[FIX] ci test (#2843)

* fix test_llama3_2_exllamav3.py

Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>

* fix test_falcon.py and test_gpt_oss.py

Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>

---------

Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>

Author: ZX-ModelCloud

gptqmodel/looper/exllamav3_processor.py

@@ -141,6 +141,7 @@ def preprocess(self, module: NamedModule, fallback=None, **kwargs):
             sym=True,
             device=module_qcfg.device,
             pack_dtype=module_qcfg.pack_dtype,
+            act_group_aware=False,
         )
 
         task = GPTQ(module=module, qcfg=capture_qcfg)

tests/models/test_falcon.py

@@ -4,6 +4,8 @@
 # Contact: qubitium@modelcloud.ai, x.com/qubitium
 
 import torch  # noqa: E402
+
+from gptqmodel import BACKEND
 from model_test import ModelTest
 
 
@@ -22,6 +24,7 @@ class TestFalcon(ModelTest):
     EVAL_BATCH_SIZE = 6
     USE_VLLM = False
     ACT_GROUP_AWARE = False
+    LOAD_BACKEND = BACKEND.AUTO
 
     def test_falcon(self):
         self.quantize_and_evaluate()

tests/models/test_gpt_oss.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: 2024-2025 qubitium@modelcloud.ai
 # SPDX-License-Identifier: Apache-2.0
 # Contact: qubitium@modelcloud.ai, x.com/qubitium
-
+from gptqmodel import BACKEND
 from model_test import ModelTest
 
 
@@ -21,6 +21,7 @@ class TestGPTOSS(ModelTest):
     EVAL_BATCH_SIZE = 6
     USE_VLLM = False
     ACT_GROUP_AWARE = False
+    LOAD_BACKEND = BACKEND.AUTO
 
     def test_gpt_oss(self):
         self.quantize_and_evaluate()