cpu offloading

deepspeed/zero3_cpu.json

@@ -0,0 +1,42 @@
+{
+  "zero_optimization": {
+    "stage": 3,
+    "offload_optimizer": {
+      "device": "cpu"
+    },
+    "overlap_comm": true,
+    "contiguous_gradients": true,
+    "sub_group_size": 0,
+    "reduce_bucket_size": "auto",
+    "stage3_prefetch_bucket_size": "auto",
+    "stage3_param_persistence_threshold": "auto",
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
+  },
+  "bf16": {
+    "enabled": "auto"
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
+}

examples/yayi2-30b/fft.yml

@@ -54,7 +54,7 @@ eval_table_size:
 eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
-deepspeed: deepspeed/zero3.json
+deepspeed: deepspeed/zero3_cpu.json
 weight_decay: 0.0
 fsdp:
 fsdp_config: