Wan2.2
A large AI model whose specific details are not elaborated in the source text; it may be intended for multimodal or domain-specific tasks.
Timeline (last 90 days)
At 64-GPU scale, TIDE reduces step time by up to 27.1% on Llama-3.1 70B, Wan2.2 and Qwen3-30B-A3B (compared with Stock FSDP2).
TIDE's peak memory is up to 4.7 GiB lower than FSDP2+SymmMem, and on Wan2.2 with Slot-Reuse enabled it is even lower than Stock FSDP2.
Users can run Krea-2, HunyuanImage-2.1, and Wan2.2 expert pairs, as well as ComfyUI's NVFP4 and MXFP8 models, within Unsloth.
All Facts (3)
At 64-GPU scale, TIDE reduces step time by up to 27.1% on Llama-3.1 70B, Wan2.2 and Qwen3-30B-A3B (compared with Stock FSDP2).
50%UnverifiedTIDE's peak memory is up to 4.7 GiB lower than FSDP2+SymmMem, and on Wan2.2 with Slot-Reuse enabled it is even lower than Stock FSDP2.
50%UnverifiedUsers can run Krea-2, HunyuanImage-2.1, and Wan2.2 expert pairs, as well as ComfyUI's NVFP4 and MXFP8 models, within Unsloth.
50%