高质量语料的语料库进行持续预训练,并在300万个多样化的微调样本上进行微调。与Yi相比,Yi-1.5在编程、数学、推理和指令遵循能力方面表现更强,同时仍然保持了出色的语言理解、常识推理和阅读理解能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/01-ai/modelmanage/model_readme_tmpl?name=Yi-1.5-34B-Chat"
- },
- {
- "name":"Yi-1.5-34B",
- "descr":"Yi-1.5的340亿参数基座模型。Yi-1.5是Yi的升级版本。它在Yi的基础上,使用包含5000亿高质量语料的语料库进行持续预训练,并在300万个多样化的微调样本上进行微调。与Yi相比,Yi-1.5在编程、数学、推理和指令遵循能力方面表现更强,同时仍然保持了出色的语言理解、常识推理和阅读理解能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/01-ai/modelmanage/model_readme_tmpl?name=Yi-1.5-34B"
- }
- ]
- },
- {
- "category":"Mistral AI",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/mistral-ai.png",
- "style":{
- "height":"26px"
- }
- },
- "models":[
- {
- "name":"Mistral-7B-Instruct-v0.2",
- "descr":"Mistral.AI 7B大模型第二代的指令调优版。快速部署且易于定制。尽管小巧,但对各种用例都非常强大。在英语和编码方面性能出色。支持32k上下文窗口。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/mistralai/modelmanage/model_readme_tmpl?name=Mistral-7B-Instruct-v0.2"
- },
- {
- "name":"Mixtral-8x7B-Instruct-v0.1",
- "descr":"7B的稀疏专家混合模型(SMoE)指令调优版。在总共45B的参数中使用了12.9B个活跃参数。精通英语、法语、意大利语、德语、西班牙语,对编码有很强的掌握。支持32k上下文窗口。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/mistralai/modelmanage/model_readme_tmpl?name=Mixtral-8x7B-Instruct-v0.1"
- },
- {
- "name":"Mixtral-8x22B-Instruct-v0.1",
- "descr":"22B的稀疏专家混合模型(SMoE)指令调优版,目前是 Mixtral.AI 性能最好的开源模型。仅使用了141B中的39B活跃参数。精通英语、法语、意大利语、德语、西班牙语,对编码有很强的掌握。支持64k上下文窗口,原生函数调用能力。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/mistralai/modelmanage/model_readme_tmpl?name=Mixtral-8x22B-Instruct-v0.1"
- }
- ]
- },
- {
- "category":"Gemma",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/gemma.png"
- },
- "models":[
- {
- "name":"gemma-7b",
- "descr":"Google Gemma 70亿参数版本. Gemma is a family of open-weights Large Language Model (LLM) by Google DeepMind, based on Gemini research and technology.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/gemma/modelmanage/model_readme_tmpl?name=gemma-7b"
- },
- {
- "name":"gemma-7b-it",
- "descr":"Google Gemma 70亿参数Instruct版本. Gemma is a family of open-weights Large Language Model (LLM) by Google DeepMind, based on Gemini research and technology.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/gemma/modelmanage/model_readme_tmpl?name=gemma-7b-it"
- },
- {
- "name":"gemma-2b",
- "descr":"Google Gemma 20亿参数版本. Gemma is a family of open-weights Large Language Model (LLM) by Google DeepMind, based on Gemini research and technology.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/gemma/modelmanage/model_readme_tmpl?name=gemma-2b"
- },
- {
- "name":"gemma-2b-it",
- "descr":"Google Gemma 20亿参数Instruct版本. Gemma is a family of open-weights Large Language Model (LLM) by Google DeepMind, based on Gemini research and technology.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/gemma/modelmanage/model_readme_tmpl?name=gemma-2b-it"
- }
- ]
- },
- {
- "category":"讯飞星火",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/xunfei.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"iFlytekSpark-13B",
- "descr":"讯飞星火开源-13B(iFlytekSpark-13B)拥有130亿参数,在经过累计超过3万亿以上tokens海量高质量数据集上进行预训练,然后在精调得多元化对齐数据上进行微调得到。iFlytekSpark-13B不仅具备通用任务处理能力如聊天、问答、文本提取和分类等,还具备数据分析和代码生成等生产力功能。",
- "detail":"https://openi.pcl.ac.cn/iflytek/iFlytekSpark-13B"
- },
- {
- "name":"Spark13B_0206",
- "descr":"讯飞星火开源-13B(iFlytekSpark-13B)拥有130亿参数,在经过累计超过3万亿以上tokens海量高质量数据集上进行预训练,然后在精调得多元化对齐数据上进行微调得到。iFlytekSpark-13B不仅具备通用任务处理能力如聊天、问答、文本提取和分类等,还具备数据分析和代码生成等生产力功能。",
- "detail":"https://openi.pcl.ac.cn/iflytek/iFlytekSpark-13B"
- }
- ]
- },
- {
- "category":"ChatGLM",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/chatglm.png",
- "style":{
- "border-radius":"8px",
- "height":"32px"
- }
- },
- "models":[
- {
- "name":"ChatGLM-6B",
- "descr":"ChatGLM-6B 是一个开源的、支持中英双语问答的对话语言模型,基于 General Language Model (GLM) 架构,具有 62 亿参数。结合模型量化技术,用户可以在消费级的显卡上进行本地部署(INT4 量化级别下最低只需 6GB 显存)。ChatGLM-6B 使用了和 ChatGLM 相同的技术,针对中文问答和对话进行了优化。经过约 1T 标识符的中英双语训练,辅以监督微调、反馈自助、人类反馈强化学习等技术的加持,62 亿参数的 ChatGLM-6B 已经能生成相当符合人类偏好的回答。",
- "detail":"https://openi.pcl.ac.cn/Zhipu.AI/ChatGLM-6B/modelmanage/model_readme_tmpl?name=ChatGLM-6B"
- },
- {
- "name":"ChatGLM2-6B",
- "descr":"ChatGLM2-6B 是开源中英双语对话模型 ChatGLM-6B 的第二代版本,在保留了初代模型对话流畅、部署门槛较低等众多优秀特性的基础之上,ChatGLM2-6B 引入了新特性:更强大的性能、更长的上下文、更高效的推理、更开放的协议。",
- "detail":"https://openi.pcl.ac.cn/Zhipu.AI/ChatGLM2-6B/modelmanage/model_readme_tmpl?name=ChatGLM2-6B"
- },
- {
- "name":"ChatGLM3-6B",
- "descr":"ChatGLM3-6B 是智谱AI和清华大学 KEG 实验室联合发布的新一代对话预训练模型。ChatGLM3-6B 是 ChatGLM3 系列中的开源模型,在保留了前两代模型对话流畅、部署门槛低等特性的基础上,ChatGLM3-6B 引入了:1. 更强大的基础模型:更多样的训练数据、更充分的训练步数和更合理的训练策略;2. 更完整的功能支持:ChatGLM3-6B 采用了全新对话格式,除多轮对话外。同时原生支持工具调用、代码解释器和智能体任务等。",
- "detail":"https://openi.pcl.ac.cn/OpenIOSSG/OpenI_LLM_Finetune_Example/modelmanage/model_readme_tmpl?name=chatglm3-6b",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"ChatGLM2-6B-32k",
- "descr":"ChatGLM2-6B长对话版本;ChatGLM2-6B-32K在ChatGLM2-6B的基础上进一步强化了对于长文本的理解能力,能够更好的处理最多32K长度的上下文。具体地,我们基于位置插值(Positional Interpolation)的方法对位置编码进行了更新,并在对话阶段使用 32K 的上下文长度训练。在实际的使用中,如果您面临的上下文长度基本在 8K 以内,我们推荐使用ChatGLM2-6B;如果您需要处理超过 8K 的上下文长度,我们推荐使用ChatGLM2-6B-32K。",
- "detail":"https://openi.pcl.ac.cn/Zhipu.AI/ChatGLM2-6B/modelmanage/model_readme_tmpl?name=ChatGLM2-6B-32k"
- },
- {
- "name":"ChatGLM3-6B-32k",
- "descr":"ChatGLM3-6B长对话版本;ChatGLM3-6B-32K在ChatGLM3-6B的基础上进一步强化了对于长文本的理解能力,能够更好的处理最多32K长度的上下文。具体地,我们对位置编码进行了更新,并设计了更有针对性的长文本训练方法,在对话阶段使用 32K 的上下文长度训练。在实际的使用中,如果您面临的上下文长度基本在 8K 以内,我们推荐使用ChatGLM3-6B;如果您需要处理超过 8K 的上下文长度,我们推荐使用ChatGLM3-6B-32K。",
- "detail":"https://openi.pcl.ac.cn/Zhipu.AI/ChatGLM3/modelmanage/model_readme_tmpl?name=chatglm3-6b-32k"
- },
- {
- "name":"CodeGeeX2-6B",
- "descr":"CodeGeeX2 是多语言代码生成模型 CodeGeeX (KDD’23) 的第二代模型。CodeGeeX2 基于 ChatGLM2 架构加入代码预训练实现,得益于 ChatGLM2 的更优性能,CodeGeeX2 在多项指标上取得性能提升(+107% > CodeGeeX;仅60亿参数即超过150亿参数的 StarCoder-15B 近10%)",
- "detail":"https://openi.pcl.ac.cn/Zhipu.AI/CodeGeeX2-6B/modelmanage/model_readme_tmpl?name=CodeGeeX2-6B"
- }
- ]
- },
- {
- "category":"Baichuan2",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/baichuan.png",
- "style":{
- "border-radius":"8px"
- }
- },
- "models":[
- {
- "name":"Baichuan2-NPU",
- "descr":"Baichuan2-NPU版本,包含7B与13B模型。Baichuan2 是由百川智能开发的开源可商用的大规模预训练语言模型,基于 Transformer 结构,支持中英双语,上下文窗口长度为 4096。目前支持Baichuan2-7B和Baichuan2-13B模型,参数量分别为70亿和130亿。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Baichuan2/modelmanage/model_readme_tmpl?name=Baichuan2-npu"
- },
- {
- "name":"Baichuan2-7B-Chat",
- "descr":"Baichuan2-7B-对话增强版。Baichuan 2 是百川智能推出的新一代开源大语言模型,采用 2.6 万亿 Tokens 的高质量语料训练,在权威的中文和英文 benchmark上均取得同尺寸最好的效果。本次发布包含有 7B、13B 的 Base 和 Chat 版本,并提供了 Chat 版本的 4bits量化,所有版本不仅对学术研究完全开放,开发者也仅需邮件申请并获得官方商用许可后,即可以免费商用。",
- "detail":"https://openi.pcl.ac.cn/OpenIOSSG/OpenI_LLM_Finetune_Example/modelmanage/model_readme_tmpl?name=Baichuan2-7b-chat",
- "experience":{
- "url":"",
- "type":"chat",
- "recommend":true
- }
- },
- {
- "name":"Baichuan2-7B-Base",
- "descr":"Baichuan2-7B-基础版。Baichuan 2 是百川智能推出的新一代开源大语言模型,采用 2.6 万亿 Tokens 的高质量语料训练,在权威的中文和英文 benchmark上均取得同尺寸最好的效果。本次发布包含有 7B、13B 的 Base 和 Chat 版本,并提供了 Chat 版本的 4bits量化,所有版本不仅对学术研究完全开放,开发者也仅需邮件申请并获得官方商用许可后,即可以免费商用。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Baichuan2/modelmanage/model_readme_tmpl?name=Baichuan2-7b-base"
- },
- {
- "name":"Baichuan2-13B-Chat",
- "descr":"Baichuan2-13B-对话增强版。Baichuan 2 是百川智能推出的新一代开源大语言模型,采用 2.6 万亿 Tokens 的高质量语料训练,在权威的中文和英文 benchmark上均取得同尺寸最好的效果。本次发布包含有 7B、13B 的 Base 和 Chat 版本,并提供了 Chat 版本的 4bits量化,所有版本不仅对学术研究完全开放,开发者也仅需邮件申请并获得官方商用许可后,即可以免费商用。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Baichuan2/modelmanage/model_readme_tmpl?name=Baichuan2-13b-chat"
- },
- {
- "name":"Baichuan2-13B-Base",
- "descr":"Baichuan2-13B-基础版。Baichuan 2 是百川智能推出的新一代开源大语言模型,采用 2.6 万亿 Tokens 的高质量语料训练,在权威的中文和英文 benchmark上均取得同尺寸最好的效果。本次发布包含有 7B、13B 的 Base 和 Chat 版本,并提供了 Chat 版本的 4bits量化,所有版本不仅对学术研究完全开放,开发者也仅需邮件申请并获得官方商用许可后,即可以免费商用。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Baichuan2/modelmanage/model_readme_tmpl?name=Baichuan2-13b-base"
- }
- ]
- },
- {
- "category":"Vicuna",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/vicuna.png",
- "style":{
- "border-radius":"8px",
- "height":"32px"
- }
- },
- "models":[
- {
- "name":"vicuna-7B-1.1",
- "descr":"Vicuna is an open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT. It is an auto-regressive language model, based on the transformer architecture.",
- "detail":"https://openi.pcl.ac.cn/Learning-Develop-Union/LangChain-ChatGLM-Webui/modelmanage/model_readme_tmpl?name=vicuna-7B-1.1"
- },
- {
- "name":"vicuna-13b-1.1",
- "descr":"Vicuna is an open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT. It is an auto-regressive language model, based on the transformer architecture.",
- "detail":"https://openi.pcl.ac.cn/Learning-Develop-Union/LangChain-ChatGLM-Webui/modelmanage/model_readme_tmpl?name=vicuna-13b-1.1"
- }
- ]
- }
- ]
- },
- {
- "type":"CV",
- "name":"计算机视觉",
- "list":[
- {
- "category":"Stable Diffusion",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/stable-diffusion.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"stabilityai-stable_diffusion-3.5-large",
- "descr":"stabilityai-stable_diffusion-3.5-large是目前最新的sd文生图模型",
- "detail":"https://openi.pcl.ac.cn/OpenIOSSG/sd-scripts-server/modelmanage/model_readme_tmpl?name=stabilityai-stable_diffusion-3.5-large"
- },
- {
- "name":"Stable-Diffusion-XL-Base-1.0",
- "descr":"Stable-Diffusion-XL-Base-1.0是基础的文生图模型",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/stabilityai/modelmanage/model_readme_tmpl?name=stable-diffusion-xl-base-1.0",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- },
- {
- "name":"Stable-Diffusion-3-Medium-Diffusers",
- "descr":"Stable Diffusion 3 Medium is a Multimodal Diffusion Transformer (MMDiT) text-to-image model that features greatly improved performance in image quality, typography, complex prompt understanding, and resource-efficiency.",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/v2ray/modelmanage/model_readme_tmpl?name=stable-diffusion-3-medium-diffusers",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- },
- {
- "name":"HelloworldXL60",
- "descr":"基于Stable-Diffusion-XL-Base-1.0,对包括超现实、私房、小朋友、合影、口罩、折纸、3D render、汽车、龙、孕照摄影等等一些主题进行了效果提升",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/misri/modelmanage/model_readme_tmpl?name=leosamsHelloworldXL_helloworldXL60",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- },
- {
- "name":"AAM_XL_AnimeMix",
- "descr":"基于Stable-Diffusion-XL-Base-1.0,对2D动画人物进行效果提升",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Lykon/modelmanage/model_readme_tmpl?name=AAM_XL_AnimeMix",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- }
- ]
- },
- {
- "category":"DeepSeek-CV",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/deepseek.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"Janus-Pro-7B",
- "descr":"Janus Pro是一个统一的理解和生成MLLM,它将视觉编码解耦,用于多模态理解和生成。Janus Pro是基于DeepSeek-LLM-1.5b-base/DeepSeek-LLM-7b-base构建的。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=Janus-Pro-7B",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- },
- {
- "name":"Janus-Pro-1B",
- "descr":"Janus Pro是一个统一的理解和生成MLLM,它将视觉编码解耦,用于多模态理解和生成。Janus Pro是基于DeepSeek-LLM-1.5b-base/DeepSeek-LLM-7b-base构建的。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/deepseek-ai/modelmanage/model_readme_tmpl?name=Janus-Pro-1B",
- "experience":{
- "url":"",
- "type":"sd",
- "recommend":true
- }
- }
- ]
- },
- {
- "category":"FLUX",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/black-forest-labs.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"black-forest-labs_FLUX.1-dev",
- "descr":"FLUX.1 [dev] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions.",
- "detail":"https://openi.pcl.ac.cn/OpenIOSSG/sd-scripts-server/modelmanage/model_readme_tmpl?name=black-forest-labs_FLUX.1-dev"
- }
- ]
- },
- {
- "category":"文心",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/wenxin.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"ERNIE-ViLG 2.0",
- "descr":"全球规模最大的中文跨模态生成模型,可通过自然语言实现图像生成与编辑,由鹏城云脑Ⅱ提供算力支持 。",
- "detail":"",
- "experience":{
- "url":"https://openi.pcl.ac.cn/extension/wenxin",
- "recommend":true
- }
- },
- {
- "name":"鹏城-百度·文心",
- "descr":"全球首个知识增强超大模型,参数规模2600亿,在60多项典型任务中取得了世界领先效果,在各类AI应用场景中均具备极强的泛化能力。",
- "detail":"https://openi.pcl.ac.cn/PCLNLP/ernie3.0_torch"
- },
- {
- "name":"ernie-3.0-xbase-zh",
- "descr":"280M参数重量级通用模型。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/ERNIE-3.0/modelmanage/model_readme_tmpl?name=ernie-3.0-xbase-zh"
- }
- ]
- },
- {
- "category":"CLIP(Contrastive Language-Image Pre-Training)",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/clip.png",
- "style":{
- "height":"16px"
- }
- },
- "models":[
- {
- "name":"Chinese-CLIP ViT-L/14@336px",
- "descr":"CLIP模型的中文版本,使用大规模中文数据进行训练(~2亿图文对),旨在帮助用户快速实现中文领域的图文特征&相似度计算、跨模态检索、零样本图片分类等任务。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Chinese-CLIP/modelmanage/model_readme_tmpl?name=CN-CLIP_ViT-L%2F14%40336px"
- },
- {
- "name":"Chinese-CLIP_ViT-H/14",
- "descr":"CLIP模型的中文版本,使用大规模中文数据进行训练(~3亿图文对),旨在帮助用户快速实现中文领域的图文特征&相似度计算、跨模态检索、零样本图片分类等任务。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Chinese-CLIP/modelmanage/model_readme_tmpl?name=Chinese-CLIP_ViT-H%2F14"
- },
- {
- "name":"Chinese-CLIP_ViT-L/14",
- "descr":"CLIP模型的中文版本,使用大规模中文数据进行训练(~4亿图文对),旨在帮助用户快速实现中文领域的图文特征&相似度计算、跨模态检索、零样本图片分类等任务。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/Chinese-CLIP/modelmanage/model_readme_tmpl?name=Chinese-CLIP_ViT-L%2F14"
- }
- ]
- },
- {
- "category":"悟道",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/wudao.png",
- "style":{
- "border-radius":"50%"
- }
- },
- "models":[
- {
- "name":"CogView",
- "descr":"CogView参数量为40亿,模型可实现文本生成图像,经过微调后可实现国画、油画、水彩画、轮廓画等图像生成。目前在公认MS COCO文生图任务上取得了超过OpenAI DALL·E的成绩,获得世界第一。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/CogView"
- }
- ]
- },
- {
- "category":"封神榜",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/fengshenbang.png"
- },
- "models":[
- {
- "name":"太乙系列",
- "descr":"太乙系列模型主要应用于跨模态场景,包括文本图像生成,蛋白质结构预测, 语音-文本表示等。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/taiyi/modelmanage/show_model"
- }
- ]
- },
- {
- "category":"其它AI模型",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/default.png"
- },
- "models":[
- {
- "name":"Swin Transformer V2",
- "descr":"Swin Transformer achieves strong performance on COCO object detection (58.7 box AP and 51.1 mask AP on test-dev) and ADE20K semantic segmentation (53.5 mIoU on val), surpassing previous models by a large margin.",
- "detail":"https://openi.pcl.ac.cn/BaiJin/Large_Model/modelmanage/model_readme_tmpl?name=Swin%20Transformer%20V2"
- }
- ]
- }
- ]
- },
- {
- "type":"TTS",
- "name":"语音合成",
- "list":[
- {
- "category":"Coqui-AI",
- "icon":{
- "url":"https://openi.pcl.ac.cn/OpenIOSSG/promote/raw/branch/master/model/icons/coquiai.png",
- "style":{
- "height":"20px"
- }
- },
- "models":[
- {
- "name":"XTTS-v2",
- "descr":"XTTS 是一种语音生成模型,让您只需使用 6 秒的快速音频剪辑即可将语音克隆为不同的语言。不需要跨越无数个小时的大量训练数据。",
- "detail":"https://openi.pcl.ac.cn/FoundationModel/coqui/modelmanage/model_readme_tmpl?name=XTTS-v2",
- "experience":{
- "url":"",
- "type":"tts",
- "recommend":true
- }
- }
- ]
- }
- ]
- }
- ]
|