[Core] Update outlines and increase its threadpool size (vllm-project…

…#11140)

Signed-off-by: Russell Bryant <[email protected]>

requirements-common.txt

@@ -18,7 +18,7 @@ prometheus_client >= 0.18.0
 prometheus-fastapi-instrumentator >= 7.0.0
 tiktoken >= 0.6.0  # Required for DBRX tokenizer
 lm-format-enforcer >= 0.10.9, < 0.11
-outlines == 0.1.9
+outlines == 0.1.11
 xgrammar >= 0.1.6; platform_machine == "x86_64"
 typing_extensions >= 4.10
 filelock >= 3.10.4 # filelock starts to support `mode` argument from 3.10.4

vllm/model_executor/guided_decoding/outlines_decoding.py

@@ -1,5 +1,6 @@
 import asyncio
 import concurrent.futures
+import os
 from enum import Enum
 from json import dumps as json_dumps
 from re import escape as regex_escape
@@ -48,6 +49,11 @@ class GuidedDecodingMode(Enum):
 
 global_thread_pool = None  # used for generating logits processor fsm
 
+# It's not yet clear that using more provides a benefit, and it could
+# potentially starve other processes on the machine. We'll cap this for now and
+# adjust later if testing proves it to help overcome a bottleneck.
+_MAX_THREADPOOL_WORKERS = 16
+
 
 async def get_outlines_guided_decoding_logits_processor(
     guided_params: GuidedDecodingParams, tokenizer: PreTrainedTokenizerBase
@@ -65,8 +71,11 @@ async def get_outlines_guided_decoding_logits_processor(
         return None
 
     if global_thread_pool is None:
+        max_workers = os.cpu_count() or 2
+        if max_workers > _MAX_THREADPOOL_WORKERS:
+            max_workers = _MAX_THREADPOOL_WORKERS
         global_thread_pool = concurrent.futures.ThreadPoolExecutor(
-            max_workers=2)
+            max_workers=max_workers)
     loop = asyncio.get_running_loop()
 
     return await loop.run_in_executor(global_thread_pool,