AMD Unveils AMD-135M: Compact AI Language Model

MetaversePlanet October 2, 2024

0 1 minute read

AMD has announced the AMD-135M, a compact AI language model. Built on Meta’s Llama framework, the AI is specifically designed for private businesses.

In a significant move, the US-based tech giant AMD unveiled the small language model, named AMD-135M, at a recent event. As part of Meta’s “Llama” model, the AMD-135M is expected to be widely adopted by private enterprises. The model comes in two distinct versions: AMD-Llama-135M and AMD-Llama-135M-code. According to AMD, the AMD-Llama-135M was trained on 670 billion public data tokens, utilizing four AMD Instinct MI250s during the training process. The AMD-Llama-135M-code version, on the other hand, incorporates an additional 20 billion tokens specifically tailored for coding tasks.

Can be optimized for specific tasks

AMD’s small language models can be optimized and used for specific tasks. Naturally, the AMD-Llama-135M-code will be primarily used for coding-related functions. According to AMD, the new language model employs predictive decoding technology, enabling the models to operate at high speed.

AMD also stated that the AMD-135M is still in its early stages of development. The company plans to further enhance the small language model in the future, aiming for better results in both performance and speed. It remains to be seen if AMD, which is striving to establish itself in the artificial intelligence sector, will achieve the success it seeks with its small language model.