Upload 2 files

config/deepspeed/zero2_config.json

@@ -0,0 +1,15 @@
+{
+    "bf16": {
+      "enabled": "auto"
+    },
+    "zero_optimization": {
+      "stage": 2,
+      "offload_optimizer": {
+          "device": "none"
+      },
+      "contiguous_gradients": true,
+      "overlap_comm": true
+    },
+    "train_micro_batch_size_per_gpu": 1,
+    "gradient_accumulation_steps": "auto"
+}

config/deepspeed/zero3_config.json

@@ -0,0 +1,33 @@
+{
+    "bf16": {
+        "enabled": "auto",
+        "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,
+        "reduce_bucket_size": 16777216, 
+        "stage3_prefetch_bucket_size": 15099494,
+        "stage3_param_persistence_threshold": 40960,
+        "sub_group_size": 1e9,
+        "stage3_max_live_parameters": 1e9,
+        "stage3_max_reuse_distance": 1e9,
+        "stage3_gather_16bit_weights_on_model_save": true
+    },
+    "gradient_accumulation_steps": "auto",
+    "train_micro_batch_size_per_gpu": 1
+}