[V1] Fix profiling.py

Signed-off-by: Alexander Matveev <alexm@neuralmagic.com>

examples/offline_inference/profiling.py

@@ -12,7 +12,7 @@ from typing import Any, Optional, TypeAlias
 import torch
 import tqdm
 
-from vllm import LLM, SamplingParams
+from vllm import LLM, SamplingParams, envs
 from vllm.engine.arg_utils import EngineArgs
 from vllm.profiler import layerwise_profile
 from vllm.utils import FlexibleArgumentParser
@@ -261,8 +261,13 @@ def run_profile(context: ProfileContext, csv_output: Optional[str],
 
     decode_profs = []
     for _ in tqdm.tqdm(range(num_steps_to_profile - 1)):
-        num_running_seqs = llm.llm_engine.scheduler[
+        if envs.VLLM_USE_V1:
-            0].get_num_unfinished_seq_groups()
+            num_running_seqs = llm.llm_engine.scheduler[
+                0].get_num_unfinished_requests()
+        else:
+            num_running_seqs = llm.llm_engine.scheduler[
+                0].get_num_unfinished_seq_groups()
+
         with layerwise_profile(
                 num_running_seqs=num_running_seqs) as decode_prof:
             llm.llm_engine.step()

vllm/v1/engine/llm_engine.py

@@ -69,6 +69,8 @@ class LLMEngine:
             self.dp_group = None
         self.should_execute_dummy_batch = False
 
+        self.scheduler_config = vllm_config.scheduler_config
+
         # Tokenizer (+ ensure liveness if running in another process).
         self.tokenizer = init_tokenizer_from_configs(
             model_config=vllm_config.model_config,
@@ -98,6 +100,9 @@ class LLMEngine:
         if not multiprocess_mode:
             # for v0 compatibility
             self.model_executor = self.engine_core.engine_core.model_executor  # type: ignore
+            self.scheduler = [
+                self.engine_core.engine_core.scheduler  # type: ignore
+            ]  # type: ignore
 
     @classmethod
     def from_vllm_config(