refactoring, now running

model/supervised_finetuning/configs/config.yaml

@@ -23,8 +23,10 @@ defaults:
   eval_size:
   log_dir: "base"
   quantization: false
+  seq2seqmodel: false
+  poly_eps: 1.0
 
-galactica-125:
+galactica-125m:
   learning_rate: 5e-5
   model_name: facebook/galactica-125m
   weight_decay: 0.01
@@ -58,8 +60,8 @@ codegen:
 
 debug:
   eval_steps: 20
-  eval_size: 100
+  eval_size: 20
   gradient_accumulation_steps: 1
   per_device_train_batch_size: 1
   per_device_eval_batch_size: 1
-  quantization: false
+  quantization: false