Amazon has taken a significant step in the field of artificial intelligence by announcing that its Amazon Bedrock platform now allows for the importing of custom models. This new feature includes compatibility with OpenAI models that have open weights, such as the GPT-OSS variants with 20 and 120 billion parameters, which focus on advanced reasoning capabilities. This functionality makes it easier for organizations to migrate their existing applications to Amazon Web Services (AWS) infrastructure, offering benefits such as enterprise-level security, scalability, and cost control.
The option to import custom models allows developers to integrate these models into a serverless environment. This simplifies management by providing a single API, eliminating the need to manage multiple endpoints or separate infrastructures. To use this new functionality, users must upload their model files to Amazon S3 and then initiate the import process through the Amazon Bedrock console. AWS takes care of technical aspects such as provisioning GPU and setting up inference servers, allowing developers to focus on improving their applications.
The GPT-OSS models are the first from OpenAI with open weights since the release of GPT-2, allowing for download, modification, and application at no additional cost. This approach is especially valuable in a commercial context. The GPT-OSS-20B model stands out for its speed and efficiency, ideal for applications that require agile performance, while the GPT-OSS-120B is more robust and designed for complex reasoning tasks such as coding and tool usage.
Both models use a mixture of experts architecture, activating only the most relevant experts for each request, which not only improves performance but also helps control computational costs. Importing these models into Amazon Bedrock follows a defined process that includes file preparation, uploading to S3, and invocation through API calls compatible with OpenAI.
Furthermore, the migration from OpenAI to Amazon Bedrock offers advantages such as continuity in request structure and improved data privacy. Companies have the ability to customize and adjust the models to their specific needs, improving performance and maintaining total control over their artificial intelligence tools. This is especially relevant in a context where the proliferation of AI-driven applications is becoming increasingly noticeable. The Amazon Bedrock Custom Model Import functionality is already available in several regions, with plans for future expansion, offering organizations an efficient and secure path to develop their technological capabilities in the field of artificial intelligence.
via: MiMub in Spanish
