[misc] LoRA - Skip LoRA kernels when not required (#15152)

Signed-off-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>

vllm/lora/ops/triton_ops/lora_expand.py

@@ -136,6 +136,7 @@ def _lora_expand(
     num_tokens_per_lora: torch.Tensor,  # shape [max-loras + 1]
     lora_token_start_loc: torch.Tensor,  # shape [max-loras + 2]
     lora_ids: torch.Tensor,  # shape [max-loras + 1]
+    no_lora_flag_cpu: torch.Tensor,  # shape [1] 
     offset_start: int = 0,
     add_inputs: bool = False,
 ) -> None:
@@ -157,11 +158,19 @@ def _lora_expand(
             identifies the the region in token_indices_sorted_by_lora_ids that
             LoRA lora_ids[i] should process.
         lora_ids (torch.Tensor): LoRA ids to process.
+        no_lora_flag_cpu (torch.Tensor): A CPU tensor of size 1, that indicates
+            if there are any requests that require LoRA.
         offset_start (int, optional): Offset start for output_tensor. 
             Defaults to 0.
         add_inputs (bool, optional): Whether to add the input tensor to the 
             output tensor. Defaults to False.
     """
+
+    assert no_lora_flag_cpu.numel() == 1
+    if no_lora_flag_cpu.item():
+        # None of the inputs require LoRA.
+        return
+
     assert inputs.dtype in [torch.float16, torch.bfloat16, torch.float32]
     for weight in lora_b_weights:
         assert weight.dtype in [torch.float16, torch.bfloat16]
@@ -170,6 +179,8 @@ def _lora_expand(
     assert output_tensor.is_contiguous()
 
     # metadata sanity check.
+    M = inputs.size(1)
+    assert token_lora_mapping.size(0) == M
     assert token_lora_mapping.size(0) == token_indices_sorted_by_lora_ids.size(
         0)
     assert lora_ids.size(0) == num_tokens_per_lora.size(0)
@@ -181,7 +192,6 @@ def _lora_expand(
                                            inputs.device)
 
     K = lora_b_weights[0].shape[-1]  # K= rank
-    M = inputs.size(1)
     ADD_INPUTS = add_inputs
     MAX_LORAS = lora_ids.size(0)
     CAST_TYPE = False
@@ -263,6 +273,7 @@ def _lora_expand_fake(
     num_tokens_per_lora: torch.Tensor,
     lora_token_start_loc: torch.Tensor,
     lora_ids: torch.Tensor,
+    no_lora_flag_cpu: torch.Tensor,
     offset_start: int = 0,
     add_inputs: bool = False,
 ) -> None:

vllm/lora/ops/triton_ops/lora_kernel_metadata.py

@@ -17,6 +17,17 @@ class LoRAKernelMeta:
     num_tokens_per_lora: torch.Tensor
     lora_token_start_loc: torch.Tensor
 
+    # The V1 architecture uses the traced torch.compile graphs to execute
+    # a forward pass. Things to note about this process,
+    # 1. The tracing infers all python scalar datatype objects into a constant
+    # value.
+    # 2. The tracing cannot handle dynamic control flow. (dynamic control flow
+    # is an experimental feature in pytorch)
+    # 3. The internals of torch.ops functions are not traced.
+    # We disguise the "no_lora" flag as a cpu tensor and leverage point number 3
+    # to early exit from inside the lora_expand / lora_shrink torch operation.
+    no_lora_flag_cpu: torch.Tensor
+
     @staticmethod
     def make(max_loras: int, max_num_tokens: int,
              device: Union[torch.device, str]) -> "LoRAKernelMeta":
@@ -47,17 +58,24 @@ class LoRAKernelMeta:
         lora_token_start_loc = torch.zeros(max_loras + 2,
                                            dtype=torch.int32,
                                            device=device)
+
+        no_lora_flag_cpu = torch.tensor([False],
+                                        dtype=torch.bool,
+                                        device='cpu')
+
         return LoRAKernelMeta(
             token_lora_mapping=token_lora_mapping,
             token_indices_sorted_by_lora_ids=token_indices_sorted_by_lora_ids,
             active_lora_ids=active_lora_ids,
             num_tokens_per_lora=num_tokens_per_lora,
-            lora_token_start_loc=lora_token_start_loc)
+            lora_token_start_loc=lora_token_start_loc,
+            no_lora_flag_cpu=no_lora_flag_cpu)
 
     def _reset(self):
         self.active_lora_ids.fill_(-1)
         self.num_tokens_per_lora.fill_(0)
         self.lora_token_start_loc.fill_(0)
+        self.no_lora_flag_cpu.fill_(False)
 
     def prepare_tensors(self, token_lora_mapping: torch.Tensor) -> None:
         """
@@ -70,6 +88,14 @@ class LoRAKernelMeta:
 
         self._reset()
 
+        # Check and record no-lora case.
+        no_lora = torch.all(token_lora_mapping == -1)
+        self.no_lora_flag_cpu[0] = no_lora
+
+        if no_lora:
+            # Early exit. LoRA kernels will not be run.
+            return
+
         num_tokens = token_lora_mapping.size(0)
 
         # copy token lora mapping
@@ -100,7 +126,7 @@ class LoRAKernelMeta:
     def meta_args(
         self, token_nums: int
     ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor,
-               torch.Tensor]:
+               torch.Tensor, torch.Tensor]:
         """
         This function returns the kernel metadata required for the current
         forward pass execution of the kernel. The function returns all the
@@ -111,7 +137,11 @@ class LoRAKernelMeta:
             token_nums (int): Number of input tokens in the current forward
             pass. 
         """
-        return (self.token_lora_mapping[:token_nums],
+        return (
+            self.token_lora_mapping[:token_nums],
             self.token_indices_sorted_by_lora_ids[:token_nums],
-                self.num_tokens_per_lora, self.lora_token_start_loc,
-                self.active_lora_ids)
+            self.num_tokens_per_lora,
+            self.lora_token_start_loc,
+            self.active_lora_ids,
+            self.no_lora_flag_cpu,
+        )

vllm/lora/ops/triton_ops/lora_shrink.py

@@ -106,6 +106,7 @@ def _lora_shrink(
     num_tokens_per_lora: torch.Tensor,  # shape [max-loras + 1]
     lora_token_start_loc: torch.Tensor,  # shape [max-loras + 2]
     lora_ids: torch.Tensor,  # shape [max-loras + 1]
+    no_lora_flag_cpu: torch.Tensor,  # shape [1]
     scaling: float,
 ) -> None:
     """
@@ -126,8 +127,16 @@ def _lora_shrink(
             identifies the region in token_indices_sorted_by_lora_ids that
             LoRA lora_ids[i] should process.
         lora_ids (torch.Tensor): LoRA ids to process.
+        no_lora_flag_cpu (torch.Tensor): A CPU tensor of size 1, that indicates
+            if there are any requests that require LoRA.
         scaling (float): Scaling factor.
     """
+
+    assert no_lora_flag_cpu.numel() == 1
+    if no_lora_flag_cpu.item():
+        # None of the inputs require LoRA.
+        return
+
     assert inputs.dtype == lora_a_weights[0].dtype
     assert inputs.dtype in [torch.float16, torch.bfloat16]
     for weight in lora_a_weights:
@@ -138,6 +147,8 @@ def _lora_shrink(
     assert output_tensor.is_contiguous()
 
     # metadata sanity check
+    M = inputs.size(0)
+    assert token_lora_mapping.size(0) == M
     assert token_lora_mapping.size(0) == token_indices_sorted_by_lora_ids.size(
         0)
     assert lora_ids.size(0) == num_tokens_per_lora.size(0)
@@ -146,7 +157,6 @@ def _lora_shrink(
     (lora_ptr_tensor, lora_strides_d0, lora_strides_d1,
      lora_strides_d2) = _get_lora_a_ptr(lora_a_weights, inputs.device)
     N, K = lora_a_weights[0].shape[-2:]  # K=hidden_size,N=rank
-    M = inputs.size(0)
     NUM_SLICES = len(lora_a_weights)
     MAX_LORAS = lora_ids.size(0)
 
@@ -218,6 +228,7 @@ def _lora_shrink_fake(
     num_tokens_per_lora: torch.Tensor,
     lora_token_start_loc: torch.Tensor,
     lora_ids: torch.Tensor,
+    no_lora_flag_cpu: torch.Tensor,
     scaling: float,
 ) -> None:
     return

vllm/worker/model_runner.py

@@ -1242,6 +1242,29 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
         max_num_seqs = self.scheduler_config.max_num_seqs
         self._dummy_run(max_num_batched_tokens, max_num_seqs)
 
+    def _add_dummy_loras(self, num_loras: int) -> list[LoRARequest]:
+        assert num_loras > 0
+        assert self.lora_manager is not None
+
+        dummy_lora_requests: list[LoRARequest] = []
+        with self.lora_manager.dummy_lora_cache():
+            for idx in range(num_loras):
+                lora_id = idx + 1
+                dummy_lora_request = LoRARequest(
+                    lora_name=f"warmup_{lora_id}",
+                    lora_int_id=lora_id,
+                    lora_path="/not/a/real/path",
+                )
+                self.lora_manager.add_dummy_lora(dummy_lora_request,
+                                                 rank=LORA_WARMUP_RANK)
+                dummy_lora_requests.append(dummy_lora_request)
+        return dummy_lora_requests
+
+    def _remove_dummy_loras(self):
+        # Remove dummy loras.
+        assert self.lora_manager is not None
+        self.remove_all_loras()
+
     def _dummy_run(self,
                    max_num_batched_tokens: int,
                    max_num_seqs: int = 1) -> None:
@@ -1251,24 +1274,16 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
                 SamplingParams(top_p=0.99, top_k=self.vocab_size - 1)
 
             # This represents the maximum number of different requests
-            # that will have unique loras, an therefore the max amount of memory
-            # consumption create dummy lora request copies from the lora request
-            # passed in, which contains a lora from the lora warmup path.
+            # that will have unique loras, and therefore the max amount of
+            # memory consumption. Create dummy lora request copies from the
+            # lora request passed in, which contains a lora from the lora
+            # warmup path.
             dummy_lora_requests: List[LoRARequest] = []
             dummy_lora_requests_per_seq: List[LoRARequest] = []
             if self.lora_config:
-                assert self.lora_manager is not None
-                with self.lora_manager.dummy_lora_cache():
-                    for idx in range(self.lora_config.max_loras):
-                        lora_id = idx + 1
-                        dummy_lora_request = LoRARequest(
-                            lora_name=f"warmup_{lora_id}",
-                            lora_int_id=lora_id,
-                            lora_path="/not/a/real/path",
-                        )
-                        self.lora_manager.add_dummy_lora(dummy_lora_request,
-                                                         rank=LORA_WARMUP_RANK)
-                        dummy_lora_requests.append(dummy_lora_request)
+                dummy_lora_requests = self._add_dummy_loras(
+                    self.lora_config.max_loras)
+                assert len(dummy_lora_requests) == self.lora_config.max_loras
                 dummy_lora_requests_per_seq = [
                     dummy_lora_requests[idx % len(dummy_lora_requests)]
                     for idx in range(max_num_seqs)
@@ -1354,9 +1369,8 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
             self.execute_model(model_input, kv_caches, intermediate_tensors)
             torch.cuda.synchronize()
             if self.lora_config:
-                # Remove dummy loras.
-                assert self.lora_manager is not None
-                self.remove_all_loras()
+                self._remove_dummy_loras()
+
             return
 
     def remove_all_loras(self):
@@ -1479,6 +1493,16 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
                 dtype=self.model_config.dtype,
                 device=self.device)
 
+        dummy_lora_id: Optional[int] = None
+        dummy_lora_request: LoRARequest = []
+        if self.lora_config:
+            # The goal is to capture the LoRA kernels in cuda graphs.
+            # for this purpose, as single dummy lora is sufficient.
+            dummy_lora_requests = self._add_dummy_loras(num_loras=1)
+            assert len(dummy_lora_requests) == 1
+            dummy_lora_request = dummy_lora_requests[0]
+            dummy_lora_id = dummy_lora_request.lora_int_id
+
         with self.attn_state.graph_capture(max_batch_size), graph_capture(
                 self.device) as graph_capture_context:
             # NOTE: Capturing the largest batch size first may help reduce the
@@ -1503,10 +1527,11 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
                     attn_metadata.enable_kv_scales_calculation = False
                     if self.lora_config:
                         lora_mapping = LoRAMapping(
-                            **dict(index_mapping=[0] * batch_size,
-                                   prompt_mapping=[0] * batch_size,
+                            **dict(index_mapping=[dummy_lora_id] * batch_size,
+                                   prompt_mapping=[dummy_lora_id] * batch_size,
                                    is_prefill=False))
-                        self.set_active_loras(set(), lora_mapping)
+                        self.set_active_loras(set([dummy_lora_request]),
+                                              lora_mapping)
 
                     if self.prompt_adapter_config:
                         prompt_adapter_mapping = PromptAdapterMapping(
@@ -1562,6 +1587,9 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
                     self.graph_runners[virtual_engine][batch_size] = (
                         graph_runner)
 
+        if self.lora_config:
+            self._remove_dummy_loras()
+
         end_time = time.perf_counter()
         end_free_gpu_memory = torch.cuda.mem_get_info()[0]
         elapsed_time = end_time - start_time