[CI/Build] Fix and re-enable v1 PP test on CI (#25496)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>

tests/distributed/test_pipeline_parallel.py

@@ -382,7 +382,6 @@ def test_tp_language_generation(
     test_options: PPTestOptions,
     num_gpus_available,
 ):
-    pytest.skip("Skipping the test until V1 passes it.")
     _compare_tp(model_id,
                 parallel_setup,
                 distributed_backend,
@@ -410,7 +409,6 @@ def test_tp_language_embedding(
     test_options: PPTestOptions,
     num_gpus_available,
 ):
-    pytest.skip("Skipping the test until V1 passes it.")
     _compare_tp(model_id,
                 parallel_setup,
                 distributed_backend,
@@ -438,7 +436,6 @@ def test_tp_multimodal_generation(
     test_options: PPTestOptions,
     num_gpus_available,
 ):
-    pytest.skip("Skipping the test until V1 passes it.")
     _compare_tp(model_id,
                 parallel_setup,
                 distributed_backend,

vllm/model_executor/models/granite.py

@@ -308,13 +308,11 @@ class GraniteModel(nn.Module):
                 hidden_states = inputs_embeds
             else:
                 hidden_states = self.get_input_embeddings(input_ids)
-            residual = None
 
             hidden_states *= self.config.embedding_multiplier
         else:
             assert intermediate_tensors is not None
             hidden_states = intermediate_tensors["hidden_states"]
-            residual = intermediate_tensors["residual"]
 
         for layer in islice(self.layers, self.start_layer, self.end_layer):
             hidden_states = layer(positions, hidden_states)
@@ -322,7 +320,6 @@ class GraniteModel(nn.Module):
         if not get_pp_group().is_last_rank:
             return IntermediateTensors({
                 "hidden_states": hidden_states,
-                "residual": residual
             })
 
         hidden_states = self.norm(hidden_states)
@@ -475,10 +472,6 @@ class GraniteForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             torch.zeros((batch_size, self.config.hidden_size),
                         dtype=dtype,
                         device=device),
-            "residual":
-            torch.zeros((batch_size, self.config.hidden_size),
-                        dtype=dtype,
-                        device=device),
         })
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/granitemoe.py

@@ -298,17 +298,14 @@ class GraniteMoeModel(nn.Module):
             else:
                 hidden_states = self.get_input_embeddings(input_ids)
             hidden_states *= self.embedding_multiplier
-            residual = None
         else:
             assert intermediate_tensors is not None
             hidden_states = intermediate_tensors["hidden_states"]
-            residual = intermediate_tensors["residual"]
         for layer in islice(self.layers, self.start_layer, self.end_layer):
             hidden_states = layer(positions, hidden_states)
         if not get_pp_group().is_last_rank:
             return IntermediateTensors({
                 "hidden_states": hidden_states,
-                "residual": residual
             })
         hidden_states = self.norm(hidden_states)
         return hidden_states
@@ -523,10 +520,6 @@ class GraniteMoeForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             torch.zeros((batch_size, self.config.hidden_size),
                         dtype=dtype,
                         device=device),
-            "residual":
-            torch.zeros((batch_size, self.config.hidden_size),
-                        dtype=dtype,
-                        device=device),
         })
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/granitemoeshared.py

@@ -195,17 +195,14 @@ class GraniteMoeSharedModel(nn.Module):
             else:
                 hidden_states = self.get_input_embeddings(input_ids)
             hidden_states *= self.embedding_multiplier
-            residual = None
         else:
             assert intermediate_tensors is not None
             hidden_states = intermediate_tensors["hidden_states"]
-            residual = intermediate_tensors["residual"]
         for layer in islice(self.layers, self.start_layer, self.end_layer):
             hidden_states = layer(positions, hidden_states)
         if not get_pp_group().is_last_rank:
             return IntermediateTensors({
                 "hidden_states": hidden_states,
-                "residual": residual
             })
         hidden_states = self.norm(hidden_states)
         return hidden_states
@@ -323,10 +320,6 @@ class GraniteMoeSharedForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             torch.zeros((batch_size, self.config.hidden_size),
                         dtype=dtype,
                         device=device),
-            "residual":
-            torch.zeros((batch_size, self.config.hidden_size),
-                        dtype=dtype,
-                        device=device),
         })
 
     def load_weights(self, weights: Iterable[tuple[str,