Skip to content

feat: ms-swift格式数据增加格式校验 #522

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
shijinpjlab merged 2 commits into from
Oct 27, 2026
Merged

feat: ms-swift格式数据增加格式校验 #522

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
da5b2d4
feat: ms-swift格式数据增加格式校验
shijinpjlab Oct 27, 2026
e070ff5
feat: lint
shijinpjlab Oct 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion AGENTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,8 @@ dingo/
│ │ ├── dataset/ ← Dataset implementations per source
│ │ ├── searcher.py ← Searcher protocol + SearchResult + registry (mock, ES)
│ │ └── converter/ ← Format converters (JSON, JSONL, CSV, Parquet, MinerU, etc.)
│ │ ├── ms_swift.py ← ms-swift evaluator-field preparation
│ │ └── ms_swift_schema.py ← Full standardized-delivery validation
│ ├── model/
│ │ ├── model.py ← Model registry (rule_register, llm_register)
│ │ ├── rule/ ← Rule-based evaluators (80+ built-in)
Expand Down Expand Up @@ -171,7 +173,9 @@ input_data = {

MinerU formats automatically flatten blocks into `Data` objects, preserving all original fields (type, bbox, page_idx, etc.) via Pydantic `extra="allow"`. Use `dataset.mineru_config.include_types` to filter by block type.

The `ms_swift` converter lives in `dingo/data/converter/ms_swift.py`. It prepares
The `ms_swift` converter lives in `dingo/data/converter/ms_swift.py` and validates
each record against the full standardized delivery contract in
`dingo/data/converter/ms_swift_schema.py` before preparing any derived fields. It prepares
independent `<EvaluatorName>_<field>` fields for all 9 deterministic Agent rules
and 7 Agent LLM metrics, only when selected in `evaluator[].evals`.
Each deterministic rule owns its parsing and field assembly in its `prepare_*`
Expand Down
176 changes: 45 additions & 131 deletions dingo/data/converter/ms_swift.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,7 @@

from dingo.config import InputArgs
from dingo.data.converter.base import BaseConverter
from dingo.data.converter.ms_swift_schema import validate_ms_swift_record
from dingo.io import Data


Expand All @@ -22,6 +23,19 @@ def _messages_content(record: Dict[str, Any]) -> str:
return json.dumps(messages, ensure_ascii=False)


def _tool_call_name_and_arguments(call: Dict[str, Any]):
"""Read fields already guaranteed by the standardized-format validator."""
function = call["function"]
arguments = function["arguments"]
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError:
# A string is itself a valid JSON value under the standard.
pass
return function["name"], arguments


def prepare_llm_agent_task_completion(record: Dict[str, Any]) -> Dict[str, str]:
"""Prepare the initial conversation-only task completion input."""
return {
Expand All @@ -41,22 +55,8 @@ def prepare_rule_agent_trace_loop_detection(record: Dict[str, Any]) -> Dict[str,
tool_calls = []
if not isinstance(tool_calls, list):
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
for call in tool_calls:
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
calls.append(item)
return {
Expand Down Expand Up @@ -85,22 +85,8 @@ def prepare_rule_agent_trace_destructive_action(record: Dict[str, Any]) -> Dict[
tool_calls = []
if not isinstance(tool_calls, list):
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
for call in tool_calls:
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
calls.append(item)
return {
Expand Down Expand Up @@ -131,20 +117,7 @@ def prepare_rule_agent_trace_gateway_bypass(record: Dict[str, Any]) -> Dict[str,
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
call_id = call.get("id")
observation = {}
Expand Down Expand Up @@ -192,20 +165,7 @@ def prepare_rule_agent_trace_integrity(record: Dict[str, Any]) -> Dict[str, str]
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
name, arguments = _tool_call_name_and_arguments(call)
tool_spans_recorded += 1
call_id = call.get("id")
if isinstance(call_id, str) and call_id:
Expand Down Expand Up @@ -248,20 +208,7 @@ def prepare_rule_agent_trace_latency_anomaly(record: Dict[str, Any]) -> Dict[str
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
name, arguments = _tool_call_name_and_arguments(call)
step = {"name": name}
call_id = call.get("id")
observation = {}
Expand All @@ -273,9 +220,16 @@ def prepare_rule_agent_trace_latency_anomaly(record: Dict[str, Any]) -> Dict[str
# Copy recorded timings and groups, preferring matched tool results.
for key in ("duration", "duration_seconds", "group", "type"):
if key in observation:
step[key] = observation[key]
value = observation[key]
elif key in call:
step[key] = call[key]
value = call[key]
else:
continue
# Standard ms-swift reserves call.type="function" for the
# invocation kind; it is not a latency peer-group label.
if key == "type" and value == "function":
continue
step[key] = value
steps.append(step)
return {"RuleAgentTraceLatencyAnomaly_content": json.dumps({"steps": steps}, ensure_ascii=False)}

Expand All @@ -291,22 +245,8 @@ def prepare_rule_agent_trace_sandbox_explicitly_disabled(record: Dict[str, Any])
tool_calls = []
if not isinstance(tool_calls, list):
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
for call in tool_calls:
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
calls.append(item)
return {
Expand Down Expand Up @@ -337,20 +277,7 @@ def prepare_rule_agent_trace_secret_exposure(record: Dict[str, Any]) -> Dict[str
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
call_id = call.get("id")
observation = {}
Expand Down Expand Up @@ -387,22 +314,8 @@ def prepare_rule_agent_trace_sensitive_access(record: Dict[str, Any]) -> Dict[st
tool_calls = []
if not isinstance(tool_calls, list):
raise ValueError(f"ms_swift messages[{message_index}].tool_calls must be a list")
for call_index, call in enumerate(tool_calls):
path = f"ms_swift messages[{message_index}].tool_calls[{call_index}].function"
function = call.get("function") if isinstance(call, dict) else None
if not isinstance(function, dict):
raise ValueError(f"{path} must be an object")
name = function.get("name")
if not isinstance(name, str) or not name:
raise ValueError(f"{path}.name must be a non-empty string")
arguments = function.get("arguments", {})
if isinstance(arguments, str):
try:
arguments = json.loads(arguments)
except json.JSONDecodeError as error:
raise ValueError(f"{path}.arguments must contain a JSON object") from error
if not isinstance(arguments, dict):
raise ValueError(f"{path}.arguments must be an object")
for call in tool_calls:
name, arguments = _tool_call_name_and_arguments(call)
item = {"tool_name": name, "args": arguments}
calls.append(item)
return {
Expand Down Expand Up @@ -493,15 +406,16 @@ def convertor(cls, input_args: InputArgs) -> Callable:
preparers = [prepare for name, prepare in _PREPARERS.items() if name in selected_names]

def _convert(raw: Union[str, Dict[str, Any]]) -> Data:
record = json.loads(raw) if isinstance(raw, str) else raw
if not isinstance(record, dict):
raise ValueError("ms_swift record must be a JSON object")
messages = record.get("messages")
if not isinstance(messages, list):
raise ValueError("ms_swift messages must be a list")
for index, message in enumerate(messages):
if not isinstance(message, dict):
raise ValueError(f"ms_swift messages[{index}] must be an object")
if isinstance(raw, str):
try:
record = json.loads(raw)
except json.JSONDecodeError as error:
raise ValueError(
f"ms_swift record is malformed: invalid JSON ({error.msg})"
) from error
else:
record = raw
validate_ms_swift_record(record)

converted = dict(record)
derived_fields = set()
Expand Down
Loading
Loading