devotd.
Recent developments, checked against their sources.
Perplexity changed its request format and training data to assess multiple questions about the same context at once. Grouping related attributes helps, but independent candidates fare better separately. Long inputs and execution-path differences need checking.
NVIDIA introduced roughly 2.8 trillion tokens of translation-based training data for 22 Indian languages and an open generation process. The planned release supplements locally authored text. Reported gains come from a training stage using both types of data.
Reflection released its first open-weight model in three precisions, enabling self-hosting and further training. Deployment control leaves operations teams responsible for checking hardware needs, task quality, and safety.
IBM’s new open-weight model offers chat, choice-based decisions and speech recognition through one endpoint. It can simplify server setups, but response-specific handling and English-only speech input require consideration.
DeepSeek released a model it says improves coding scores while reducing task time and output length. Web search and page fetching are now public APIs, with temporarily free server-side calls through designated tools.
Support for a new model's text, image, and video inputs has been added. The cache manager's Python implementation is gone, so existing settings and event collection paths need checking before an upgrade.
Transformers adds support for Meta's multimodal model and native loading for InclusionAI's model. Input handling and loading are unified. Deployments still need separate checks for large-model memory needs and expert-parallel inference slowdowns.
NVIDIA released weights for training and evaluation, plus memory-saving deployment weights, for a model handling long documents, images, and video. Built-in prediction can accelerate generation, but actual speed and quantization's quality impact must be assessed by deployment conditions and task.
A small language model stores weight signs in 1 bit, with a 292.1MB distribution file that also runs on CPUs. Its average across ten company-evaluated tasks fell by 8.28 points. Runtime memory and generation speed were not measured.
Support for distributing model expert computations across GPUs has expanded on Rubin and Blackwell workstations. The workstation path leaves input delivery and output summation to the caller. Published single-layer performance figures do not represent speedups for an entire distributed service.