Fix flux tuning device issue (#2352)

Author: mengniwang95

examples/pytorch/diffusion_model/diffusers/flux/README.md

@@ -8,10 +8,10 @@ This example quantizes and validates the accuracy of Flux.
 
 ```shell
 pip install -r requirements.txt
-# Use `INC_PT_ONLY=1 pip install git+https://github.com/intel/neural-compressor.git@v3.6rc` for the latest updates before neural-compressor v3.6 release
-pip install neural-compressor-pt==3.6
-# Use `pip install git+https://github.com/intel/auto-round.git@v0.8.0rc2` for the latest updates before auto-round v0.8.0 release
-pip install auto-round==0.8.0
+# Use `INC_PT_ONLY=1 pip install git+https://github.com/intel/neural-compressor.git@master` for the latest updates before neural-compressor v3.6 release
+pip install neural-compressor-pt==3.7
+# Use `pip install git+https://github.com/intel/auto-round.git@main` for the latest updates before auto-round v0.8.0 release
+pip install auto-round==0.9.3
 ```
 
 ## 2. Prepare Model

examples/pytorch/diffusion_model/diffusers/flux/main.py

@@ -81,8 +81,8 @@ def inference_worker(eval_file, pipe, image_save_dir):
             output.images[idx].save(os.path.join(image_save_dir, str(image_id) + ".png"))
 
 
-def tune():
-    pipe = AutoPipelineForText2Image.from_pretrained(args.model, torch_dtype=torch.bfloat16)
+def tune(device):
+    pipe = AutoPipelineForText2Image.from_pretrained(args.model, torch_dtype=torch.bfloat16).to(device)
     model = pipe.transformer
     layer_config = {}
     kwargs = {}
@@ -116,7 +116,7 @@ def tune():
 
     if args.quantize:
         print(f"Start to quantize {args.model}.")
-        tune()
+        tune(device)
         exit(0)
 
     if args.inference:

examples/pytorch/diffusion_model/diffusers/flux/run_benchmark.sh

@@ -86,7 +86,7 @@ function run_benchmark {
 
 	echo "Start calculating final score..."
 
-    python3 main.py --output_image_path ${output_image_path} --accuracy
+    python3 main.py --output_image_path ${output_image_path} --accuracy --eval_dataset ${dataset_location}
 }
 
 main "$@"