freebearr/tw-judgment-entities-extraction-model

See axolotl config

axolotl version: 0.8.0.dev0

adapter: lora
base_model: 01-ai/Yi-1.5-9B
bf16: auto
dataset_processes: 32

datasets:
- path: freebearr/tw-judgment-entities-extraction-dataset
  type: alpaca

gradient_accumulation_steps: 8
gradient_checkpointing: true
learning_rate: 0.00002
lisa_layers_attribute: model.layers
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: true
lora_alpha: 16
lora_dropout: 0.05
lora_r: 8
lora_target_modules:
- q_proj
- v_pro
loraplus_lr_embedding: 1.0e-06
lr_scheduler: cosine
max_prompt_len: 4096
mean_resizing_embeddings: false
micro_batch_size: 6
num_epochs: 5.0
optimizer: adamw_bnb_8bit
output_dir: ./outputs/mymodel
pretrain_multipack_attn: true
pretrain_multipack_buffer_size: 10000
qlora_sharded_model_loading: false
ray_num_workers: 1
resources_per_worker:
  GPU: 1
sample_packing_bin_size: 200
sample_packing_group_size: 100000
save_only_model: false
save_safetensors: true
sequence_len: 4096
shuffle_merged_datasets: true
skip_prepare_dataset: false
strict: false
train_on_inputs: false
trl:
  log_completions: false
  ref_model_mixup_alpha: 0.9
  ref_model_sync_steps: 64
  sync_ref_model: false
  use_vllm: false
  vllm_device: auto
  vllm_dtype: auto
  vllm_gpu_memory_utilization: 0.9
use_ray: false
val_set_size: 0.1
weight_decay: 0.0

evals_per_epoch: 8
saves_per_epoch: 2

outputs/mymodel

This model is a fine-tuned version of 01-ai/Yi-1.5-9B on the freebearr/tw-judgment-entities-extraction-dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.2812

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-05
• train_batch_size: 6
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 48
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 3
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.4225	0.05	1	0.7396
0.4794	0.15	3	0.7314
0.467	0.3	6	0.7306
0.4193	0.45	9	0.6863
0.5038	0.6	12	0.6593
0.4206	0.75	15	0.5906
0.356	0.9	18	0.5472
0.3543	1.05	21	0.5003
0.3266	1.2	24	0.4621
0.2281	1.35	27	0.4333
0.4226	1.5	30	0.4044
0.2775	1.65	33	0.3886
0.228	1.8	36	0.3748
0.1459	1.95	39	0.3590
0.2063	2.1	42	0.3477
0.2423	2.25	45	0.3412
0.1629	2.4	48	0.3322
0.2195	2.55	51	0.3187
0.2382	2.7	54	0.3159
0.1635	2.85	57	0.3102
0.1404	3.0	60	0.3097
0.115	3.15	63	0.2995
0.1688	3.3	66	0.2971
0.1681	3.45	69	0.3023
0.2178	3.6	72	0.2916
0.1684	3.75	75	0.2925
0.1468	3.9	78	0.2863
0.1565	4.05	81	0.2881
0.1859	4.2	84	0.2834
0.5019	4.35	87	0.2906
0.1854	4.5	90	0.2891
0.1601	4.65	93	0.2800
0.1347	4.8	96	0.2800
0.1398	4.95	99	0.2812

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0