add benchmark for sft, dpo, simpo, orpo. Add benchmarking result. Support lora with gradient checkpoint

applications/ColossalChat/benchmarks/dummy_dataset.py

@@ -0,0 +1,21 @@
+import torch
+from torch.utils.data import Dataset, DataLoader
+
+class DummyLLMDataset(Dataset):
+    def __init__(self, keys, seq_len, size=500):
+        self.keys = keys
+        self.seq_len = seq_len
+        self.data = self._generate_data()
+        self.size = size
+
+    def _generate_data(self):
+        data = {}
+        for key in self.keys:
+            data[key] = torch.ones(self.seq_len, dtype = torch.long)
+        return data
+
+    def __len__(self):
+        return self.size
+
+    def __getitem__(self, idx):
+        return {key: self.data[key] for key in self.keys}