renyiyu/llama-2-7b-sft-lora

Model Details

• Supervised fine-tuning (sft) based on meta-llama/Llama-2-7b-hf on yahma/alpaca-cleaned
• Trained with Deepspeed ZeRO-1 + TRL + QLoRA + Flash-Attntion 2 within 1h with 3090x4
• The LoRa adapter is uploaded

Model and Training Details

• Finetuned from model: meta-llama/Llama-2-7b-hf

• Dataset: yahma/alpaca-cleaned

Preprocessing

• preprocessed and packed the sft dataset with trl.trainer.ConstantLengthDataset

Results

Compute Infrastructure

The model is trained using 4 * RTX 3090 - 24GB

Model Card Authors

Yiyu (Michael) Ren

Model Card Contact

Email: renyiyuap@gmail.com

Framework versions

• PEFT 0.8.2