NAS
A research direction that automates the design of neural network architectures. First proposed by Google in 2017, NAS has become a significant branch of deep learning research, aiming to automatically discover optimal neural network structures.
Core Facts
Timeline (last 90 days)
The 35-watt power estimate in the article usually covers only standby or light-load states; when the model is actually performing inference, actual power consumption under high CPU/GPU load may rise to 60–100 watts or even higher, depending on the chip model and model scale.
Calculated at 35 watts of power consumption, the NAS costs less than one yuan a day in electricity.
The Bilibili video demonstrated deploying a local large model on a NAS with two commands, with the API exposed on the local machine's port 11434 and compatible with the OpenAI API.
UPS products in the $200 range usually support NAS auto-shutdown signals, effectively preventing data corruption caused by power outages.
NAS and home server hardware specifications have improved year over year, and multi-core CPUs, large-capacity memory, and even consumer-grade GPUs are no longer uncommon.
Data from a survey paper shows that more than 3,000 new model architectures were proposed in the NAS field over approximately five years.
The Transformer was not discovered through NAS search but is the product of human researchers' design.
Neural Architecture Search (NAS) emerged around 2017, with the core idea of using algorithms to automatically find optimal neural network structures within a predefined search space.
DAS (Direct-Attached Storage) is a multi-bay storage device connected directly to the host via USB or eSATA interfaces, distinct from NAS which is accessed over a network.
The large-scale entry of autonomous aircraft into the national airspace system involves legal issues such as detect-and-avoid (DAA) rules, emergency response protocols for communication link failures, and accident liability attribution.
5 more timeline events
All Facts (17)
Neural Architecture Search (NAS) emerged around 2017, with the core idea of using algorithms to automatically find optimal neural network structures within a predefined search space.
75%UnverifiedA full rebuild of a 6TB drive on a consumer NAS may take 12-24 hours or longer
80%Unverified该自进化架构借鉴了元学习（Meta-Learning）和神经架构搜索（NAS）领域的核心思想
70%UnverifiedNAS (Network Attached Storage) is the most entry-level use case choice in the self-hosted server community
60%UnverifiedThe 35-watt power estimate in the article usually covers only standby or light-load states; when the model is actually performing inference, actual power consumption under high CPU/GPU load may rise to 60–100 watts or even higher, depending on the chip model and model scale.
50%UnverifiedCalculated at 35 watts of power consumption, the NAS costs less than one yuan a day in electricity.
50%UnverifiedUPS products in the $200 range usually support NAS auto-shutdown signals, effectively preventing data corruption caused by power outages.
50%UnverifiedNAS and home server hardware specifications have improved year over year, and multi-core CPUs, large-capacity memory, and even consumer-grade GPUs are no longer uncommon.
50%UnverifiedData from a survey paper shows that more than 3,000 new model architectures were proposed in the NAS field over approximately five years.
50%UnverifiedThe Transformer was not discovered through NAS search but is the product of human researchers' design.
50%UnverifiedDAS (Direct-Attached Storage) is a multi-bay storage device connected directly to the host via USB or eSATA interfaces, distinct from NAS which is accessed over a network.
50%UnverifiedThe large-scale entry of autonomous aircraft into the national airspace system involves legal issues such as detect-and-avoid (DAA) rules, emergency response protocols for communication link failures, and accident liability attribution.
50%UnverifiedA hybrid home data center strategy keeps core data self-hosted on a local NAS while synchronizing critical data to offsite backups on object storage services such as Backblaze B2 or Wasabi.
50%UnverifiedCommon technical approaches to model lightweighting include knowledge distillation, model pruning, quantization, and neural architecture search (NAS)
50%UnverifiedSelf-hosted file conversion services (such as those deployed on resource-limited VPS or home NAS devices) will affect the normal operation of other services on the same machine (file synchronization, databases, reverse proxies, etc.) if resource limits are not applied.
50%UnverifiedNAS是一种连接在局域网上、可供多设备共享访问的专用存储设备，全天候运行并支持远程访问
50%UnverifiedThe Bilibili video demonstrated deploying a local large model on a NAS with two commands, with the API exposed on the local machine's port 11434 and compatible with the OpenAI API.
50%