Training in progress, epoch 1

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bbad7a51936e9d60cc0f7618e20fe0a56e387481ec70f21cb5b60f23b6a7a931
+oid sha256:27e2ac7cb643fd30f84baae2da820e5bb4bb007195fa5e93edce5a2ea4aac2c7
 size 6651

upload.py

@@ -1,7 +1,7 @@
 from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
-ckpt = 'checkpoint-70'
+ckpt = 'checkpoint-36'
 origin_model_id = 'beomi/Llama-3-Open-Ko-8B'
 model = AutoModelForCausalLM.from_pretrained(ckpt)
 tokenizer = AutoTokenizer.from_pretrained(origin_model_id)
@@ -12,5 +12,5 @@ model.resize_token_embeddings(len(tokenizer))
 # state_dict = get_fp32_state_dict_from_zero_checkpoint(ckpt)
 # model.load_state_dict(state_dict)
 
-tokenizer.push_to_hub('ingeol/kosaul_ft_v0.1')
-model.push_to_hub('ingeol/kosaul_ft_v0.1')
+tokenizer.push_to_hub('ingeol/kosaul_ft_v0.3')
+model.push_to_hub('ingeol/kosaul_ft_v0.3')