youngkingdom/dify
1
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-05-05 09:58:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

vibe: implement file structured output

Browse Source
This commit is contained in:
Stream
2026-02-01 02:47:28 +08:00
parent b6465327c1
commit b66db183c9
8 changed files with 554 additions and 794 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
api/core/workflow/nodes/llm/llm_utils.py
View File

@ -257,8 +257,8 @@ def _build_file_descriptions(files: Sequence[Any]) -> str:
"""
Build a text description of generated files for inclusion in context.
The description includes file_id which can be used by subsequent nodes
to reference the files via structured output.
The description includes file_id for context; structured output file paths
are only supported in sandbox mode.
"""
if not files:
return ""

78
api/core/workflow/nodes/llm/node.py
View File

@ -21,7 +21,11 @@ from core.app_assets.constants import AppAssetsAttrs
from core.file import FileTransferMethod, FileType, file_manager
from core.helper.code_executor import CodeExecutor, CodeLanguage
from core.llm_generator.output_parser.errors import OutputParserError
from core.llm_generator.output_parser.file_ref import convert_file_refs_in_output
from core.llm_generator.output_parser.file_ref import (
adapt_schema_for_sandbox_file_paths,
convert_sandbox_file_paths_in_output,
detect_file_path_fields,
)
from core.llm_generator.output_parser.structured_output import invoke_llm_with_structured_output
from core.memory.base import BaseMemory
from core.model_manager import ModelInstance, ModelManager
@ -297,13 +301,20 @@ class LLMNode(Node[LLMNodeData]):
)
structured_output_schema: Mapping[str, Any] | None
structured_output_file_paths: list[str] = []
if self.node_data.structured_output_enabled:
if not self.node_data.structured_output:
raise ValueError("structured_output_enabled is True but structured_output is not set")
structured_output_schema = LLMNode.fetch_structured_output_schema(
structured_output=self.node_data.structured_output
)
raw_schema = LLMNode.fetch_structured_output_schema(structured_output=self.node_data.structured_output)
if self.node_data.computer_use:
structured_output_schema, structured_output_file_paths = adapt_schema_for_sandbox_file_paths(
raw_schema
)
else:
if detect_file_path_fields(raw_schema):
raise LLMNodeError("Structured output file paths are only supported in sandbox mode.")
structured_output_schema = raw_schema
else:
structured_output_schema = None
@ -319,8 +330,10 @@ class LLMNode(Node[LLMNodeData]):
stop=stop,
variable_pool=variable_pool,
tool_dependencies=tool_dependencies,
structured_output_schema=structured_output_schema
structured_output_schema=structured_output_schema,
structured_output_file_paths=structured_output_file_paths,
)
elif self.tool_call_enabled:
generator = self._invoke_llm_with_tools(
model_instance=model_instance,
@ -330,7 +343,7 @@ class LLMNode(Node[LLMNodeData]):
variable_pool=variable_pool,
node_inputs=node_inputs,
process_data=process_data,
structured_output_schema=structured_output_schema
structured_output_schema=structured_output_schema,
)
else:
# Use traditional LLM invocation
@ -532,8 +545,8 @@ class LLMNode(Node[LLMNodeData]):
model_parameters=node_data_model.completion_params,
stop=list(stop or []),
user=user_id,
tenant_id=tenant_id,
)
else:
request_start_time = time.perf_counter()
@ -1880,7 +1893,7 @@ class LLMNode(Node[LLMNodeData]):
variable_pool: VariablePool,
node_inputs: dict[str, Any],
process_data: dict[str, Any],
structured_output_schema: Mapping[str, Any] | None
structured_output_schema: Mapping[str, Any] | None,
) -> Generator[NodeEventBase, None, LLMGenerationData]:
"""Invoke LLM with tools support (from Agent V2).
@ -1906,14 +1919,14 @@ class LLMNode(Node[LLMNodeData]):
files=prompt_files,
max_iterations=self._node_data.max_iterations or 10,
context=ExecutionContext(user_id=self.user_id, app_id=self.app_id, tenant_id=self.tenant_id),
structured_output_schema=structured_output_schema
structured_output_schema=structured_output_schema,
)
# Run strategy
outputs = strategy.run(
prompt_messages=list(prompt_messages),
model_parameters=self._node_data.model.completion_params,
stop=list(stop or [])
stop=list(stop or []),
)
result = yield from self._process_tool_outputs(outputs)
@ -1927,10 +1940,12 @@ class LLMNode(Node[LLMNodeData]):
stop: Sequence[str] | None,
variable_pool: VariablePool,
tool_dependencies: ToolDependencies | None,
structured_output_schema: Mapping[str, Any] | None
structured_output_schema: Mapping[str, Any] | None,
structured_output_file_paths: Sequence[str] | None,
) -> Generator[NodeEventBase | LLMStructuredOutput, None, LLMGenerationData]:
result: LLMGenerationData | None = None
sandbox_output_files: list[File] = []
structured_output_files: list[File] = []
# FIXME(Mairuis): Async processing for bash session.
with SandboxBashSession(sandbox=sandbox, node_id=self.id, tools=tool_dependencies) as session:
@ -1948,17 +1963,31 @@ class LLMNode(Node[LLMNodeData]):
max_iterations=self._node_data.max_iterations or 100,
agent_strategy=AgentEntity.Strategy.FUNCTION_CALLING,
context=ExecutionContext(user_id=self.user_id, app_id=self.app_id, tenant_id=self.tenant_id),
structured_output_schema=structured_output_schema
structured_output_schema=structured_output_schema,
)
outputs = strategy.run(
prompt_messages=list(prompt_messages),
model_parameters=self._node_data.model.completion_params,
stop=list(stop or [])
stop=list(stop or []),
)
result = yield from self._process_tool_outputs(outputs)
if result and result.structured_output and structured_output_file_paths:
structured_output_payload = result.structured_output.structured_output or {}
try:
converted_output, structured_output_files = convert_sandbox_file_paths_in_output(
output=structured_output_payload,
file_path_fields=structured_output_file_paths,
file_resolver=session.download_file,
)
except ValueError as exc:
raise LLMNodeError(str(exc)) from exc
result = result.model_copy(
update={"structured_output": LLMStructuredOutput(structured_output=converted_output)}
)
# Collect output files from sandbox before session ends
# Files are saved as ToolFiles with valid tool_file_id for later reference
sandbox_output_files = session.collect_output_files()
@ -1971,7 +2000,7 @@ class LLMNode(Node[LLMNodeData]):
yield structured_output
# Merge sandbox output files into result
if sandbox_output_files:
if sandbox_output_files or structured_output_files:
result = LLMGenerationData(
text=result.text,
reasoning_contents=result.reasoning_contents,
@ -1979,7 +2008,7 @@ class LLMNode(Node[LLMNodeData]):
sequence=result.sequence,
usage=result.usage,
finish_reason=result.finish_reason,
files=result.files + sandbox_output_files,
files=result.files + sandbox_output_files + structured_output_files,
trace=result.trace,
)
@ -2056,9 +2085,12 @@ class LLMNode(Node[LLMNodeData]):
structured_output=self._node_data.structured_output or {},
)
tool_instances.extend(
build_agent_output_tools(tenant_id=self.tenant_id, invoke_from=self.invoke_from,
tool_invoke_from=ToolInvokeFrom.WORKFLOW,
structured_output_schema=structured_output_schema)
build_agent_output_tools(
tenant_id=self.tenant_id,
invoke_from=self.invoke_from,
tool_invoke_from=ToolInvokeFrom.WORKFLOW,
structured_output_schema=structured_output_schema,
)
)
return tool_instances
@ -2564,15 +2596,7 @@ class LLMNode(Node[LLMNodeData]):
raise ValueError("No agent result found in tool outputs")
output_payload = agent_result.output
if isinstance(output_payload, dict):
state.aggregate.structured_output = LLMStructuredOutput(
structured_output=convert_file_refs_in_output(
output=output_payload,
json_schema=LLMNode.fetch_structured_output_schema(
structured_output=self._node_data.structured_output or {},
),
tenant_id=self.tenant_id,
)
)
state.aggregate.structured_output = LLMStructuredOutput(structured_output=output_payload)
state.aggregate.text = json.dumps(output_payload)
elif isinstance(output_payload, str):
state.aggregate.text = output_payload