fix comments

Author: pan-x-c

src/agentscope/tuner/_algorithm.py

@@ -18,7 +18,7 @@ class Algorithm(BaseModel):
     group_size: int = Field(
         description=(
             "The group size for algorithms "
-            "required group rollout, e.g., GRPO."
+            "requiring group rollout, e.g., GRPO."
         ),
         default=8,
     )

src/agentscope/tuner/_config.py

@@ -25,11 +25,6 @@ def to_trinity_config(
     experiment_name: str | None = None,
 ) -> Any:
     """Convert to Trinity-RFT compatible configuration."""
-    if workflow_func is not None:
-        check_workflow_function(workflow_func)
-    if judge_func is not None:
-        check_judge_function(judge_func)
-
     from trinity.common.config import (
         Config,
         TasksetConfig,
@@ -54,12 +49,19 @@ def to_trinity_config(
 
     workflow_name = "agentscope_workflow_adapter_v1"
     if train_dataset is not None:
-        config.buffer.explorer_input.taskset = TasksetConfig(
-            name="train_taskset",
-            path=train_dataset.path,
-            split=train_dataset.split,
-            subset_name=train_dataset.name,
-        )
+        if config.buffer.explorer_input.taskset is None:
+            config.buffer.explorer_input.taskset = TasksetConfig(
+                name="train_taskset",
+                path=train_dataset.path,
+                split=train_dataset.split,
+                subset_name=train_dataset.name,
+            )
+        else:
+            config.buffer.explorer_input.taskset.path = train_dataset.path
+            config.buffer.explorer_input.taskset.split = train_dataset.split
+            config.buffer.explorer_input.taskset.subset_name = (
+                train_dataset.name
+            )
         config.buffer.total_epochs = train_dataset.total_epochs
         config.buffer.total_steps = train_dataset.total_steps
     config.buffer.explorer_input.taskset.default_workflow_type = workflow_name

src/agentscope/tuner/_dataset.py

@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 """Dataset definition for tuner."""
-from typing import Optional, Dict
+from itertools import islice
+from typing import Optional, List
 from pydantic import BaseModel, Field
 
 
@@ -34,7 +35,7 @@ class Dataset(BaseModel):
         default=None,
     )
 
-    def preview(self, n: int = 5) -> Dict:
+    def preview(self, n: int = 5) -> List:
         """Preview the dataset information.
 
         Args:
@@ -53,7 +54,8 @@ def preview(self, n: int = 5) -> Dict:
             path=self.path,
             name=self.name,
             split=self.split,
+            streaming=True,
         )
-        samples = ds[:n]
+        samples = list(islice(ds, n))
         print(json.dumps(samples, indent=2))
         return samples

src/agentscope/tuner/_tune.py

@@ -4,7 +4,11 @@
 from ._judge import JudgeType
 from ._model import TunerChatModel
 from ._dataset import Dataset
-from ._config import to_trinity_config
+from ._config import (
+    to_trinity_config,
+    check_judge_function,
+    check_workflow_function,
+)
 from ._algorithm import Algorithm
 
 
@@ -35,7 +39,8 @@ def tune(
             Defaults to None.
         auxiliary_models (dict[str, TunerChatModel], optional): A
             dictionary of auxiliary chat models for LLM-as-a-Judge
-            usage. Defaults to None.
+            or acting other agents in multi-agent scenarios.
+            Defaults to None.
         algorithm (Algorithm, optional): The tuning algorithm
             configuration. Defaults to None.
         config_path (str, optional): Path to the learning configuration
@@ -49,6 +54,10 @@ def tune(
             "`pip install trinity-rft`.",
         ) from e
 
+    check_workflow_function(workflow_func)
+    if judge_func is not None:
+        check_judge_function(judge_func)
+
     config = to_trinity_config(
         config_path=config_path,
         workflow_func=workflow_func,