DandinPower's picture
End of training
2e18deb verified
metadata
language:
  - zh
license: apache-2.0
library_name: peft
tags:
  - trl
  - sft
  - nycu-112-2-deeplearning-hw2
  - generated_from_trainer
base_model: yentinglin/Taiwan-LLM-7B-v2.1-chat
datasets:
  - DandinPower/ZH-Reading-Comprehension-TaiwanLlama-chat
model-index:
  - name: taiwanllm_7b_lora_completion_only
    results: []

taiwanllm_7b_lora_completion_only

This model is a fine-tuned version of yentinglin/Taiwan-LLM-7B-v2.1-chat on the DandinPower/ZH-Reading-Comprehension-TaiwanLlama-chat dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0867

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: 0.0001
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 2
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 16
  • total_eval_batch_size: 2
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 700
  • num_epochs: 3.0

Training results

Training Loss Epoch Step Validation Loss
0.1316 0.3690 250 0.1124
0.0988 0.7380 500 0.1145
0.0827 1.1070 750 0.0971
0.0744 1.4760 1000 0.0897
0.0482 1.8450 1250 0.0893
0.0337 2.2140 1500 0.0924
0.0229 2.5830 1750 0.0911
0.0385 2.9520 2000 0.0867

Framework versions

  • PEFT 0.10.0
  • Transformers 4.40.0
  • Pytorch 2.2.2+cu121
  • Datasets 2.19.0
  • Tokenizers 0.19.1