Upload training_args.json with huggingface_hub

training_args.json

@@ -0,0 +1,30 @@
+{
+  "run_name": "kanana-cpt-qlora-sft-lora-concrete-250409",
+  "model_name": "Onuii/DAMI-base-merge-0408",
+  "dataset_name": "Onuii/DAMI-lecture-DB-Dataset-0408-concrete",
+  "max_seq_length": 8192,
+  "dtype": "bfloat16",
+  "load_in_4bit": true,
+  "r": 16,
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "use_rslora": true,
+  "batch_size": 16,
+  "gradient_accumulation_steps": 2,
+  "learning_rate": 1e-05,
+  "warmup_ratio": 0.05,
+  "num_train_epochs": 4,
+  "optimizer": "adamw_8bit",
+  "weight_decay": 0.01,
+  "lr_scheduler_type": "linear",
+  "output_dir": "../outputs/kanana-cpt-qlora-sft-lora-concrete-250409",
+  "logging_dir": "../logs",
+  "logging_steps": 1,
+  "save_steps": 200,
+  "save_total_limit": 2,
+  "report_to": "wandb",
+  "use_gradient_checkpointing": true,
+  "save_strategy": "steps",
+  "random_seed": 562,
+  "hyperparameter_dir": "../hyperparameters"
+}