Add use_mse_search to GPTQ Config

examples/huggingface/pytorch/text-generation/quantization/run_generation_cpu_woq.py

@@ -133,6 +133,11 @@
     action="store_true",
     help="Use determined group to do quantization",
 )
+parser.add_argument(
+    "--use_mse_search",
+    action="store_true",
+    help="Enables mean squared error (MSE) search.",
+)
 # ============AUTOROUND configs==============
 parser.add_argument(
     "--lr",
@@ -261,6 +266,7 @@
             sym=True if args.scheme == "sym" else False,
             blocksize=args.blocksize,
             static_groups=args.static_groups,
+            use_mse_search=args.use_mse_search,
             group_size=args.group_size,
             n_samples=args.n_samples,
             seq_len=args.seq_len,

intel_extension_for_transformers/transformers/llm/quantization/utils.py

@@ -636,6 +636,7 @@ def convert_to_quantized_model(model, config, device="cpu"):
                 percdamp=config.damp_percent,
                 block_size=config.blocksize,
                 static_groups=config.static_groups,
+                use_mse_search=config.use_mse_search,
             )
             if config.llm_int8_skip_modules != []:
                 for module in config.llm_int8_skip_modules:

intel_extension_for_transformers/transformers/utils/config.py

@@ -879,6 +879,7 @@ def __init__(
         n_samples: int = 128,
         seq_len: int = 2048,
         static_groups: bool = False,
+        use_mse_search: bool = False,
         true_sequential: bool = False,
         layer_wise: bool = False,
         use_ggml: bool = False,
@@ -908,6 +909,7 @@ def __init__(
         self.damp_percent = damp_percent
         self.desc_act = desc_act
         self.static_groups = static_groups
+        self.use_mse_search = use_mse_search
         self.true_sequential = true_sequential
         self.layer_wise = layer_wise
         self.seq_len = seq_len