Upload training_config.yml with huggingface_hub

training_config.yml

@@ -24,12 +24,12 @@ checkpointer:
 resume_from_checkpoint: false
 interim_checkpoint_steps: 5000
 interim_gen_steps: null
-max_new_tokens: 180
+max_new_tokens: 200
 temperature: 0.8
 top_k: 200
 dataset:
   _component_: ds.EvenBatcher
-  buffer_size: 64
+  buffer_size: 72
   dataset:
     _component_: ds.RoundRobinDataset
     datasets:
@@ -53,7 +53,7 @@ optimizer:
   lr: 0.0003
 lr_scheduler:
   _component_: torchtune.modules.get_cosine_schedule_with_warmup
-  num_warmup_steps: 100
+  num_warmup_steps: 150
 loss:
   _component_: torch.nn.CrossEntropyLoss
 epochs: 6