Research paper introducing Program-as-Weights, a novel programming paradigm treating weights as executable programs for fuzzy functions. Achieved 91 upvotes on HuggingFace.
New model GigaChat3.5-432B with 28B active parameters released with immediate GGUF quantization support. Enables local deployment from launch.
Tencent released Hy3, a new open model with 295B total parameters and 21B active parameters under Apache 2.0 license. Addresses efficiency in large model deployment.