esunn/1min-scriptgen-luxia-8b

See axolotl config

axolotl version: 0.4.0

base_model: maywell/Llama-3-Ko-Luxia-Instruct
trust_remote_code: true
load_in_8bit: false
load_in_4bit: false
strict: false
datasets:
  - path: "../data/output_fix_real.json"
    type: alpaca
    conversation: chatml
dataset_prepared_path: ../data/1min-luxia-data-pre
val_set_size: 0.1
output_dir: ../data/output/1min-luxia-8b
sequence_len: 1024
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false
wandb_project: 
wandb_entity: 
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 10
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 2e-6
train_on_inputs: false
group_by_length: false
bf16: auto
fp16: null
tf32: false
gradient_checkpointing: true
early_stopping_patience: null
resume_from_checkpoint: null
local_rank: null
logging_steps: 1
xformers_attention: null
flash_attention: true
warmup_steps: 10
evals_per_epoch: 4
eval_table_size: null
eval_max_new_tokens: 128
saves_per_epoch: 1
save_total_limit: 4
debug: true
deepspeed: deepspeed_configs/zero2.json
weight_decay: 0.0
special_tokens:
  pad_token: <|end_of_text|>

data/output/1min-luxia-8b

This model is a fine-tuned version of maywell/Llama-3-Ko-Luxia-Instruct on the modified maywell/ko_youtube_transcription_sample dataset. It achieves the following results on the evaluation set:

• Loss: 2.5280

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 7
• gradient_accumulation_steps: 8
• total_train_batch_size: 56
• total_eval_batch_size: 7
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
2.9998	0.2051	1	3.0382
3.0081	0.4103	2	3.0379
2.9024	0.6154	3	3.0356
2.9814	0.8205	4	3.0280
2.9813	1.0256	5	3.0136
2.9137	1.1795	6	2.9918
2.9909	1.3846	7	2.9426
2.8925	1.5897	8	2.9047
2.825	1.7949	9	2.8790
2.8329	2.0	10	2.7949
2.6496	2.1538	11	2.7632
2.6857	2.3590	12	2.7388
2.679	2.5641	13	2.7193
2.6802	2.7692	14	2.6748
2.6269	2.9744	15	2.6452
2.5546	3.1282	16	2.6286
2.574	3.3333	17	2.6168
2.5548	3.5385	18	2.6054
2.5145	3.7436	19	2.5952
2.452	3.9487	20	2.5863
2.4647	4.1026	21	2.5786
2.423	4.3077	22	2.5715
2.4104	4.5128	23	2.5648
2.3664	4.7179	24	2.5592
2.4211	4.9231	25	2.5536
2.4291	5.0769	26	2.5492
2.3475	5.2821	27	2.5455
2.3665	5.4872	28	2.5417
2.3862	5.6923	29	2.5387
2.3784	5.8974	30	2.5360
2.354	6.0513	31	2.5343
2.3442	6.2564	32	2.5321
2.3499	6.4615	33	2.5312
2.3312	6.6667	34	2.5297
2.3551	6.8718	35	2.5289
2.3363	7.0256	36	2.5289
2.3691	7.2308	37	2.5284
2.3267	7.4359	38	2.5281
2.3389	7.6410	39	2.5281
2.1969	7.8462	40	2.5280

Framework versions

• Transformers 4.40.2
• Pytorch 2.1.2+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1