|
Work item:
|
Q.CPN-BNG-DLLMT
|
|
Subject/title:
|
Signalling requirements of border network gateway for distributed large language model training in computing power network
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-09 (Medium priority)
|
|
Liaison:
|
ITU-T SG13, ITU-T SG2
|
|
Supporting members:
|
China Telecom; China Unicom; University of Science and Technology Beijing
|
|
Summary:
|
With the rapid development of Large Language Models (LLMs), single-site capacity is hitting the limits in terms of training LLM with extremely large parameter scale. As a result, distributed training across multiple data centers becomes an alternative technical scheme. This document aims to study the architecture and signalling requirements of border network gateway for distributed LLM training in computing power network to meet increasing bandwidth and latency demands, brings the flexibility and agility deployment to the network for distributed LLM training.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-03-19 13:55:46
|
|
Last update:
2026-03-19 14:04:52
|
