[None][fix] Cherry-pick #20203, #20581, #20575 to release/1.3

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()}")

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)

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)

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())

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)

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)

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)

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)

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)

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)

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)

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)

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)

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)

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)

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)

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())

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)

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)

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)

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)

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)

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)

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)

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)

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)

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)

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)