update configs

Author: Nghi Bui

codetf/configs/inference/causal_lm.yaml

@@ -7,8 +7,8 @@ causallm-codegen-350M-multi-pretrained:
     tokenizer_url: "Salesforce/codegen-350M-multi"
     max_prediction_length: 512
 causallm-codegen-350M-nl-pretrained:
-    huggingface_url: "Salesforce/codegen-350-nl"
-    tokenizer_url: "Salesforce/codegen-350-nl"
+    huggingface_url: "Salesforce/codegen-350M-nl"
+    tokenizer_url: "Salesforce/codegen-350M-nl"
     max_prediction_length: 512
 causallm-codegen-2B-mono-pretrained:
     huggingface_url: "Salesforce/codegen-2B-mono"

codetf/configs/inference/codet5.yaml

@@ -140,20 +140,20 @@ codet5-plus-770M-python-pretrained:
     max_prediction_length: 512
     beam_size: 5
     trust_remote_code: False
-    device_map: True
+    device_map: False
 codet5-plus-770M-pretrained:
     huggingface_url: "Salesforce/codet5p-770m"
     tokenizer_url: "Salesforce/codet5p-770m"
     max_source_length: 512
     max_prediction_length: 512
     beam_size: 5
     trust_remote_code: False
-    device_map: True
+    device_map: False
 codet5-plus-220M-pretrained:
     huggingface_url: "Salesforce/codet5p-220m"
     tokenizer_url: "Salesforce/codet5p-220m"
     max_source_length: 512
     max_prediction_length: 512
     beam_size: 5
     trust_remote_code: False
-    device_map: True
+    device_map: False

codetf/configs/training/causal_lm.yaml

@@ -6,9 +6,6 @@ hyperparameters:
   num_train_epochs: 10
   auto_find_batch_size: True
   batch_size: 4
-  max_steps: 1000
-  eval_steps: 100
-  save_steps: 1000
   logging_steps: 100
   per_device_train_batch_size: 8
   per_device_eval_batch_size: 8
@@ -30,3 +27,4 @@ hyperparameters:
   beam_size: 5
   max_grad_norm: 5.0
   adam_epsilon : 1e-06
+  load_best_model_at_end: True

codetf/configs/training/codet5.yaml

@@ -6,9 +6,6 @@ hyperparameters:
   num_train_epochs: 1
   auto_find_batch_size: True
   batch_size: 4
-  max_steps: 1000
-  eval_steps: 100
-  save_steps: 1000
   logging_steps: 100
   per_device_train_batch_size: 8
   per_device_eval_batch_size: 8
@@ -30,6 +27,7 @@ hyperparameters:
   beam_size: 5
   max_grad_norm: 5.0
   adam_epsilon : 1e-06
+  load_best_model_at_end: True
 lora:
   r: 8
   lora_alpha: 32