Zhipu AI announced on Wednesday that it trained GLM-Image, a multimodal "autoregressive + diffusion decoder" hybrid model, entirely using Huawei hardware including Ascend Atlas 800T A2 servers and Ascend 910 processors. The open-source model pairs a 9B-parameter autoregressive generator with a 7B-parameter diffusion decoder, but Zhipu has not disclosed training scale, speed, or cost details.
Key Points
- 1Trains GLM-Image hybrid model combining 9B autoregressive and 7B diffusion decoder on Huawei Ascend servers.
- 2Demonstrates feasibility of end-to-end training on domestically produced Chinese hardware, though scale details absent.
- 3Signals potential supply-chain independence that could challenge NVIDIA and AMD dominance in specialized model training.
Scoring Rationale
Notable open-source multimodal model and full-stack domestic training; limited disclosure on scale and performance reduces certainty.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
High-Value Overnight OrdersEasyDelivered International ShipmentsMediumOn-Time Delivery Rate by CarrierHard
250 free problems · No credit card
See all Logistics & Shipping problems
