Ali-Bhai/toolace-ragtruth-style-hallucinations
ToolACE RAGTruth-style Tool Hallucination Dataset This dataset was built from ToolACE tool-use dialogues and converted into a RAGTruth-style format for hallucination detection in tool calling. Task Given: query: user query context: tool response output: final assistant answer the goal is to classify whether the answer is grounded in the tool output or belongs to one of three hallucination types. Labels clean tool_output_conflict overgeneration… See the full description on the dataset page: https://huggingface.co/datasets/Ali-Bhai/toolace-ragtruth-style-hallucinations.
ToolACE RAGTruth-style Tool Hallucination Dataset
This dataset was built from ToolACE tool-use dialogues and converted into a RAGTruth-style format for hallucination detection in tool calling.
Task
Given:
query: user querycontext: tool responseoutput: final assistant answer
the goal is to classify whether the answer is grounded in the tool output or belongs to one of three hallucination types.
Labels
cleantool_output_conflictovergenerationmissing_tool_action_recommendation
Synthetic hallucinated examples contain RAGTruth-style character-span labels in hallucination_labels.
Splits
The default files are the mixed clean + synthetic splits:
Split label counts
{
"train": {
"overgeneration": 276,
"clean": 834,
"missing_tool_action_recommendation": 276,
"tool_output_conflict": 276
},
"validation": {
"clean": 103,
"overgeneration": 37,
"tool_output_conflict": 37,
"missing_tool_action_recommendation": 37
},
"test": {
"clean": 96,
"missing_tool_action_recommendation": 37,
"tool_output_conflict": 37,
"overgeneration": 37
}
}Loading
from datasets import load_dataset
dataset = load_dataset("Ali-Bhai/toolace-ragtruth-style-hallucinations")
print(dataset)If automatic split detection is not available in the runtime, use:
from datasets import load_dataset
dataset = load_dataset(
"json",
data_files={
"train": "hf://datasets/Ali-Bhai/toolace-ragtruth-style-hallucinations/train.jsonl",
"validation": "hf://datasets/Ali-Bhai/toolace-ragtruth-style-hallucinations/validation.jsonl",
"test": "hf://datasets/Ali-Bhai/toolace-ragtruth-style-hallucinations/test.jsonl",
},
)Source and construction
The clean records were extracted from ToolACE dialogues with exact tool-call/tool-output matching. Synthetic hallucinations were injected using three controlled policies:
- numeric/date tool-output conflict,
- natural unsupported overgeneration,
- recommendation of an unavailable tool.
See metadata/dataset_manifest.json for the full file manifest.
