Update --int8 flag to --optimized flag

.azure-pipelines/model-test-3x.yml

@@ -33,7 +33,6 @@ parameters:
     default:
       - opt_125m_woq_gptq_int4
       - opt_125m_woq_gptq_nf4_dq_bnb
-      - opt_125m_woq_gptq_int4_dq_ggml
 
 stages:
   - stage: PyTorchModels

examples/3.x_api/pytorch/cv/fp8_quant/main.py

@@ -60,7 +60,7 @@
 parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
                     help='evaluate model on validation set')
 parser.add_argument('-t', '--tune', dest='tune', action='store_true',
-                    help='tune best int8 model on calibration dataset')
+                    help='tune best fp8 model on calibration dataset')
 parser.add_argument('--pretrained', dest='pretrained', action='store_true',
                     help='use pre-trained model')
 parser.add_argument('--world-size', default=-1, type=int,
@@ -94,7 +94,7 @@
                     help='For accuracy measurement only.')
 parser.add_argument("--tuned_checkpoint", default='./saved_results', type=str, metavar='PATH',
                     help='path to checkpoint tuned by Neural Compressor (default: ./)')
-parser.add_argument('--int8', dest='int8', action='store_true',
+parser.add_argument('--optimized', dest='optimized', action='store_true',
                     help='run benchmark')
 parser.add_argument('--device', default='hpu', type=str,
                     help='use hpu device for fp8 quantization')
@@ -205,7 +205,7 @@ def eval_func(model):
 
     if args.performance or args.accuracy:
         model.eval()
-        if args.int8:
+        if args.optimized:
             from neural_compressor.utils.pytorch import load
             new_model = load(os.path.abspath(os.path.expanduser(args.tuned_checkpoint)),
                              model,

examples/3.x_api/pytorch/cv/static_quant/main.py

@@ -86,8 +86,8 @@
                     help='For accuracy measurement only.')
 parser.add_argument("--tuned_checkpoint", default='./saved_results', type=str, metavar='PATH',
                     help='path to checkpoint tuned by Neural Compressor (default: ./)')
-parser.add_argument('--int8', dest='int8', action='store_true',
-                    help='Load int8 model.')
+parser.add_argument('--optimized', dest='optimized', action='store_true',
+                    help='Load optimized model.')
 parser.add_argument("--calib_iters", default=128, type=int,
                     help="For calibration only.")
 parser.add_argument("--iters", default=100, type=int,
@@ -222,7 +222,7 @@ def eval_func(model):
         return
 
     if args.performance or args.accuracy:
-        if args.int8:
+        if args.optimized:
             from neural_compressor.torch.quantization import load
             q_model = load(args.tuned_checkpoint)
 

examples/3.x_api/pytorch/cv/static_quant/run_benchmark.sh

@@ -35,8 +35,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -63,8 +63,8 @@ function run_benchmark {
         echo "Error: No such mode: ${mode}"
         exit 1
     fi
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
     echo $extra_cmd
 

examples/3.x_api/pytorch/diffusion_model/diffusers/stable_diffusion/smooth_quant/README.md

@@ -24,7 +24,7 @@ sh run_quant.sh --alpha=0.44
 ```
 To load a quantized model:
 ```bash
-python sdxl_smooth_quant.py --model_name_or_path stabilityai/stable-diffusion-xl-base-1.0 --quantize --load --int8
+python sdxl_smooth_quant.py --model_name_or_path stabilityai/stable-diffusion-xl-base-1.0 --quantize --load --optimized
 ```
 or
 ```bash
@@ -79,5 +79,5 @@ python clip/clip_score.py \
 ```
 Or you can use the bash script for all steps above:
 ```bash
-sh run_benchmark.sh --mode=accuracy --int8=true
+sh run_benchmark.sh --mode=accuracy --optimized=true
 ```

examples/3.x_api/pytorch/diffusion_model/diffusers/stable_diffusion/smooth_quant/main.py

@@ -325,7 +325,7 @@ def __call__(
 parser.add_argument('--precision', default='fp32', type=str)
 parser.add_argument('--base-output-dir', default="./output", type=str)
 parser.add_argument('--quantized-unet', default="./saved_results", type=str)
-parser.add_argument("--int8", action="store_true", help="Load quantized model.")
+parser.add_argument("--optimized", action="store_true", help="Load quantized model.")
 parser.add_argument("--load", action="store_true")
 parser.add_argument('--iters', default=5000, type=int, help="Num of image generated.")
 parser.add_argument('--output-dir-name', default=None, type=str)
@@ -411,7 +411,7 @@ def __call__(
                                                                     variant="fp16" if args.precision == 'fp16' else None,
                                                                     torch_dtype=dtype)
 
-if args.int8 and args.load:
+if args.optimized and args.load:
     from neural_compressor.torch.quantization import load
     example_inputs = {"sample": torch.randn((2, 4, 128, 128), dtype=dtype),
                 "timestep": torch.tensor(951.0),

examples/3.x_api/pytorch/diffusion_model/diffusers/stable_diffusion/smooth_quant/run_benchmark.sh

@@ -20,8 +20,8 @@ function init_params {
       --iters=*)
           iters=$(echo $var | cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo $var | cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo $var | cut -f2 -d=)
       ;;
       --mode=*)
           mode=$(echo $var | cut -f2 -d=)
@@ -43,8 +43,8 @@ function run_benchmark {
     latent="latents.pt"
     base_output_dir="./output/"
 
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
     echo $extra_cmd
 

examples/3.x_api/pytorch/diffusion_model/diffusers/stable_diffusion/smooth_quant/sdxl_smooth_quant.py

@@ -339,7 +339,7 @@ def main():
     )
     parser.add_argument("--quantize", action="store_true")
     parser.add_argument("--load", action="store_true")
-    parser.add_argument("--int8", action="store_true", help="Load quantized model.")
+    parser.add_argument("--optimized", action="store_true", help="Load quantized model.")
     parser.add_argument("--performance", action="store_true")
     parser.add_argument("--n_steps", type=int, default=20)
     parser.add_argument("--batch-size", type=int, default=1)
@@ -404,7 +404,7 @@ def forward_loop(model):
         q_unet.save(args.output_dir)
 
     if args.load:
-        if args.int8:
+        if args.optimized:
             from neural_compressor.torch.quantization import load
             q_unet = load(os.path.abspath(os.path.expanduser(args.output_dir)))
         else:

examples/3.x_api/pytorch/image_recognition/torchvision_models/quantization/static_quant/ipex/README.md

@@ -59,7 +59,7 @@ python main.py -t -a resnet18 --ipex --pretrained /path/to/imagenet
 or
 ```shell
 bash run_quant.sh --input_model=resnet18 --dataset_location=/path/to/imagenet
-bash run_benchmark.sh --input_model=resnet18 --dataset_location=/path/to/imagenet --mode=performance/accuracy --int8=true/false
+bash run_benchmark.sh --input_model=resnet18 --dataset_location=/path/to/imagenet --mode=performance/accuracy --optimized=true/false
 ```
 
 ### 2. ResNet50 With Intel PyTorch Extension
@@ -70,7 +70,7 @@ python main.py -t -a resnet50 --ipex --pretrained /path/to/imagenet
 or
 ```shell
 bash run_quant.sh --input_model=resnet50 --dataset_location=/path/to/imagenet
-bash run_benchmark.sh --input_model=resnet50 --dataset_location=/path/to/imagenet --mode=performance/accuracy --int8=true/false
+bash run_benchmark.sh --input_model=resnet50 --dataset_location=/path/to/imagenet --mode=performance/accuracy --optimized=true/false
 ```
 
 ### 3. ResNext101_32x16d With Intel PyTorch Extension
@@ -81,7 +81,7 @@ python main.py -t -a resnext101_32x16d_wsl --hub --ipex --pretrained /path/to/im
 or
 ```shell
 bash run_quant.sh --input_model=resnext101_32x16d_wsl --dataset_location=/path/to/imagenet
-bash run_benchmark.sh --input_model=resnext101_32x16d_wsl --dataset_location=/path/to/imagenet --mode=performance/accuracy --int8=true/false
+bash run_benchmark.sh --input_model=resnext101_32x16d_wsl --dataset_location=/path/to/imagenet --mode=performance/accuracy --optimized=true/false
 ```
 
 # Run with Intel GPU
@@ -96,5 +96,5 @@ python main.py -t -a resnet18 --ipex --pretrained /path/to/imagenet --xpu
 or
 ```shell
 bash run_quant.sh --input_model=resnet18 --dataset_location=/path/to/imagenet
-bash run_benchmark.sh --input_model=resnet18 --dataset_location=/path/to/imagenet --mode=performance/accuracy --int8=true/false --xpu=true/false
+bash run_benchmark.sh --input_model=resnet18 --dataset_location=/path/to/imagenet --mode=performance/accuracy --optimized=true/false --xpu=true/false
 ```

examples/3.x_api/pytorch/image_recognition/torchvision_models/quantization/static_quant/ipex/main.py

@@ -102,7 +102,7 @@
                     help='For accuracy measurement only.')
 parser.add_argument("--tuned_checkpoint", default='./saved_results', type=str, metavar='PATH',
                     help='path to checkpoint tuned by Neural Compressor (default: ./)')
-parser.add_argument('--int8', dest='int8', action='store_true',
+parser.add_argument('--optimized', dest='optimized', action='store_true',
                     help='run benchmark')
 parser.add_argument('--ipex', dest='ipex', action='store_true',
                     help='tuning or benchmark with Intel PyTorch Extension')
@@ -196,7 +196,7 @@ def main_worker(gpu, ngpus_per_node, args):
             else:
                 model = quantize_models.__dict__[args.arch]()
 
-    if args.ipex and not args.int8:
+    if args.ipex and not args.optimized:
         model = model.to(memory_format=torch.channels_last)
 
     if not torch.cuda.is_available():
@@ -333,8 +333,8 @@ def run_fn(model):
 
     if args.performance or args.accuracy:
         model.eval()
-        if args.int8:
-            print("load int8 model")
+        if args.optimized:
+            print("load optimized model")
             from neural_compressor.torch.quantization import load
             model = load(os.path.abspath(os.path.expanduser(args.tuned_checkpoint)))
         else:

examples/3.x_api/pytorch/image_recognition/torchvision_models/quantization/static_quant/ipex/run_benchmark.sh

@@ -34,8 +34,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --xpu=*)
           xpu=$(echo ${var} |cut -f2 -d=)
@@ -66,8 +66,8 @@ function run_benchmark {
         extra_cmd=$extra_cmd" --hub"
     fi
 
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
 
     if [[ ${xpu} == "true" ]]; then

examples/3.x_api/pytorch/multimodal-modeling/quantization/auto_round/run_benchmark.sh

@@ -35,8 +35,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -64,7 +64,7 @@ function run_benchmark {
         exit 1
     fi
 
-    if [[ ${int8} == "true" ]]; then
+    if [[ ${optimized} == "true" ]]; then
         extra_cmd=$extra_cmd" --load"
     fi
     echo $extra_cmd

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/fp8_quant/run_benchmark.sh

@@ -35,8 +35,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -94,7 +94,7 @@ function run_benchmark {
         python_cmd="deepspeed --num_gpus 8"
     fi
 
-    if [[ ${int8} == "true" ]]; then
+    if [[ ${optimized} == "true" ]]; then
         ${python_cmd} quantize.py \
             --model ${tuned_checkpoint} \
             --load\

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/smooth_quant/run_benchmark.sh

@@ -37,8 +37,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -68,8 +68,8 @@ function run_benchmark {
         exit 1
     fi
 
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
     echo $extra_cmd
 

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/smooth_quant/run_clm_no_trainer.py

@@ -29,7 +29,7 @@
 parser.add_argument(
     "--approach", type=str, default="static", help="Select from ['dynamic', 'static', 'weight-only']"
 )
-parser.add_argument("--int8", action="store_true")
+parser.add_argument("--optimized", action="store_true")
 parser.add_argument("--ipex", action="store_true", help="Use intel extension for pytorch.")
 parser.add_argument("--load", action="store_true", help="Load quantized model.")
 parser.add_argument("--accuracy", action="store_true")
@@ -217,8 +217,8 @@ def eval_func(model):
 
 
 if args.load:
-    if args.int8 or args.int8_bf16_mixed:
-        print("load int8 model")
+    if args.optimized or args.int8_bf16_mixed:
+        print("load optimized model")
         from neural_compressor.torch.quantization import load
 
         tokenizer = AutoTokenizer.from_pretrained(args.model)

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/static_quant/ipex/run_benchmark.sh

@@ -37,8 +37,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -68,8 +68,8 @@ function run_benchmark {
         exit 1
     fi
 
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
     echo $extra_cmd
 

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/static_quant/ipex/run_clm_no_trainer.py

@@ -35,7 +35,7 @@
 )
 parser.add_argument("--approach", type=str, default='static',
                     help="Select from ['dynamic', 'static', 'weight-only']")
-parser.add_argument("--int8", action="store_true")
+parser.add_argument("--optimized", action="store_true")
 parser.add_argument("--ipex", action="store_true", help="Use intel extension for pytorch.")
 parser.add_argument("--load", action="store_true", help="Load quantized model.")
 parser.add_argument("--accuracy", action="store_true")
@@ -198,8 +198,8 @@ def run_fn(model):
     user_model.save(args.output_dir)
 
 if args.load:
-    if args.int8 or args.int8_bf16_mixed:
-        print("load int8 model")
+    if args.optimized or args.int8_bf16_mixed:
+        print("load optimized model")
         from neural_compressor.torch.quantization import load
         tokenizer = AutoTokenizer.from_pretrained(args.model)
         config = AutoConfig.from_pretrained(args.model)

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/static_quant/pt2e/README.md

@@ -32,7 +32,7 @@ python run_clm_no_trainer.py --model facebook/opt-125m --quantize --output_dir q
 python run_clm_no_trainer.py --model facebook/opt-125m --accuracy  --tasks lambada_openai
 
 # Measure the accuracy of the quantized model
-python run_clm_no_trainer.py --model facebook/opt-125m --accuracy  --tasks lambada_openai --int8 --output_dir qmodel_save_path 
+python run_clm_no_trainer.py --model facebook/opt-125m --accuracy  --tasks lambada_openai --optimized --output_dir qmodel_save_path 
 ```
 
 #### Performance
@@ -41,5 +41,5 @@ python run_clm_no_trainer.py --model facebook/opt-125m --accuracy  --tasks lamba
 python run_clm_no_trainer.py --model facebook/opt-125m --performance
 
 # Measure the performance of the quantized model
-python run_clm_no_trainer.py --model facebook/opt-125m --performance --int8 --output_dir qmodel_save_path 
+python run_clm_no_trainer.py --model facebook/opt-125m --performance --optimized --output_dir qmodel_save_path 
 ```

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/static_quant/pt2e/run_benchmark.sh

@@ -36,8 +36,8 @@ function init_params {
       --iters=*)
           iters=$(echo ${var} |cut -f2 -d=)
       ;;
-      --int8=*)
-          int8=$(echo ${var} |cut -f2 -d=)
+      --optimized=*)
+          optimized=$(echo ${var} |cut -f2 -d=)
       ;;
       --config=*)
           tuned_checkpoint=$(echo $var |cut -f2 -d=)
@@ -67,8 +67,8 @@ function run_benchmark {
         exit 1
     fi
 
-    if [[ ${int8} == "true" ]]; then
-        extra_cmd=$extra_cmd" --int8"
+    if [[ ${optimized} == "true" ]]; then
+        extra_cmd=$extra_cmd" --optimized"
     fi
     echo $extra_cmd