AMD is on the scene with a new AI model that is sure to make waves. The Llama-135m small language model is AMD’s first self-developed AI model. The model, which attracts attention especially with its “predictive decoding” capability, will significantly reduce the use of RAM in AI operations. In other words, it becomes possible to achieve more efficient results using fewer resources.

AMD enters the world of artificialAI with the Llama-135m model

AMD Llama-135m runs on 6700 billion tokens and is available to developers under the Apache 2.0 open source license. So what does the predictive decoding feature of this model do? In a nutshell, after the model generates a set of tokens, these tokens are verified by a large target model.

Thus, multiple tokens are generated and validated in each process, resulting in faster and more efficient results without any performance loss. The new AI model was trained on the AMD Instinct MI250 for six days.

In addition, a coding-optimized variant, AMD-Llama-135m-code, has also been optimized at the micro level with four more days of training. AMD’s new model will create a huge opportunity, especially for software developers and AI projects.

This step is an important demonstration of AMD’s growing strength in AI. The Llama-135m will heat up the competition by offering faster, more efficient and resource-optimized solutions for AI projects.

What do you think about this? Let us know in the comments section below.