End of training

README.md

@@ -18,11 +18,11 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [riotu-lab/ArabianGPT-01B](https://huggingface.co/riotu-lab/ArabianGPT-01B) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Bleu: 0.3119
-- Loss: 2.0654
-- Rouge1: 0.5862
-- Rouge2: 0.3489
-- Rougel: 0.5479
+- Loss: 2.1411
+- Bleu: 0.2987
+- Rouge1: 0.5831
+- Rouge2: 0.3405
+- Rougel: 0.5413
 
 ## Model description
 

all_results.json

@@ -5,15 +5,15 @@
     "eval_rouge1": 0.5830931277059221,
     "eval_rouge2": 0.3405124611999597,
     "eval_rougeL": 0.5412563338799627,
-    "eval_runtime": 29.7624,
+    "eval_runtime": 30.3764,
     "eval_samples": 884,
-    "eval_samples_per_second": 29.702,
-    "eval_steps_per_second": 3.73,
+    "eval_samples_per_second": 29.102,
+    "eval_steps_per_second": 3.654,
     "perplexity": 8.50903297033769,
     "total_flos": 2.7664555180032e+16,
-    "train_loss": 0.09870077279897836,
-    "train_runtime": 270.9621,
+    "train_loss": 0.0,
+    "train_runtime": 0.0941,
     "train_samples": 3531,
-    "train_samples_per_second": 260.627,
-    "train_steps_per_second": 32.624
+    "train_samples_per_second": 112613.962,
+    "train_steps_per_second": 14096.678
 }

egy_training_log.txt

@@ -617,3 +617,5 @@ WARNING:__main__:The tokenizer picked seems to have a very large `model_max_leng
 INFO:datasets.arrow_dataset:Loading cached processed dataset at /home/iais_marenpielka/.cache/huggingface/datasets/text/default-93ed01be52df6f6e/0.0.0/96636a050ef51804b84abbfd4f4ad440e01153c24b86293eb5c3b300a41f9101/cache-038f8e8385bf6638.arrow
 INFO:datasets.arrow_dataset:Loading cached processed dataset at /home/iais_marenpielka/.cache/huggingface/datasets/text/default-93ed01be52df6f6e/0.0.0/96636a050ef51804b84abbfd4f4ad440e01153c24b86293eb5c3b300a41f9101/cache-51f1e2b6546273ed.arrow
 WARNING:accelerate.utils.other:Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
+INFO:__main__:*** Evaluate ***
+INFO:absl:Using default tokenizer.

eval_results.json

@@ -5,9 +5,9 @@
     "eval_rouge1": 0.5830931277059221,
     "eval_rouge2": 0.3405124611999597,
     "eval_rougeL": 0.5412563338799627,
-    "eval_runtime": 29.7624,
+    "eval_runtime": 30.3764,
     "eval_samples": 884,
-    "eval_samples_per_second": 29.702,
-    "eval_steps_per_second": 3.73,
+    "eval_samples_per_second": 29.102,
+    "eval_steps_per_second": 3.654,
     "perplexity": 8.50903297033769
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 20.0,
     "total_flos": 2.7664555180032e+16,
-    "train_loss": 0.09870077279897836,
-    "train_runtime": 270.9621,
+    "train_loss": 0.0,
+    "train_runtime": 0.0941,
     "train_samples": 3531,
-    "train_samples_per_second": 260.627,
-    "train_steps_per_second": 32.624
+    "train_samples_per_second": 112613.962,
+    "train_steps_per_second": 14096.678
 }

trainer_state.json

@@ -371,16 +371,16 @@
       "epoch": 20.0,
       "step": 8840,
       "total_flos": 2.7664555180032e+16,
-      "train_loss": 0.09870077279897836,
-      "train_runtime": 270.9621,
-      "train_samples_per_second": 260.627,
-      "train_steps_per_second": 32.624
+      "train_loss": 0.0,
+      "train_runtime": 0.0941,
+      "train_samples_per_second": 112613.962,
+      "train_steps_per_second": 14096.678
     }
   ],
   "logging_steps": 500,
-  "max_steps": 8840,
+  "max_steps": 1326,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 20,
+  "num_train_epochs": 3,
   "save_steps": 500,
   "stateful_callbacks": {
     "EarlyStoppingCallback": {