Add conditional_generation example xpu support by kaixuanliu · Pull Request #2684 · huggingface/peft

kaixuanliu · 2025-07-31T07:13:44Z

No description provided.

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

review-notebook-app · 2025-07-31T07:13:49Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan

Thanks for updating the examples. I found a handful of issues, otherwise LGTM.

BenjaminBossan · 2025-07-31T09:36:42Z

examples/conditional_generation/multitask_prompt_tuning.ipynb

@@ -179,7 +180,979 @@
   "metadata": {
    "tags": []
   },
-   "outputs": [],
+   "outputs": [


Let's clear the cells here. Normally, I prefer to see the outputs but this notebook already has cleared cells and having the given output in this case is not very helpful.

BenjaminBossan · 2025-07-31T09:37:23Z

examples/conditional_generation/peft_adalora_seq2seq.py

    lora_dropout=0.1,
    task_type=TaskType.SEQ_2_SEQ_LM,
    inference_mode=False,
+    total_step=len(dataset['train']) * num_epochs,


Thanks for fixing!

BenjaminBossan · 2025-07-31T09:55:32Z

examples/conditional_generation/peft_lora_seq2seq_accelerate_big_model_inference.ipynb

-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "ValueError",


It looks like this checkpoint has been deleted and I don't even know which model was used or how it was trained. I'd say, let's delete this notebook, as it is non-functional. The same is true for examples/causal_language_modeling/peft_lora_clm_accelerate_big_model_inference.ipynb.

yao-matrix · 2025-07-31T16:33:19Z

examples/conditional_generation/multitask_prompt_tuning.ipynb

    "\n",
    "set_seed(42)\n",
-    "\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


how about we use this? device = torch.accelerator.current_accelerator().type if hasattr(torch, "accelerator") else "cuda"

yao-matrix · 2025-07-31T16:33:37Z

examples/conditional_generation/peft_adalora_seq2seq.py

 os.environ["TOKENIZERS_PARALLELISM"] = "false"

-device = "cuda"
+device = "xpu" if torch.xpu.is_available() else "cuda"


yao-matrix · 2025-07-31T16:34:07Z

examples/conditional_generation/peft_lora_seq2seq.ipynb

    "from datasets import load_dataset\n",
    "\n",
-    "device = \"cuda\"\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


yao-matrix · 2025-07-31T16:35:01Z

examples/conditional_generation/peft_prefix_tuning_seq2seq.ipynb

    "from datasets import load_dataset\n",
    "\n",
-    "device = \"cuda\"\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


yao-matrix · 2025-07-31T16:35:28Z

examples/conditional_generation/peft_prompt_tuning_seq2seq.ipynb

    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\"\n",
    "\n",
-    "device = \"cuda\"\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


yao-matrix · 2025-07-31T16:38:14Z

examples/conditional_generation/peft_prompt_tuning_seq2seq_with_generate.ipynb

    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\"\n",
    "\n",
-    "device = \"cuda\"\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


yao-matrix · 2025-07-31T16:43:09Z

examples/conditional_generation/peft_ia3_seq2seq.ipynb

    "from datasets import load_dataset\n",
    "\n",
-    "device = \"cuda\"\n",
+    "device = \"xpu\" if torch.xpu.is_available() else \"cuda\"\n",


Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan · 2025-08-04T09:26:31Z

@kaixuanliu Could you please run make style?

HuggingFaceDocBuilderDev · 2025-08-04T09:27:25Z

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

kaixuanliu · 2025-08-04T12:54:19Z

@BenjaminBossan Oops, my fault, have fixed the formatting issue.

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan · 2025-08-05T09:25:13Z

Thanks, but there still seem to be issues with formatting. Maybe let's check that you have matching ruff version (0.9.10) and that the settings from the pyproject.toml are being used.

BenjaminBossan

Thanks for making the examples XPU-compatible and also updating them where necessary. The PR LGTM. @yao-matrix anything else from your side?

yao-matrix · 2025-08-05T17:39:42Z

Thanks for making the examples XPU-compatible and also updating them where necessary. The PR LGTM. @yao-matrix anything else from your side?

looks good for me, thx @BenjaminBossan

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

kaixuanliu added 2 commits July 31, 2025 07:01

enable conditional_generation example for XPU

000f631

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

update log

381571f

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

use orig param

006ab11

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan requested changes Jul 31, 2025

View reviewed changes

yao-matrix reviewed Jul 31, 2025

View reviewed changes

kaixuanliu added 2 commits July 31, 2025 19:29

delete file and output from notebook

366ec6d

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

delete extra output

b005da2

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

kaixuanliu requested a review from BenjaminBossan August 1, 2025 08:11

kaixuanliu added 3 commits August 1, 2025 10:13

update device setting

0382dc7

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

replace model.bin with safetensors file

c0156b1

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

adjust

86cab06

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

run make style

c06c312

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan approved these changes Aug 5, 2025

View reviewed changes

BenjaminBossan requested a review from yao-matrix August 5, 2025 10:00

yao-matrix approved these changes Aug 5, 2025

View reviewed changes

fix format issue by upgrading ruff version

ac81931

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>

BenjaminBossan merged commit e3d8fc9 into huggingface:main Aug 6, 2025
2 of 14 checks passed

kaixuanliu deleted the conditional_generation_xpu branch August 11, 2025 01:36

Conversation

kaixuanliu commented Jul 31, 2025

Uh oh!

review-notebook-app bot commented Jul 31, 2025

Uh oh!

BenjaminBossan left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

BenjaminBossan commented Aug 4, 2025

Uh oh!

HuggingFaceDocBuilderDev commented Aug 4, 2025

Uh oh!

kaixuanliu commented Aug 4, 2025

Uh oh!

BenjaminBossan commented Aug 5, 2025

Uh oh!

BenjaminBossan left a comment

Choose a reason for hiding this comment

Uh oh!

yao-matrix commented Aug 5, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants