Quick take: DeepSeek struggles to train models on Huawei
Subscribe to keep reading.
Sign up to stay briefed on what the Chinese government is doing, thinking, and saying.
Already a subscriber? Log in.
What happened: On August 14, the FT reported that the launch of DeepSeek's R2 model has been delayed due to difficulties getting Huawei’s Ascend AI chips working for model training.
The company has switched back to Nvidia for training, but will keep using Ascend for inference.
ICYDK: Training fron...