End of training

final_metrics.json

@@ -1,11 +1,8 @@
 {
-    "eval_loss": 22.220468521118164,
-    "eval_accuracy": 0.0,
-    "eval_rouge1": 0.0,
-    "eval_rouge2": 0.0,
-    "eval_rougeL": 0.0,
-    "eval_runtime": 20.7252,
-    "eval_samples_per_second": 4.825,
-    "eval_steps_per_second": 1.206,
-    "epoch": 2.986666666666667
+    "eval_loss": 23.53748321533203,
+    "eval_model_preparation_time": 0.004,
+    "eval_runtime": 4.9923,
+    "eval_samples_per_second": 10.015,
+    "eval_steps_per_second": 2.604,
+    "epoch": 2.9734513274336285
 }

training_args.json

@@ -6,11 +6,11 @@
     "do_predict": false,
     "eval_strategy": "steps",
     "prediction_loss_only": false,
-    "per_device_train_batch_size": 2,
-    "per_device_eval_batch_size": 2,
+    "per_device_train_batch_size": 4,
+    "per_device_eval_batch_size": 4,
     "per_gpu_train_batch_size": null,
     "per_gpu_eval_batch_size": null,
-    "gradient_accumulation_steps": 32,
+    "gradient_accumulation_steps": 16,
     "eval_accumulation_steps": null,
     "eval_delay": 0,
     "torch_empty_cache_steps": null,
@@ -29,13 +29,13 @@
     "log_level": "passive",
     "log_level_replica": "warning",
     "log_on_each_node": true,
-    "logging_dir": "logs_sft_8",
+    "logging_dir": "./logs",
     "logging_strategy": "steps",
     "logging_first_step": false,
-    "logging_steps": 10,
+    "logging_steps": 100,
     "logging_nan_inf_filter": true,
     "save_strategy": "steps",
-    "save_steps": 10,
+    "save_steps": 500,
     "save_total_limit": 3,
     "save_safetensors": true,
     "save_on_each_node": false,
@@ -61,7 +61,7 @@
     "tpu_metrics_debug": false,
     "debug": [],
     "dataloader_drop_last": false,
-    "eval_steps": 10,
+    "eval_steps": 100,
     "dataloader_num_workers": 0,
     "dataloader_prefetch_factor": null,
     "past_index": -1,
@@ -112,7 +112,7 @@
     "hub_token": "<HUB_TOKEN>",
     "hub_private_repo": false,
     "hub_always_push": false,
-    "gradient_checkpointing": true,
+    "gradient_checkpointing": false,
     "gradient_checkpointing_kwargs": null,
     "include_inputs_for_metrics": false,
     "include_for_metrics": [],