diff --git a/tensorrt_llm/media/encoding.py b/tensorrt_llm/media/encoding.py
index 0e2e32490909..5a5484f79e1f 100644
--- a/tensorrt_llm/media/encoding.py
+++ b/tensorrt_llm/media/encoding.py
@@ -181,7 +181,7 @@ def encode_video(
                     stdout=subprocess.PIPE,
                     stderr=subprocess.PIPE,
                 )
-                raw_frames = video_np.tobytes()
+                raw_frames = memoryview(video_np.reshape(-1))
                 _, stderr = process.communicate(input=raw_frames)
                 if process.returncode != 0:
                     raise RuntimeError(f"ffmpeg encoding failed: {stderr.decode()}")
diff --git a/tests/integration/defs/examples/visual_gen/test_visual_gen_multi_gpu.py b/tests/integration/defs/examples/visual_gen/test_visual_gen_multi_gpu.py
index b5a1c211665f..23c46bb31613 100644
--- a/tests/integration/defs/examples/visual_gen/test_visual_gen_multi_gpu.py
+++ b/tests/integration/defs/examples/visual_gen/test_visual_gen_multi_gpu.py
@@ -146,7 +146,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt
index 3696b661b7ad..a3a6045b692f 100644
--- a/tests/integration/test_lists/waives.txt
+++ b/tests/integration/test_lists/waives.txt
@@ -228,10 +228,6 @@ unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py:
 unittest/_torch/speculative/hw_agnostic/test_dflash.py::test_dflash_qwen3_5_4b[False] SKIP (https://nvbugs/6535767)
 unittest/_torch/speculative/hw_agnostic/test_dflash.py::test_dflash_qwen3_5_4b[True] SKIP (https://nvbugs/6535767)
 unittest/_torch/speculative/test_eagle3.py SKIP (https://nvbugs/5461761)
-unittest/_torch/thop/parallel/test_fp4_linear.py::test_fp4_gemm_bias_per_backend[mnk2-cublaslt] SKIP (https://nvbugs/6581067)
-unittest/_torch/thop/parallel/test_fp4_linear.py::test_fp4_gemm_bias_per_backend[mnk2-cutlass] SKIP (https://nvbugs/6581067)
-unittest/_torch/thop/parallel/test_fp4_linear.py::test_fp4_gemm_bias_per_backend[mnk3-cublaslt] SKIP (https://nvbugs/6581067)
-unittest/_torch/thop/parallel/test_fp4_linear.py::test_fp4_gemm_bias_per_backend[mnk3-cutlass] SKIP (https://nvbugs/6581067)
 unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py::TestAttn2DAttentionGQAAndCrossAttention::test_attn2d_cross_attention SKIP (https://nvbugs/6890768)
 unittest/_torch/visual_gen/test_cosmos3_t2v_offload.py::TestCosmos3Offload::test_cosmos3_offload_matches_baseline SKIP (https://nvbugs/6702264)
 unittest/bindings/test_transfer_agent_bindings.py::TestMooncakeFunctionalTransfer::test_mooncake_wait_in_progress_on_zero_timeout SKIP (https://nvbugs/6517836)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py
index f71f61ecd714..32113154f788 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py
@@ -120,7 +120,7 @@ def forward_with_lse(self, q, k, v, batch_size=None, seq_len=None, **kwargs):
 
 
 def _init_worker(rank: int, world_size: int, backend: str, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     if backend == "nccl" and torch.cuda.is_available():
         torch.cuda.set_device(rank % torch.cuda.device_count())
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py
index 22801ba49903..a9f3d1efd9c8 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py
@@ -171,7 +171,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py
index dd33d28f2375..14bd1bfc184c 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py
@@ -63,7 +63,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py
index ffab5efaf4b3..384da03b07f2 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py
@@ -69,7 +69,7 @@ def _cleanup_mpi_env():
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
     """Initialize distributed environment for a worker process."""
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py
index 34afe03f0d5b..3d54e917020e 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py
@@ -45,7 +45,7 @@ def _cleanup_mpi_env():
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
     """Initialize distributed environment for a worker process."""
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py
index 92215e3183fb..a625de782272 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py
@@ -56,7 +56,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_parallel_vae.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_parallel_vae.py
index 562dc6d0e328..047541dca183 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_parallel_vae.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_parallel_vae.py
@@ -42,7 +42,7 @@ def _cleanup_mpi_env():
 # Distributed helpers
 # ---------------------------------------------------------------------------
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py
index 211d19bbae71..14204a027207 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py
@@ -45,7 +45,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py
index 7747f0af6c7f..1b6078b259f4 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py
@@ -34,7 +34,7 @@ def _cleanup_mpi_env():
 
 
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py
index 4a8eeb7e61bf..81a8aa2416c8 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py
@@ -45,7 +45,7 @@ def _cleanup_mpi_env():
 
 
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py
index e3eec279711c..87429988d4f3 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py
@@ -29,7 +29,7 @@ def _cleanup_mpi_env():
 
 
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py
index f7ef7160d253..b4db6fb5960e 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py
@@ -42,7 +42,7 @@ def _cleanup_mpi_env():
 
 
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_qwen_image_attention_parallel.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_qwen_image_attention_parallel.py
index 74412b67946c..08613f5a4ecc 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_qwen_image_attention_parallel.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_qwen_image_attention_parallel.py
@@ -58,7 +58,7 @@ def _cleanup_mpi_env():
 def _init_distributed_worker(
     rank: int, world_size: int, backend: str = "nccl", port: int = 29500
 ) -> None:
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py
index dc512935fa26..ceeb15d1897d 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py
@@ -107,7 +107,7 @@ def forward_with_lse(self, q, k, v, batch_size=None, seq_len=None, **kwargs):
 
 
 def _init_worker(rank: int, world_size: int, backend: str, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     if backend == "nccl" and torch.cuda.is_available():
         torch.cuda.set_device(rank % torch.cuda.device_count())
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py
index f5523fe02973..8878cd788963 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py
@@ -58,7 +58,7 @@ def _cleanup_mpi_env():
 
 
 def _init_worker(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py
index 0f22d38a843f..8439e8cf73b4 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py
@@ -37,7 +37,7 @@
 
 
 def _init_dist(rank: int, world_size: int, port: int):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py
index 8415bd6bc52a..6f022a380259 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py
@@ -36,7 +36,7 @@ def _cleanup_mpi_env():
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "gloo", port: int = 29500):
     """Initialize distributed environment for a worker process."""
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py
index e2669bc337c5..7559c8182162 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py
@@ -64,7 +64,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py
index 75b4d6a58ff0..64fa4978cad0 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py
@@ -27,7 +27,7 @@ def _cleanup_mpi_env():
 
 
 def _init_dist(rank, world_size, port):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py
index 96fba95677f5..c7ec49a250ec 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py
@@ -54,7 +54,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py
index d0d26f646190..0b4045c75f0b 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py
@@ -88,7 +88,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py
index 6f3aef886476..34b58e4dcc2f 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py
@@ -60,7 +60,7 @@ def _cleanup_mpi_env():
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
     """Initialize distributed environment for a worker process."""
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py
index 0bda70d2729d..a4db10e4a50e 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py
@@ -71,7 +71,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_vsa_ulysses.py b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_vsa_ulysses.py
index a04eca96a0c1..6b5910233206 100644
--- a/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_vsa_ulysses.py
+++ b/tests/unittest/_torch/visual_gen/multi_gpu/test_wan_vsa_ulysses.py
@@ -64,7 +64,7 @@ def _cleanup_mpi_env():
 
 
 def init_distributed_worker(rank: int, world_size: int, backend: str = "nccl", port: int = 29500):
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)
diff --git a/tests/unittest/_torch/visual_gen/test_flux_pipeline.py b/tests/unittest/_torch/visual_gen/test_flux_pipeline.py
index 786bb2af4d94..f2d76151b783 100644
--- a/tests/unittest/_torch/visual_gen/test_flux_pipeline.py
+++ b/tests/unittest/_torch/visual_gen/test_flux_pipeline.py
@@ -1007,7 +1007,7 @@ def test_flux2_num_images_per_prompt_batch(self, flux2_pipeline):
 def _setup_distributed(rank, world_size, master_port, backend="nccl"):
     """Initialize distributed process group for multi-GPU tests."""
     os.environ["TLLM_DISABLE_MPI"] = "1"
-    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(master_port)
     os.environ["RANK"] = str(rank)
     os.environ["WORLD_SIZE"] = str(world_size)