DeepSeek, the Chinese AI start-up, is facing challenges developing its R2 AI reasoning model. The company has enlisted Huawei engineers to help overcome these challenges. However, delays in the product’s launch are expected.
Delays in DeepSeek’s R2 Model
The DeepSeek R2 model was scheduled to launch by the end of the month, but delays have occurred. The main reason is the difficulty DeepSeek is having while training the AI model with Huawei Ascend chips. Even with assistance from Huawei engineers, the chipsets have not produced the desired results.
DeepSeek’s first AI model, R1, successfully used both Huawei Ascend and Nvidia chips. However, the new R2 model is facing compatibility issues with Ascend chips, causing setbacks in development.
Huawei Ascend Chips vs. Nvidia
The delay comes after DeepSeek chose Huawei Ascend chips over Nvidia’s technology. Local authorities encouraged the use of Ascend chips following the success of the R1 model, despite Nvidia’s dominance in AI hardware.
However, the R2 model has struggled to work effectively with Ascend chips. The issues began in May and remain a major reason for the delay of the AI model.
Turning to Nvidia for Inference
Due to these issues, DeepSeek is considering using Nvidia chips for inference. Inference involves making predictions based on trained data. This shift might allow DeepSeek to repeat the R1 model strategy by combining Huawei Ascend chips for training and Nvidia technology for inference.
Although Huawei engineers are assisting with Ascend chips, DeepSeek might rely on Nvidia for the inference phase. This mixed-chip approach seems to be the best way to overcome the current challenges.
The Future of DeepSeek’s R2 Model
Currently, DeepSeek is not satisfied with the progress of the R2 model. The company’s founder, Liang Wenfeng, has expressed concerns about the model’s performance. He has requested more time to ensure the model remains competitive in the fast-evolving AI landscape.
Despite setbacks, DeepSeek is committed to refining the R2 model. The collaboration with Huawei engineers may still be key to overcoming these technical challenges. With additional time and resources, the R2 model may soon match or exceed previous models.
Conclusion
DeepSeek’s R2 model development faces setbacks due to issues with Huawei Ascend chips. However, Huawei engineers are working with the company to help resolve these problems. The R2 model may soon be ready for launch. While Nvidia chips may be used for inference, DeepSeek hopes that working with Huawei will lead to a successful product.
