Update README.md

README.md

@@ -137,18 +137,18 @@ A partir desses valores é calculada a Precisão, o Recall e F1-score do modelo.
 
 ### Domain
 
--  Restaurant:     82.418006
-- Laptop:         75.098765
-- book:           40.076923
-- home:           39.204082
-- pet:            35.227273
-- beauty:         33.675676
-- grocery:        33.384615
-- electronics:    32.947368
-- fashion:        32.480000
-- toy:            30.520000
-
-### Training Details:
+-  Restaurant:     82.42
+- Laptop:         75.10
+- book:           40.08
+- home:           39.21
+- pet:            35.23
+- beauty:         33.68
+- grocery:        33.39
+- electronics:    32.95
+- fashion:        32.49
+- toy:            30.52
+
+### Training Details
 
 - Model: "google/flan-t5-base"
 - Tokenizer: T5Tokenizer / T5ForConditionalGeneration
@@ -159,6 +159,61 @@ A partir desses valores é calculada a Precisão, o Recall e F1-score do modelo.
 - Max_length: 512
 - Num_Epochs: 10
 
+### Global Parameters
+
+L_RATE = 3e-4
+
+BATCH_SIZE = batch_size
+
+PER_DEVICE_EVAL_BATCH = batch_size
+
+WEIGHT_DECAY = 0.01
+
+SAVE_TOTAL_LIM = 1
+
+NUM_EPOCHS = num_epochs
+
+MAX_NEW_TOKENS = max_length
+
+### Set up training arguments
+
+training_args = Seq2SeqTrainingArguments(
+
+   output_dir="./results/absa_domain_parafrase_v512_2",
+   
+   evaluation_strategy="epoch",
+   
+   save_strategy="epoch",
+   
+   learning_rate=L_RATE,
+   
+   per_device_train_batch_size=BATCH_SIZE,
+   
+   per_device_eval_batch_size=PER_DEVICE_EVAL_BATCH,
+   
+   gradient_accumulation_steps=8,
+   
+   weight_decay=WEIGHT_DECAY,
+   
+   save_total_limit=SAVE_TOTAL_LIM,
+   
+   load_best_model_at_end=True,
+   
+   num_train_epochs=NUM_EPOCHS,
+   
+   predict_with_generate=True,
+   
+   logging_dir="./logs",
+   
+   logging_strategy="epoch",
+   
+   logging_steps=100,
+   
+   report_to = 'wandb',
+   
+   push_to_hub=True
+   
+)
 ### Lora config:
 
 - lora_config =