Add LoRA training config

skythought/train/LLaMA-Factory/data/dataset_info.json

@@ -656,5 +656,15 @@
       "chosen": "chosen",
       "rejected": "rejected"
     }
-  }
+  }, 
+  "Sky-T1-Lora": {
+    "file_name": "your data path here",
+    "ranking": true,
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected"
+    }
+  },
 }

skythought/train/LLaMA-Factory/examples/train_lora/qwen2_lora_config.yaml

@@ -0,0 +1,37 @@
+### model
+model_name_or_path: Qwen/Qwen2.5-32B-Instruct
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: all
+lora_rank: 64
+deepspeed: examples/deepspeed/ds_z3_offload_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
+
+### dataset
+dataset: Sky-T1-Lora
+template: qwen25
+cutoff_len: 16384
+max_samples: 1000000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+### output
+output_dir: saves/Sky-T1-Lora/full/original
+logging_steps: 1
+save_steps: 100
+plot_loss: true
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 12
+learning_rate: 1.0e-4
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+report_to: wandb
+run_name: Sky-T1-Lora