HenryJJ's picture
Update README.md
10ac2c8
metadata
license: apache-2.0
datasets:
  - databricks/databricks-dolly-15k

Instruct_Phi2_Dolly15K

Fine-tuned from phi2,used Dolly15k for the dataset. 90% for training, 10% validation. Trained for 2.0 epochs using QLora. Trained with 1024 context window.

Model Details

  • Trained by: trained by HenryJJ.
  • Model type: Instruct_Phi2_Dolly15K is an auto-regressive language model based on the phi 2 transformer architecture.
  • Language(s): English
  • License for HenryJJ/Instruct_Phi2_Dolly15K: apache-2.0 license

Prompting

Prompt Template With Context

chatml format

<|im_start|>system
{instruction}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Prompt Template Without Context

<|im_start|>system
{instruction}<|im_end|>
<|im_start|>assistant

Training script:

Fully opensourced at: https://github.com/hengjiUSTC/learn-llm/blob/main/trl_finetune.py. Run on 1 A10G instance for 4 hours.

python3 trl_finetune.py --config configs/phi2-dolly.yml