Is FlashOptim compatible with ZeRO1/ZeRO2/ZeRO3 using deepspeed ？

[nideyongbao]

Thanks for your great work!

I consider how to train in transformers models with deepspeed, code likes this：

accelerator = Accelerator(...)
model = Qwen3_5ForConditionalGeneration.from_pretrained(...)
optimizer = FlashAdamW(model.parameters(), ...)
train_dataloader = build_dataloader(...)
model, optimizer, train_dataloader = accelerator.prepare(model, optimizer, train_dataloader)

for batch in train_dataloader:
    with accelerator.accumulate(model):
        outputs = model(**batch)
        loss = outputs.loss
        accelerator.backward(loss)
        optimizer.step()
        optimizer.zero_grad()