superiort/kullm3_4300QA_10epochs

See axolotl config

axolotl version: 0.4.0

base_model: nlpai-lab/KULLM3
base_model_config: nlpai-lab/KULLM3
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true
hub_model_id: kullm3_finetuning_test_4300QA_10epochs

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: superiort/multiplechoice-4300
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.02
output_dir: ./kullm3_finetuning_test_4300QA_10epochs

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: false

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: axolotl
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 10
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 100
eval_steps: 0.01
save_strategy: epoch
save_steps:
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
    bos_token: "<s>"
    eos_token: "</s>"
    unk_token: "<unk>"
    pad_token: "</s>"  # EOS와 PAD가 동일

kullm3_finetuning_test_4300QA_10epochs

This model is a fine-tuned version of nlpai-lab/KULLM3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4754

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
0.4883	0.01	1	0.3229
0.4139	0.11	14	0.2783
0.3475	0.21	28	0.2473
0.3427	0.32	42	0.2353
0.303	0.43	56	0.2297
0.2902	0.53	70	0.2334
0.288	0.64	84	0.2271
0.2856	0.74	98	0.2233
0.3035	0.85	112	0.2182
0.2829	0.96	126	0.2161
0.2986	1.06	140	0.2219
0.2552	1.17	154	0.2269
0.2489	1.28	168	0.2223
0.2523	1.38	182	0.2248
0.2481	1.49	196	0.2220
0.235	1.59	210	0.2209
0.2661	1.7	224	0.2165
0.2522	1.81	238	0.2231
0.2775	1.91	252	0.2190
0.1825	2.02	266	0.2228
0.1836	2.13	280	0.2331
0.1655	2.23	294	0.2378
0.1604	2.34	308	0.2376
0.1766	2.44	322	0.2356
0.1897	2.55	336	0.2344
0.1756	2.66	350	0.2375
0.1616	2.76	364	0.2387
0.1436	2.87	378	0.2371
0.166	2.98	392	0.2341
0.0828	3.08	406	0.2602
0.0893	3.19	420	0.2747
0.079	3.29	434	0.2760
0.0843	3.4	448	0.2780
0.0815	3.51	462	0.2812
0.0948	3.61	476	0.2828
0.0845	3.72	490	0.2766
0.1025	3.83	504	0.2772
0.0763	3.93	518	0.2813
0.0322	4.04	532	0.3309
0.031	4.14	546	0.3221
0.028	4.25	560	0.3348
0.031	4.36	574	0.3374
0.0309	4.46	588	0.3355
0.0331	4.57	602	0.3344
0.034	4.68	616	0.3384
0.0324	4.78	630	0.3420
0.0301	4.89	644	0.3350
0.0327	4.99	658	0.3387
0.0111	5.1	672	0.4010
0.0089	5.21	686	0.3917
0.0075	5.31	700	0.3925
0.0106	5.42	714	0.3911
0.0091	5.53	728	0.3937
0.0109	5.63	742	0.3985
0.009	5.74	756	0.4044
0.0095	5.84	770	0.3949
0.0075	5.95	784	0.3984
0.0036	6.06	798	0.4133
0.0031	6.16	812	0.4424
0.0026	6.27	826	0.4525
0.0034	6.38	840	0.4519
0.0019	6.48	854	0.4513
0.0018	6.59	868	0.4517
0.0023	6.69	882	0.4520
0.0016	6.8	896	0.4534
0.0018	6.91	910	0.4528
0.001	7.01	924	0.4537
0.0011	7.12	938	0.4581
0.0009	7.23	952	0.4631
0.0009	7.33	966	0.4662
0.0013	7.44	980	0.4680
0.0008	7.54	994	0.4700
0.001	7.65	1008	0.4711
0.0009	7.76	1022	0.4720
0.0011	7.86	1036	0.4727
0.0009	7.97	1050	0.4731
0.0011	8.08	1064	0.4735
0.001	8.18	1078	0.4739
0.001	8.29	1092	0.4741
0.001	8.39	1106	0.4746
0.0011	8.5	1120	0.4744
0.0012	8.61	1134	0.4751
0.0011	8.71	1148	0.4748
0.001	8.82	1162	0.4747
0.0009	8.93	1176	0.4754
0.0011	9.03	1190	0.4752
0.0013	9.14	1204	0.4751
0.0009	9.24	1218	0.4749
0.001	9.35	1232	0.4750
0.0017	9.46	1246	0.4750
0.0012	9.56	1260	0.4749
0.0008	9.67	1274	0.4747
0.0008	9.78	1288	0.4749
0.0011	9.88	1302	0.4754

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.2