Perplexity 在 Mac 上推出 Hybrid Compute，云端智能体把敏感步骤下发给本地模型并由端侧隐私门控把关

编排的方向是设计决策。计算机在云端启动所有任务，前沿模型负责网页搜索、规划和长期推理。当某个步骤涉及私密文件或敏感数据时，计算机将步骤下达给Mac的本地模型，且不重启任务或丢失上下文，然后将两半合并成一个结果。

这会颠倒本地计算：https：//www.perplexity.ai/hub/blog/the-data-center-moves-to-your-machine 模式 Perplexity 一周前已在 NVIDIA DGX Spark 上发布，该模式从用户硬件开始，经过许可后升级到云端模型。同一个编排器，默认设置相反。

由于电脑与iPhone配合使用，任务可以在Mac桌面执行敏感步骤时远程触发。Perplexity将始终在线的Mac mini定位为专门的本地推理节点，用于处理这种模式。

在受保护文件的任何内容进入云端之前，设备内的分类器会对其进行检查，门会执行四种结果之一：保持本地、隐藏敏感区间、拒绝操作或征求用户同意。凭据、支付卡号和政府身份证会受到最严格的处理。掩盖值在离开时被替换为替代值，云端回复返回时恢复。

PII-Tracer是一款0.6B的双向编码器，基于Qwen3骨干网，用4,096个令牌窗口内的缓冲感知双向注意力取代了因果掩码。线性标记头输出37个标签，包括一个外区标签和针对九种PII类型的BIOS位置标签，辅助头预测对话是否包含敏感内容。训练对约714,000个样本运行了三个历;约束Viterbi解码器在推断时解析标签序列。

配套的基准测试PII-TRACE包含13,148条合成对话，跨13种语言和10种书写系统，提及37,431条字符级标识符。其核心主张是，在长时间对话中找到最多PII，与找到所有PII副本不同。

关于结果：在12个检测器中，PII-Tracer记录了最高的字符F1（0.629），以及第二高的跨度重叠和跨度包含F1，仅次于GPT-5.6-sol。在一致性方面，它领先幅度很大——每个被检测出的重复标识符的提及率为79.4%，跨轮次标识符的提及率为77.6%，而GPT-5.6-sol分别为57.0%和55.1%。在最困难的区间（6–10次提及）中，它的得分为0.691，而GPT-5.6-sol为0.464，GLiNER2-PII为0.073，Claude Opus 4.8为0.045。

非常有趣的是，单窗口召回率从字符数不足1000的对话中的0.975下降到10000字符或以上的对话中的0.687。Perplexity的解决方法是解码，而不是重新训练：50%重叠滑动窗口将整体字符召回率从0.830提升至0.965，多提及一致检测从0.794提升至0.954（基于同一检查点）。

Perplexity的公告列出了三个本地模型：Gemma 4 E4B、Qwen3.6 35B-A3B，以及为计算机后训练的Perplexity模型。产品页面的设置流程指向一次点击即可下载的PPLX Qwen 3.8 27B；Perplexity的Hugging Face组织：https://huggingface.co/perplexity-ai 提供与pplx-computer-qwen-3-8-27b匹配的构建版本，以及 pplx-pii-masking-vllm：https://huggingface.co/perplexity-ai/pplx-pii-masking-vllm，这是0.6B的令牌分类模型，作为门后的核心模型。

对于企业用户，管理员可以设置全组织规则：哪些数据必须保留在设备上、哪些可以进行掩码处理、哪些需要明确批准——同时提供信息离开设备时的审计日志。这正是让其在法律、医疗和金融团队中可用，而不仅仅是有趣的原因。

需要与我们合作以推广您的GitHub仓库或Hugging Face页面或产品发布或网络研讨会等？请联系我们：https://forms.gle/wbash1wF6efRj8G58

Asif Razzaq 是 Marktechpost Media Inc. 的首席执行官。作为一位有远见的企业家和工程师，Asif 致力于利用人工智能的潜力造福社会。他最近的努力是推出人工智能媒体平台 Marktechpost，该平台因其对机器学习和深度学习新闻的深入报道而脱颖而出，既技术上可靠，又易于广大受众理解。该平台每月浏览量超过 200 万次，显示了其在观众中的受欢迎程度。

The direction of orchestration is the design decision. Computer starts every task in the cloud, where frontier models handle web search, planning, and long-horizon reasoning. When a step touches private files or sensitive data, Computer hands that step down to the local model on the Mac without restarting the task or losing context then merges both halves into one result.

This inverts the local compute：https://www.perplexity.ai/hub/blog/the-data-center-moves-to-your-machine mode Perplexity shipped on NVIDIA DGX Spark a week earlier, which starts on the user’s hardware and escalates up to cloud models with permission. Same orchestrator, opposite default.

Because Computer works with iPhone, a task can be triggered remotely while sensitive steps execute on the Mac at the desk. Perplexity positions an always-on Mac mini as a dedicated local inference node for exactly this pattern.

Before anything from a protected file reaches the cloud, an on-device classifier inspects it and the gate applies one of four outcomes: keep it local, mask the sensitive spans, refuse the action, or ask the user for consent. Credentials, payment card numbers, and government IDs get the strictest handling. Masked values are swapped for stand-ins on the way out and restored when the cloud answer returns.

PII-Tracer is a 0.6B bidirectional encoder adapted from a Qwen3 backbone, replacing the causal mask with padding-aware bidirectional attention over a 4,096-token window. A linear tagging head emits 37 labels, one outside-span label plus BIOES position labels for each of nine PII types, and an auxiliary head predicts whether a conversation contains sensitive material. Training ran three epochs on roughly 714,000 samples; a constrained Viterbi decoder resolves the label sequence at inference.

PII-TRACE , the accompanying benchmark, contains 13,148 synthetic conversations across 13 languages and 10 writing systems, with 37,431 character-level identifier mentions. Its central claim is that finding most PII in a long conversation is not the same as finding every copy of it.

On results: across 12 detectors, PII-Tracer records the highest character F1 (0.629) and the second-best span-overlap and span-containment F1, behind GPT-5.6-sol. On consistency it leads by a wide margin — every mention found for 79.4% of recurring identifiers and 77.6% of cross-turn identifiers, versus 57.0% and 55.1% for GPT-5.6-sol. In the hardest bucket (6–10 mentions) it scores 0.691 against 0.464 for GPT-5.6-sol, 0.073 for GLiNER2-PII, and 0.045 for Claude Opus 4.8.

It is very interesting to know that single-window recall drops from 0.975 on conversations under 1,000 characters to 0.687 at 10,000 characters or more. Perplexity’s fix is decoding, not retraining: 50%-overlap sliding windows lift overall character recall from 0.830 to 0.965 and multi-mention consistent detection from 0.794 to 0.954 on the same checkpoint.

Perplexity’s announcement lists three local models at launch: Gemma 4 E4B, Qwen3.6 35B-A3B, and a Perplexity model post-trained for Computer. The product page’s setup flow points to a one-click download of PPLX Qwen 3.8 27B; Perplexity’s Hugging Face org：https://huggingface.co/perplexity-ai carries matching pplx-computer-qwen-3-8-27b builds alongside pplx-pii-masking-vllm ：https://huggingface.co/perplexity-ai/pplx-pii-masking-vllm, the 0.6B token-classification model behind the gate.

For Enterprise, admins can set org-wide rules for what must stay on device, what may be masked, and what requires explicit approval — plus audit logs for when information leaves a machine. That is the piece that makes this usable for legal, healthcare, and financial teams rather than just interesting.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us ：https://forms.gle/wbash1wF6efRj8G58

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.