update examples

Former-commit-id: 3b5f138155d96b346bda18e465cf60ec7d99e19c

examples/extras/galore/llama3_full_sft.yaml

@@ -1,7 +1,7 @@
-# model
+### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 
-# method
+### method
 stage: sft
 do_train: true
 finetuning_type: full
@@ -11,7 +11,7 @@ galore_target: mlp,self_attn
 galore_rank: 128
 galore_scale: 2.0
 
-# dataset
+### dataset
 dataset: identity,alpaca_gpt4_en
 template: llama3
 cutoff_len: 1024
@@ -19,14 +19,14 @@ max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
 
-# output
+### output
 output_dir: saves/llama3-8b/full/sft
 logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 
-# train
+### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 1
 learning_rate: 0.0001
@@ -35,7 +35,7 @@ lr_scheduler_type: cosine
 warmup_steps: 0.1
 pure_bf16: true
 
-# eval
+### eval
 val_size: 0.1
 per_device_eval_batch_size: 1
 evaluation_strategy: steps