Views
No views yet
1deepspeed --include localhost:0,1,2,3 sft.py --deepspeed dp_zero3.json \
2 --model_name_or_path="/home/shenyl/cached_models/meta-llama/Llama-2-7b-chat-hf" \
3 --dataset_name="timdettmers/openassistant-guanaco" \
4 --dataset_text_field="text" \
5 --report_to="tensorboard" \
6 --learning_rate=1e-5 \
7 --per_device_train_batch_size=6 \
8 --gradient_accumulation_steps=8 \
9 --output_dir="guanaco_Llama-2-7b-chat-hf" \
10 --logging_steps=1 \
11 --num_train_epochs=15 \
12 --max_steps=-1 \
13 --gradient_checkpointing \
14 --save_steps=0.3timdettmers/openassistant-guanaco