update

examples/tutorials/dpo/ultrachat/ds_config/{deepspeed_stage_3_config.json → deepspeed_stage_2_config.json}

@@ -1,46 +1,35 @@
 {
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "steps_per_print": 200,
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false,
-  "optimizer": {
-    "type": "Adam",
-    "params": {
-      "lr": "auto",
-      "betas": "auto",
-      "eps": "auto",
-      "weight_decay": "auto"
-    }
+  "gradient_accumulation_steps": "auto",
+  "gradient_clipping": 1.0,
+  "zero_optimization": {
+    "stage": 2,
+    "offload_optimizer": {
+      "device": "cpu",
+      "pin_memory": true
+    },
+    "contiguous_gradients": true,
+    "overlap_comm": true,
+    "allgather_bucket_size": 5e8,
+    "reduce_bucket_size": 5e8
   },
   "fp16": {
-    "enabled": "auto",
+    "enabled": true,
     "loss_scale": 0,
     "loss_scale_window": 1000,
     "initial_scale_power": 16,
     "hysteresis": 2,
     "min_loss_scale": 1
   },
-  "zero_optimization": {
-    "stage": 3,
-    "offload_optimizer": {
-      "device": "cpu",
-      "pin_memory": true
-    },
-    "offload_param": {
-      "device": "cpu",
-      "pin_memory": true
-    },
-    "overlap_comm": true,
-    "contiguous_gradients": true,
-    "sub_group_size": 1e9,
-    "reduce_bucket_size": "auto",
-    "stage3_prefetch_bucket_size": "auto",
-    "stage3_param_persistence_threshold": "auto",
-    "stage3_max_live_parameters": 1e9,
-    "stage3_max_reuse_distance": 1e9,
-    "stage3_gather_16bit_weights_on_model_save": true
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": 1e-8,
+      "weight_decay": "auto"
+    }
   },
   "scheduler": {
     "type": "WarmupLR",
@@ -50,9 +39,6 @@
       "warmup_num_steps": "auto"
     }
   },
-  "activation_checkpointing": {
-    "enabled": true,
-    "partition_activations": true,
-    "contiguous_memory_optimization": true
-  }
-}
+  "steps_per_print": 100,
+  "wall_clock_breakdown": false
+}

examples/tutorials/dpo/ultrachat/step_2_train_sft_model_deepspeed.py

@@ -216,7 +216,7 @@ def main():
     sft_config = SFTConfig(
         output_dir=args.output_model_dir,
         dataset_text_field="formated_text",
-        deepspeed=None if debug_mode else "./ds_config/deepspeed_stage_3_config.json",
+        deepspeed=None if debug_mode else "./ds_config/deepspeed_stage_2_config.json",
         num_train_epochs=1,
         per_device_train_batch_size=1 if debug_mode else 2,
         gradient_accumulation_steps=1 if debug_mode else 8,