fact/README.md

metadata

license: other
library_name: peft
tags:
  - llama-factory
  - lora
  - generated_from_trainer
base_model: /workspace/xll/checkpoints/lmsys/vicuna-7b-v1.5
model-index:
  - name: fact
    results: []

fact

This model is a fine-tuned version of /workspace/xll/checkpoints/lmsys/vicuna-7b-v1.5 on the vicuna_fact_test dataset. It achieves the following results on the evaluation set:

• Loss: 0.0626

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: float16

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• training_steps: 4430
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.0941	0.02	100	0.0995
0.0903	0.04	200	0.0902
0.089	0.06	300	0.0802
0.0619	0.08	400	0.0762
0.0722	0.1	500	0.0752
0.0724	0.12	600	0.0736
0.0762	0.14	700	0.0710
0.0567	0.16	800	0.0712
0.0578	0.18	900	0.0679
0.0597	0.2	1000	0.0691
0.0538	0.22	1100	0.0660
0.0498	0.24	1200	0.0658
0.0595	0.26	1300	0.0642
0.0465	0.28	1400	0.0677
0.0533	0.3	1500	0.0651
0.0593	0.32	1600	0.0641
0.055	0.34	1700	0.0653
0.0546	0.36	1800	0.0634
0.0524	0.38	1900	0.0615
0.0432	0.4	2000	0.0632
0.0631	0.42	2100	0.0619
0.0519	0.44	2200	0.0598
0.0397	0.46	2300	0.0607
0.0467	0.48	2400	0.0616
0.049	0.5	2500	0.0636
0.0488	0.52	2600	0.0604
0.0449	0.54	2700	0.0598
0.0438	0.56	2800	0.0610
0.036	0.58	2900	0.0633
0.0464	0.6	3000	0.0603
0.0437	0.62	3100	0.0612
0.0389	0.64	3200	0.0605
0.0377	0.66	3300	0.0632
0.0382	0.68	3400	0.0626
0.051	0.7	3500	0.0612
0.047	0.72	3600	0.0633
0.0451	0.74	3700	0.0608
0.0472	0.77	3800	0.0630
0.0394	0.79	3900	0.0612
0.0471	0.81	4000	0.0614
0.031	0.83	4100	0.0602
0.0405	0.85	4200	0.0584
0.0421	0.87	4300	0.0591
0.0454	0.89	4400	0.0593

Framework versions

• PEFT 0.7.0
• Transformers 4.37.1
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1