frozenleaves/peft
1
0
Fork 0
mirror of https://github.com/huggingface/peft.git synced 2025-10-20 15:33:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

ENH Support XPU for CorDA example (#2687)

Browse Source 
---------

Signed-off-by: Yao, Matrix <matrix.yao@intel.com>
This commit is contained in:
Yao Matrix
2025-08-05 02:41:42 -07:00
committed by GitHub
parent 207b27ec2c
commit daee6367aa
2 changed files with 14 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/corda_finetuning/README.md
View File

@ -150,7 +150,10 @@ corda_config = CordaConfig(
#### Knowledge-preserved adaptation mode
```bash
CUDA_VISIBLE_DEVICES=0 python -u preprocess.py --model_id="meta-llama/Llama-2-7b-hf" \
export CUDA_VISIBLE_DEVICES=0 # force to use device 0 of CUDA GPU
export ZE_AFFINITY_MASK=0 # force to use device 0 of Intel XPU
python -u preprocess.py --model_id="meta-llama/Llama-2-7b-hf" \
--r 128 --seed 233 \
--save_model --save_path {path_to_residual_model} \
--calib_dataset "nqopen"
@ -165,7 +168,10 @@ Arguments:
#### Instruction-previewed adaptation mode
```bash
CUDA_VISIBLE_DEVICES=0 python -u preprocess.py --model_id="meta-llama/Llama-2-7b-hf" \
export CUDA_VISIBLE_DEVICES=0 # force to use device 0 of CUDA GPU
export ZE_AFFINITY_MASK=0 # force to use device 0 of Intel XPU
python -u preprocess.py --model_id="meta-llama/Llama-2-7b-hf" \
--r 128 --seed 233 \
--save_model --save_path {path_to_residual_model} \
--first_eigen --calib_dataset "MetaMATH"
@ -248,4 +254,4 @@ Note that this conversion is not supported if `rslora` is used in combination wi
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024},
}
```
```

7
examples/corda_finetuning/preprocess.py
View File

@ -38,8 +38,11 @@ def main(args):
# Setting random seed of numpy and torch
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
torch.backends.cudnn.deterministic = True
if torch.cuda.is_available():
torch.cuda.manual_seed_all(args.seed)
elif torch.xpu.is_available():
torch.xpu.manual_seed_all(args.seed)
torch.use_deterministic_algorithms(True)
# Load model
model_id = args.model_id