fix parsing args (#2264)

* fix parsing args

* fix parsing args

* [CI] build eora for test

Author: CSY-ModelCloud

.github/workflows/unit_tests.yml

@@ -71,6 +71,7 @@ env:
   BUILD_QQQ: 1
   BUILD_EORA: 1
   GPTQMODEL_BUILD_EXLLAMA_V1: 1
+  GPTQMODEL_BUILD_EORA: 1
   LEGACY_TESTS: "models/test_internlm.py,models/test_internlm2_5.py,models/test_xverse.py"
   IGNORED_TEST_FILES: "test_tgi.py,test_gptneox.py,models/test_mixtral.py,models/test_phi_3_moe.py,test_bits_new.py"
   GPTQMODEL_FORCE_BUILD: 1

tests/test_cpu_gpu_memory_copy.py

@@ -105,7 +105,7 @@ def main():
     parser.add_argument("--gpu", type=int, default=0, help="GPU id to test against")
     parser.add_argument("--total-gib", type=float, default=40.0, help="Total GiB to stream per direction per mode")
     parser.add_argument("--chunk-gib", type=float, default=1.0, help="Chunk size GiB per copy")
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
 
     if not torch.cuda.is_available():
         raise SystemExit("CUDA not available.")

tests/test_gpu_gpu_memory_copy.py

@@ -77,7 +77,7 @@ def main():
     parser.add_argument("--dst", type=int, default=1, help="destination GPU id")
     parser.add_argument("--total-gib", type=float, default=40.0, help="total GiB to stream per direction")
     parser.add_argument("--chunk-gib", type=float, default=1.0, help="chunk size GiB per copy")
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
 
     if not torch.cuda.is_available() or torch.cuda.device_count() < 2:
         raise SystemExit("Need at least 2 CUDA devices.")