Fix AutoRound scale dtype UT (#1688)

Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>

Author: Kaihui-intel

.azure-pipelines/scripts/ut/env_setup.sh

@@ -92,7 +92,7 @@ elif [[ $(echo "${test_case}" | grep -c "tf pruning") != 0 ]]; then
 fi
 
 if [[ $(echo "${test_case}" | grep -c "api") != 0 ]] || [[ $(echo "${test_case}" | grep -c "adaptor") != 0 ]]; then
-    pip install git+https://github.com/intel/auto-round.git@d02f94d4b085523df3b313863fb07f83b2989cce
+    pip install auto-round
 fi
 
 # test deps

neural_compressor/adaptor/pytorch.py

@@ -4937,6 +4937,7 @@ def autoround_quantize(self, model, tune_cfg, dataloader):
         dynamic_max_gap = self.recipes["autoround_args"].get("dynamic_max_gap", -1)
         data_type = self.recipes["autoround_args"].get("data_type", "int")  ##only support data_type
         scale_dtype = self.recipes["autoround_args"].get("scale_dtype", "fp16")
+        amp = self.recipes["autoround_args"].get("amp", True)
 
         model, autoround_config = autoround_quantize(
             model=model,
@@ -4947,6 +4948,7 @@ def autoround_quantize(self, model, tune_cfg, dataloader):
             weight_config=weight_config,
             enable_full_range=enable_full_range,
             batch_size=batch_size,
+            amp=amp,
             lr_scheduler=lr_scheduler,
             dataloader=dataloader,
             dataset_name=dataset_name,

test/adaptor/pytorch_adaptor/test_weight_only_adaptor_pytorch.py

@@ -756,7 +756,6 @@ def test_AutoRound_quant(self):
             tokenizer, seqlen=10, seed=42, train_bs=8, dataset_split="train", dataset_name="NeelNanda/pile-10k"
         )
         fp32_model = copy.deepcopy(self.gptj)
-
         conf = PostTrainingQuantConfig(
             approach="weight_only",
             op_type_dict={
@@ -781,6 +780,7 @@ def test_AutoRound_quant(self):
                     "seq_len": 10,
                     "iters": 10,
                     "scale_dtype": "fp32",
+                    "amp": False,
                 },
             },
         )