[torch.compile][TPU] Make @support_torch_compile work for XLA backend (vllm-project#15782)

Signed-off-by: Siyuan Liu <lsiyuan@google.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>

Author: 2 people

vllm/v1/worker/tpu_model_runner.py

@@ -15,6 +15,7 @@
 import vllm.envs as envs
 from vllm.attention.backends.abstract import AttentionType
 from vllm.attention.layer import Attention
+from vllm.compilation.wrapper import TorchCompileWrapperWithCustomDispatcher
 from vllm.config import VllmConfig
 from vllm.forward_context import set_forward_context
 from vllm.logger import init_logger
@@ -691,11 +692,10 @@ def execute_model(
             hidden_states = self.model(
                 input_ids=input_ids,
                 positions=self.position_ids,
-                kv_caches=self.kv_caches,
                 inputs_embeds=inputs_embeds,
             )
-        selected_token_ids = self.model.sample_from_hidden(
-            hidden_states, tpu_sampling_metadata)
+        selected_token_ids = self.sample_from_hidden(hidden_states,
+                                                     tpu_sampling_metadata)
         # Remove padding on cpu and keep dynamic op outside of xla graph.
         selected_token_ids = selected_token_ids.cpu()[:num_reqs]
 
@@ -795,17 +795,15 @@ def load_model(self) -> None:
                 "get_tensor_model_parallel_rank",
                 return_value=xm_tp_rank):
             model = get_model(vllm_config=self.vllm_config)
-        model = model.eval()
+        # Sync all pending XLA execution during model initialization and weight
+        # loading.
         xm.mark_step()
         xm.wait_device_ops()
-        model = ModelWrapperV1(model)
-        self.model = torch.compile(model,
-                                   backend="openxla",
-                                   fullgraph=True,
-                                   dynamic=False)
+        self.model = model
+        self.sampler = TPUSampler()
 
     @torch.no_grad()
-    def _dummy_run(self, kv_caches, num_tokens: int) -> None:
+    def _dummy_run(self, num_tokens: int) -> None:
         if self.is_multimodal_model:
             input_ids = None
             inputs_embeds = torch.zeros((num_tokens, self.hidden_size),
@@ -856,7 +854,6 @@ def _dummy_run(self, kv_caches, num_tokens: int) -> None:
         with set_forward_context(attn_metadata, self.vllm_config, 0):
             out = self.model(input_ids=input_ids,
                              positions=position_ids,
-                             kv_caches=kv_caches,
                              inputs_embeds=inputs_embeds)
         self._hidden_states_dtype = out.dtype
 
@@ -868,7 +865,7 @@ def capture_model(self) -> None:
         start = time.perf_counter()
         for num_tokens in self.num_tokens_paddings:
             logger.info("  -- num_tokens: %d", num_tokens)
-            self._dummy_run(self.kv_caches, num_tokens)
+            self._dummy_run(num_tokens)
             xm.mark_step()
         xm.wait_device_ops()
         end = time.perf_counter()
@@ -899,8 +896,7 @@ def capture_model(self) -> None:
                     from_input_batch(self.input_batch, indices)
                 logger.info("  -- num_tokens: %d, num_seqs: %d", num_tokens,
                             num_reqs_to_sample)
-                out = self.model.sample_from_hidden(dummy_hidden,
-                                                    sampling_meta)
+                out = self.sample_from_hidden(dummy_hidden, sampling_meta)
                 out = out.cpu()
                 # Requests can't be more than tokens. But do compile for the
                 # next bigger value in case num_tokens uses bucketed padding.
@@ -954,79 +950,48 @@ def initialize_kv_cache(self, kv_cache_config: KVCacheConfig) -> None:
             self.vllm_config.compilation_config.static_forward_context,
             self.kv_caches)
 
-
-class ModelWrapperV1(nn.Module):
-
-    def __init__(self, model: nn.Module):
-        super().__init__()
-        self.model = model
-        self.sampler = TPUSampler()
-
-    def sample(
-            self, logits: torch.Tensor,
-            sampling_metadata: TPUSupportedSamplingMetadata) -> SamplerOutput:
-        sampler_out = self.sampler(logits, sampling_metadata)
-        return sampler_out
-
-    def forward(
-        self,
-        input_ids: torch.Tensor,
-        positions: torch.Tensor,
-        kv_caches: list[torch.Tensor],
-        inputs_embeds: Optional[torch.Tensor] = None,
-    ) -> torch.Tensor:
-        """Executes the forward pass of the model.
-
-        Args:
-            input_ids: The input token IDs of shape [num_tokens].
-            positions: The input position IDs of shape [num_tokens].
-            kv_caches: The key and value caches. They can be None during the
-                memory profiling at initialization.
-            inputs_embeds: The input embeddings of shape [num_tokens,
-                hidden_size]. It is used for multimodal models.
-        """
-
-        hidden_states = self.model(
-            input_ids=input_ids,
-            positions=positions,
-            inputs_embeds=inputs_embeds,
-        )
-
-        return hidden_states
+    def reset_dynamo_cache(self):
+        if self.is_multimodal_model:
+            assert hasattr(self.model, "language_model")
+            compiled_model = self.model.language_model.model
+        else:
+            compiled_model = self.model.model
+        if isinstance(compiled_model, TorchCompileWrapperWithCustomDispatcher):
+            logger.info("Clear dynamo cache and cached dynamo bytecode.")
+            torch._dynamo.eval_frame.remove_from_cache(
+                compiled_model.original_code_object)
+            compiled_model.compiled_codes.clear()
 
     def sample_from_hidden(
         self,
         hidden_states: torch.Tensor,
         sampling_metadata: TPUSupportedSamplingMetadata,
     ) -> torch.Tensor:
         """
-        Sample with xla-friendly function. This function is to be traced 
-        separately from `forward` for lighter compilation overhead.
-        """
+            Sample with xla-friendly function. This function is to be traced 
+            separately for lighter compilation overhead.
+            """
         # Tensor `sample_hidden_states` is of fixed pre-compiled size.
         sample_hidden_states = \
             hidden_states[sampling_metadata.indices_do_sample]
-        logits = self.compute_logits(sample_hidden_states)
+        # SamplingMetadata here for pruning output in LogitsProcessor, disabled.
+        logits = self.model.compute_logits(sample_hidden_states, None)
+
+        def sample(
+                logits: torch.Tensor,
+                sampling_metadata: TPUSupportedSamplingMetadata
+        ) -> SamplerOutput:
+            sampler_out = self.sampler(logits, sampling_metadata)
+            return sampler_out
+
         # Optimized greedy sampling branch, tracing both paths in a single pass
         # NOTE all_greedy is a scalar, this is just an optimized if/else.
-        out_tokens = torch.where(sampling_metadata.all_greedy,
-                        torch.argmax(logits, dim=-1, keepdim=True),
-                        self.sample(logits, sampling_metadata)\
-                                            .sampled_token_ids)
+        out_tokens = torch.where(
+            sampling_metadata.all_greedy,
+            torch.argmax(logits, dim=-1, keepdim=True),
+            sample(logits, sampling_metadata).sampled_token_ids)
         return out_tokens
 
-    def compute_logits(self,
-                       hidden_states: torch.Tensor) -> Optional[torch.Tensor]:
-        # SamplingMetadata here for pruning output in LogitsProcessor, disabled
-        logits = self.model.compute_logits(hidden_states, None)
-        return logits
-
-    def get_multimodal_embeddings(self, *args, **kwargs):
-        return self.model.get_multimodal_embeddings(*args, **kwargs)
-
-    def get_input_embeddings(self, *args, **kwargs):
-        return self.model.get_input_embeddings(*args, **kwargs)
-
 
 def _get_padded_number(n: int, multiple: int) -> int:
     return ((n + multiple - 1) // multiple) * multiple

vllm/v1/worker/tpu_worker.py

@@ -157,13 +157,19 @@ def determine_available_memory(self) -> int:
             runner_kv_caches)
 
         self.model_runner._dummy_run(
-            runner_kv_caches,
-            num_tokens=self.scheduler_config.max_num_batched_tokens,
-        )
+            self.scheduler_config.max_num_batched_tokens)
 
         # Synchronize before measuring the memory usage.
         xm.wait_device_ops()
 
+        # During the profiling run, the model runs without KV cache. After
+        # the profiling run, the model always runs with KV cache. Here we clear
+        # the dynamo cache and cached bytecode to ensure the model always has
+        # one compiled bytecode. Having one FX graph/cached bytecode per
+        # compiled model is required for `support_torch_compile` decorator to
+        # skip dynamo guard.
+        self.model_runner.reset_dynamo_cache()
+
         # Get the maximum amount of memory used by the model weights and
         # intermediate activations.
         m = xm.get_memory_info(self.device)