Meituan has fully open-sourced LongCat-2.0 and the inference code for domestic
accelerator cards. LongCat-2.0 is the first trillion-parameter model to complete
inference on a 50,000-card domestic compute cluster. Meituan says it
co-optimized model architecture, chip adaptation and deployment strategies to
run the trillion-parameter model stably and quickly on memory- and
bandwidth-constrained domestic chips and older cards. The release includes BF16,
FP8 and INT8 multi-precision builds and open-sourced inference optimizations
targeted at domestic compute platforms to enable deployment on existing
hardware.
