reduce vlm ut

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

Author: Kaihui-intel

test/3x/torch/quantization/weight_only/test_autoround.py

@@ -238,13 +238,13 @@ def test_mllm(self):
             image_processor=None,
             dataset="liuhaotian/llava_conv_58k",
             extra_data_dir=None,
-            seqlen=512,
+            seqlen=32,
             batch_size=1,
             split=None,
             apply_template=None,
             truncation=False,
             seed=42,
-            nsamples=5,
+            nsamples=1,
             gradient_accumulate_steps=1,
             quant_nontext_module=False,
             processor=processor,
@@ -253,9 +253,9 @@ def test_mllm(self):
             bits=4,
             group_size=128,
             is_mllm=True,
-            nsamples=5,
+            nsamples=1,
             batch_size=batch_size,
-            iters=2,
+            iters=1,
             seqlen=seqlen,
             quant_nontext_module=False,
             truncation=truncation,

test/3x/torch/quantization/weight_only/test_transformers.py

@@ -227,9 +227,9 @@ def test_vlm(self):
             group_size=128,
             is_vlm=True,
             dataset="NeelNanda/pile-10k",  
-            iters=2,
-            n_samples=5,
-            seq_len=512,
+            iters=1,
+            n_samples=1,
+            seq_len=32,
             batch_size=1,
         )