Luigi
/

SmolLM2-360M-Instruct-TaiwanChat

Generated from Trainer

Model card Files Files and versions

Luigi commited on May 1

Commit

976c215

·

1 Parent(s): 38e2b45

adjust hyper-parameters

Files changed (1) hide show

train_with_unsloth.py +2 -2

train_with_unsloth.py CHANGED Viewed

@@ -92,8 +92,8 @@ class LoggingSFTTrainer(SFTTrainer):
 PROJECT_NAME = 'SmolLM2-360M-Instruct-TaiwanChat'
 BASE_MODEL_ID = "unsloth/SmolLM2-360M-Instruct"
 DATASET_ID = "yentinglin/TaiwanChat"
-N_SAMPLES = 800000
-MAX_LEN = 2048
 # CUDA and W&B setup
 os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True,max_split_size_mb:128"

 PROJECT_NAME = 'SmolLM2-360M-Instruct-TaiwanChat'
 BASE_MODEL_ID = "unsloth/SmolLM2-360M-Instruct"
 DATASET_ID = "yentinglin/TaiwanChat"
+N_SAMPLES = 600000
+MAX_LEN = 512
 # CUDA and W&B setup
 os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True,max_split_size_mb:128"