Improvement in config

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

@@ -1,6 +1,6 @@
 services:
   trinity-node-1:
-    image: trinity-rft-megatron:latest-unittest
+    image: trinity-rft-unittest:20250918
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --head --dashboard-host 0.0.0.0 --include-dashboard true --block"
     environment:
@@ -28,7 +28,7 @@ services:
             capabilities: [gpu]
 
   trinity-node-2:
-    image: trinity-rft-megatron:latest-unittest
+    image: trinity-rft-unittest:20250918
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --address=trinity-node-1:6379 --block"
     environment:

benchmark/config/countdown-template.yaml

@@ -40,7 +40,6 @@ buffer:
     experience_buffer:
       name: experience_buffer
       storage_type: queue
-      path: ''
       use_priority_queue: true
       replay_buffer_kwargs:
         priority_fn: linear_decay
@@ -133,7 +132,6 @@ trainer:
       default_hdfs_dir: null
       remove_previous_ckpt_in_save: false
       del_local_ckpt_after_load: false
-      val_before_train: false
       max_actor_ckpt_to_keep: null
       max_critic_ckpt_to_keep: null
     critic:

benchmark/config/gsm8k-template.yaml

@@ -45,7 +45,6 @@ buffer:
     experience_buffer:
       name: experience_buffer
       storage_type: queue
-      path: ''
       use_priority_queue: true
       replay_buffer_kwargs:
         priority_fn: linear_decay
@@ -131,7 +130,6 @@ trainer:
       default_hdfs_dir: null
       remove_previous_ckpt_in_save: false
       del_local_ckpt_after_load: false
-      val_before_train: false
       max_actor_ckpt_to_keep: null
       max_critic_ckpt_to_keep: null
 monitor:

docs/sphinx_doc/source/tutorial/trinity_configs.md

@@ -617,7 +617,6 @@ trainer:
   default_hdfs_dir: null
   remove_previous_ckpt_in_save: False
   del_local_ckpt_after_load: False
-  val_before_train: False
   max_actor_ckpt_to_keep: 5
   max_critic_ckpt_to_keep: 5
 ```

docs/sphinx_doc/source_zh/tutorial/trinity_configs.md

@@ -616,7 +616,6 @@ trainer:
   default_hdfs_dir: null
   remove_previous_ckpt_in_save: False
   del_local_ckpt_after_load: False
-  val_before_train: False
   max_actor_ckpt_to_keep: 5
   max_critic_ckpt_to_keep: 5
 ```

pyproject.toml

@@ -56,7 +56,7 @@ agent = [
     "agentscope"
 ]
 rm_gallery = [
-    "rm-gallery>=0.1.1"
+    "rm-gallery>=0.1.5"
 ]
 dev = [
     "pre-commit>=2.17.0",
@@ -123,6 +123,3 @@ known_third_party = ["wandb"]
 [project.urls]
 "Homepage" = "https://github.com/modelscope/Trinity-RFT"
 "Documentation" = "https://modelscope.github.io/Trinity-RFT/"
-
-[tool.uv]
-override-dependencies=["math_verify>=0.8.0"]  # rm-gallery requires math_verify<0.8.0 which is not compatible with trinity-rft

scripts/docker/Dockerfile

@@ -15,7 +15,7 @@ RUN apt update && apt install -y \
     python3 python3-pip python3-dev python3-packaging \
     libomp-dev infiniband-diags libibverbs-dev librdmacm-dev rdma-core perftest \
     && rm -rf /var/lib/apt/lists/* \
-    && ln -sf /usr/bin/python3 /usr/bin/python
+    && ln -sf /usr/bin/python3 /usr/bin/python \
     && ln -sf /usr/bin/pip3 /usr/bin/pip
 
 

scripts/docker_for_megatron/Dockerfile

@@ -18,7 +18,7 @@ RUN apt update && apt install -y \
     python3 python3-pip python3-dev python3-packaging \
     libomp-dev infiniband-diags libibverbs-dev librdmacm-dev rdma-core perftest \
     && rm -rf /var/lib/apt/lists/* \
-    && ln -sf /usr/bin/python3 /usr/bin/python
+    && ln -sf /usr/bin/python3 /usr/bin/python \
     && ln -sf /usr/bin/pip3 /usr/bin/pip
 
 # For Aliyun users: update pip mirror to aliyun to speed up pip install

tests/common/vllm_test.py

@@ -121,7 +121,7 @@ def setUp(self):
         pprint(self.config)
         self.engines, self.auxiliary_engines = create_inference_models(self.config)
         self.model_wrapper = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=self.enable_history
+            self.engines[0], engine_type="vllm", enable_history=self.enable_history
         )
 
     async def test_generate(
@@ -240,7 +240,7 @@ def setUp(self):
         self.config.check_and_update()
 
         self.engines, self.auxiliary_engines = create_inference_models(self.config)
-        self.model_wrapper = ModelWrapper(self.engines[0], model_type="vllm", enable_history=True)
+        self.model_wrapper = ModelWrapper(self.engines[0], engine_type="vllm", enable_history=True)
 
     def test_model_len(self):
         messages = [
@@ -277,7 +277,7 @@ def setUp(self):
         self.config = get_template_config()
         self.config.mode = "explore"
         self.config.model.model_path = get_model_path()
-        self.config.explorer.rollout_model.engine_type = "vllm_async"
+        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
@@ -286,11 +286,9 @@ def setUp(self):
 
         self.config.check_and_update()
         self.engines, self.auxiliary_engines = create_inference_models(self.config)
-        self.model_wrapper = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=True
-        )
+        self.model_wrapper = ModelWrapper(self.engines[0], engine_type="vllm", enable_history=True)
         self.model_wrapper_no_history = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=False
+            self.engines[0], engine_type="vllm", enable_history=False
         )
 
     def test_api(self):
@@ -348,7 +346,7 @@ def setUp(self):
         self.config = get_template_config()
         self.config.mode = "explore"
         self.config.model.model_path = get_model_path()
-        self.config.explorer.rollout_model.engine_type = "vllm_async"
+        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
@@ -357,11 +355,9 @@ def setUp(self):
 
         self.config.check_and_update()
         self.engines, self.auxiliary_engines = create_inference_models(self.config)
-        self.model_wrapper = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=True
-        )
+        self.model_wrapper = ModelWrapper(self.engines[0], engine_type="vllm", enable_history=True)
         self.model_wrapper_no_history = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=False
+            self.engines[0], engine_type="vllm", enable_history=False
         )
 
     async def test_api_async(self):
@@ -537,7 +533,7 @@ def setUp(self):
         self.config = get_template_config()
         self.config.mode = "explore"
         self.config.model.model_path = get_api_model_path()
-        self.config.explorer.rollout_model.engine_type = "vllm_async"
+        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
@@ -551,11 +547,9 @@ def setUp(self):
 
         self.config.check_and_update()
         self.engines, self.auxiliary_engines = create_inference_models(self.config)
-        self.model_wrapper = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=True
-        )
+        self.model_wrapper = ModelWrapper(self.engines[0], engine_type="vllm", enable_history=True)
         self.model_wrapper_no_history = ModelWrapper(
-            self.engines[0], model_type="vllm_async", enable_history=False
+            self.engines[0], engine_type="vllm", enable_history=False
         )
 
     def test_api_tool_calls(self):

tests/template/verl_config.yaml

@@ -92,6 +92,5 @@ trainer:
   default_hdfs_dir: null
   remove_previous_ckpt_in_save: False
   del_local_ckpt_after_load: False
-  val_before_train: False
   max_actor_ckpt_to_keep: 1
   max_critic_ckpt_to_keep: 1

trinity/buffer/storage/file.py

@@ -34,7 +34,7 @@ class FileStorage:
     """
 
     def __init__(self, storage_config: StorageConfig, config: BufferConfig) -> None:
-        if storage_config.path is None:
+        if not storage_config.path:
             storage_config.path = default_storage_path(storage_config, config)
         ext = os.path.splitext(storage_config.path)[-1]
         if ext != ".jsonl" and ext != ".json":

trinity/buffer/storage/queue.py

@@ -219,7 +219,7 @@ def __init__(self, storage_config: StorageConfig, config: BufferConfig) -> None:
         self.queue = QueueBuffer.get_queue(storage_config, config)
         st_config = deepcopy(storage_config)
         st_config.wrap_in_ray = False
-        if st_config.path is not None:
+        if st_config.path:
             if is_database_url(st_config.path):
                 from trinity.buffer.writer.sql_writer import SQLWriter