Update vLLM to 0.11

.github/workflows/docker/docker-compose.yaml

@@ -1,6 +1,6 @@
 services:
   trinity-node-1:
-    image: trinity-rft-unittest:20250924
+    image: trinity-rft-unittest:20251029
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --head --dashboard-host 0.0.0.0 --include-dashboard true --block"
     environment:
@@ -29,7 +29,7 @@ services:
             capabilities: [gpu]
 
   trinity-node-2:
-    image: trinity-rft-unittest:20250924
+    image: trinity-rft-unittest:20251029
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --address=trinity-node-1:6379 --block"
     environment:

benchmark/config/gsm8k-template.yaml

@@ -59,7 +59,6 @@ explorer:
   rollout_model:
     engine_num: 2
     tensor_parallel_size: 1
-    use_v1: true
     enforce_eager: false
     enable_prefix_caching: false
     enable_chunked_prefill: false

pyproject.toml

@@ -23,7 +23,7 @@ requires-python = ">=3.10,<3.13"
 dependencies = [
     "verl==0.5.0",
     "ray[default]>=2.48.0",
-    "vllm>=0.9.1,<=0.10.2",
+    "vllm>=0.9.1,<=0.11.0",
     "tensordict",
     "wandb",
     "omegaconf",

tests/common/vllm_test.py

@@ -276,7 +276,6 @@ def setUp(self):
         self.config.explorer.rollout_model.engine_type = "vllm"
         self.config.explorer.rollout_model.engine_num = 1
         self.config.explorer.rollout_model.tensor_parallel_size = 1
-        self.config.explorer.rollout_model.use_v1 = True
         self.config.explorer.rollout_model.chat_template = CHAT_TEMPLATE
         self.config.explorer.rollout_model.enable_openai_api = True
 
@@ -368,7 +367,6 @@ def setUp(self):
         self.config.explorer.rollout_model.engine_type = "vllm"
         self.config.explorer.rollout_model.engine_num = 1
         self.config.explorer.rollout_model.tensor_parallel_size = 1
-        self.config.explorer.rollout_model.use_v1 = True
         self.config.explorer.rollout_model.chat_template = CHAT_TEMPLATE
         self.config.explorer.rollout_model.enable_openai_api = True
 
@@ -578,7 +576,6 @@ def setUp(self):
         self.config.explorer.rollout_model.engine_type = "vllm"
         self.config.explorer.rollout_model.engine_num = 1
         self.config.explorer.rollout_model.tensor_parallel_size = 1
-        self.config.explorer.rollout_model.use_v1 = True
         self.config.explorer.rollout_model.chat_template = CHAT_TEMPLATE
         self.config.explorer.rollout_model.enable_openai_api = True
         # added for toolcalls

tests/template/config.yaml

@@ -42,7 +42,6 @@ explorer:
     enforce_eager: true
     dtype: bfloat16
     seed: 42
-    use_v1: true
 trainer:
   trainer_type: verl
   save_interval: 100

trinity/common/models/api/vllm_patch.py

@@ -347,10 +347,10 @@ async def run_api_server_in_ray_actor(
     reasoning_parser: Optional[str] = None,
 ):
     vllm_version = get_vllm_version()
-    if vllm_version < parse_version("0.8.5") or vllm_version > parse_version("0.10.2"):
+    if vllm_version < parse_version("0.8.5") or vllm_version > parse_version("0.11.0"):
         raise ValueError(
             f"Unsupported vllm version: {vllm.__version__}. "
-            "This patch requires vllm version >= 0.8.5, <= 0.10.2."
+            "This patch requires vllm version >= 0.8.5, <= 0.11.0."
         )
 
     parser = FlexibleArgumentParser(description="Run the OpenAI API server.")