fix optimizer + fsdp combination in example (#1893)

Author: winglian

examples/llama-3/fft-8b-liger-fsdp.yaml

@@ -31,7 +31,7 @@ wandb_log_model:
 gradient_accumulation_steps: 4
 micro_batch_size: 2
 num_epochs: 1
-optimizer: paged_adamw_8bit
+optimizer: adamw_torch
 lr_scheduler: cosine
 learning_rate: 2e-5