Revert "[V1][Core] Fix memory issue with logits & sampling (#13776)"

This reverts commit 8d5aa46.

Author: robertgshaw2-redhat

tests/basic_correctness/test_cumem.py

@@ -142,16 +142,7 @@ def test_end_to_end(model: str, use_v1: bool):
     used_bytes = total - free_gpu_bytes_after_sleep - used_bytes_baseline
     # now the memory usage is mostly cudagraph memory pool,
     # and it should be less than the model weights (1B model, 2GiB weights)
-
-    # NOTE: In V1, the memory buffer for logits (max_num_reqs x vocab_size)
-    # is captured but cannot be releasesd from PyTorch due to a known bug,
-    # therefore high memory usage after `llm.sleep` is called is expected.
-    # FIXME(youkaichao & ywang96): Fix memory buffer issue with sleep mode
-    # in V1.
-    if use_v1:
-        assert used_bytes < 7 * GiB_bytes
-    else:
-        assert used_bytes < 2 * GiB_bytes
+    assert used_bytes < 2 * GiB_bytes
 
     llm.wake_up()
     output2 = llm.generate(prompt, sampling_params)

vllm/v1/worker/gpu_model_runner.py

@@ -1238,42 +1238,6 @@ def _dummy_run(
             )
         return hidden_states
 
-    @torch.inference_mode()
-    def _dummy_sampler_run(
-        self,
-        hidden_states: torch.Tensor,
-    ) -> torch.Tensor:
-
-        logits = self.model.compute_logits(hidden_states, None)
-        num_reqs = logits.size(0)
-
-        dummy_tensors = lambda v: torch.full(
-            (num_reqs, ), v, device=self.device)
-
-        dummy_metadata = SamplingMetadata(
-            temperature=dummy_tensors(0.5),
-            all_greedy=False,
-            all_random=False,
-            top_p=dummy_tensors(0.9),
-            top_k=dummy_tensors(logits.size(1) - 1),
-            min_p=None,
-            generators={},
-            max_num_logprobs=None,
-            no_penalties=True,
-            prompt_token_ids=None,
-            frequency_penalties=dummy_tensors(0.1),
-            presence_penalties=dummy_tensors(0.1),
-            repetition_penalties=dummy_tensors(0.1),
-            output_token_ids=[[] for _ in range(num_reqs)],
-            min_tokens={},
-            logit_bias=[None for _ in range(num_reqs)],
-            allowed_token_ids_mask=None,
-        )
-        sampler_output = self.model.sample(logits=logits,
-                                           sampling_metadata=dummy_metadata)
-
-        return sampler_output
-
     def profile_run(self) -> None:
         # Profile with multimodal encoder & encoder cache.
         # TODO: handle encoder-decoder models once we support them.
@@ -1389,11 +1353,37 @@ def profile_run(self) -> None:
             hidden_states = self._dummy_run(self.max_num_tokens)
             if get_pp_group().is_last_rank:
                 hidden_states = hidden_states[logit_indices]
-                sampler_output = self._dummy_sampler_run(hidden_states)
+                logits = self.model.compute_logits(hidden_states, None)
+                dummy_tensors = lambda v: torch.full(
+                    (num_reqs, ), v, device=self.device)
+                dummy_metadata = SamplingMetadata(
+                    temperature=dummy_tensors(0.5),
+                    all_greedy=False,
+                    all_random=False,
+                    top_p=dummy_tensors(0.9),
+                    top_k=dummy_tensors(logits.size(1) - 1),
+                    min_p=None,
+                    generators={},
+                    max_num_logprobs=None,
+                    no_penalties=True,
+                    prompt_token_ids=torch.ones_like(logits,
+                                                     dtype=torch.int64),
+                    frequency_penalties=dummy_tensors(0.1),
+                    presence_penalties=dummy_tensors(0.1),
+                    repetition_penalties=dummy_tensors(0.1),
+                    output_token_ids=[[] for _ in range(num_reqs)],
+                    min_tokens={},
+                    logit_bias=[None for _ in range(num_reqs)],
+                    allowed_token_ids_mask=None,
+                )
+                sampler_output = self.model.sample(
+                    logits=logits, sampling_metadata=dummy_metadata)
             else:
+                logits = None
                 sampler_output = None
+                dummy_metadata = None
             torch.cuda.synchronize()
-            del hidden_states, sampler_output
+            del hidden_states, logits, sampler_output, dummy_metadata
             self.encoder_cache.clear()
         gc.collect()
 

vllm/v1/worker/gpu_worker.py

@@ -119,8 +119,6 @@ def init_device(self):
         self.model_runner: GPUModelRunner = GPUModelRunner(
             self.vllm_config, self.device)
 
-    # FIXME(youkaichao & ywang96): Use TorchDispatchMode instead of memory pool
-    # to hijack tensor allocation.
     def load_model(self) -> None:
         if self.vllm_config.model_config.enable_sleep_mode:
             allocator = CuMemAllocator.get_instance()
@@ -213,25 +211,6 @@ def compile_or_warm_up_model(self) -> None:
             self.model_runner._dummy_run(size)
         if not self.model_config.enforce_eager:
             self.model_runner.capture_model()
-
-        # Warm up sampler and preallocate memory buffer for logits and other
-        # sampling related tensors of max possible shape to avoid memory
-        # fragmentation issue.
-        # NOTE: This is called after `capture_model` on purpose to prevent
-        # memory buffers from being cleared by `torch.cuda.empty_cache`.
-        try:
-            self.model_runner._dummy_sampler_run(
-                hidden_states=self.model_runner._dummy_run(
-                    num_tokens=self.scheduler_config.max_num_seqs))
-        except RuntimeError as e:
-            if 'out of memory' in str(e):
-                raise RuntimeError(
-                    "CUDA out of memory occurred when warming up sampler. "
-                    "Please try lowering `gpu_memory_utilization` when "
-                    "initializing the engine.") from None
-            else:
-                raise e
-
         # Reset the seed to ensure that the random state is not affected by
         # the model initialization and profiling.
         set_random_seed(self.model_config.seed)