deepdata: core product code (backend, frontend, runtime, deepflow module)

deepflow/src/prose/graph/builder.py

@@ -0,0 +1,68 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import asyncio
+import logging
+
+from langgraph.graph import END, START, StateGraph
+
+from src.prose.graph.prose_continue_node import prose_continue_node
+from src.prose.graph.prose_fix_node import prose_fix_node
+from src.prose.graph.prose_improve_node import prose_improve_node
+from src.prose.graph.prose_longer_node import prose_longer_node
+from src.prose.graph.prose_shorter_node import prose_shorter_node
+from src.prose.graph.prose_zap_node import prose_zap_node
+from src.prose.graph.state import ProseState
+
+
+def optional_node(state: ProseState):
+    return state["option"]
+
+
+def build_graph():
+    """Build and return the ppt workflow graph."""
+    # build state graph
+    builder = StateGraph(ProseState)
+    builder.add_node("prose_continue", prose_continue_node)
+    builder.add_node("prose_improve", prose_improve_node)
+    builder.add_node("prose_shorter", prose_shorter_node)
+    builder.add_node("prose_longer", prose_longer_node)
+    builder.add_node("prose_fix", prose_fix_node)
+    builder.add_node("prose_zap", prose_zap_node)
+    builder.add_conditional_edges(
+        START,
+        optional_node,
+        {
+            "continue": "prose_continue",
+            "improve": "prose_improve",
+            "shorter": "prose_shorter",
+            "longer": "prose_longer",
+            "fix": "prose_fix",
+            "zap": "prose_zap",
+        },
+        END,
+    )
+    return builder.compile()
+
+
+async def _test_workflow():
+    workflow = build_graph()
+    events = workflow.astream(
+        {
+            "content": "The weather in Beijing is sunny",
+            "option": "continue",
+        },
+        stream_mode="messages",
+        subgraphs=True,
+    )
+    async for node, event in events:
+        e = event[0]
+        print({"id": e.id, "object": "chat.completion.chunk", "content": e.content})
+
+
+if __name__ == "__main__":
+    from dotenv import load_dotenv
+
+    load_dotenv()
+    logging.basicConfig(level=logging.INFO)
+    asyncio.run(_test_workflow())

deepflow/src/prose/graph/prose_continue_node.py

@@ -0,0 +1,25 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_continue_node(state: ProseState):
+    logger.info("Generating prose continue content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_continue")),
+            HumanMessage(content=state["content"]),
+        ],
+    )
+    return {"output": prose_content.content}

deepflow/src/prose/graph/prose_fix_node.py

@@ -0,0 +1,26 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_fix_node(state: ProseState):
+    logger.info("Generating prose fix content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_fix")),
+            HumanMessage(content=f"The existing text is: {state['content']}"),
+        ],
+    )
+    logger.info(f"prose_content: {prose_content}")
+    return {"output": prose_content.content}

deepflow/src/prose/graph/prose_improve_node.py

@@ -0,0 +1,26 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_improve_node(state: ProseState):
+    logger.info("Generating prose improve content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_improver")),
+            HumanMessage(content=f"The existing text is: {state['content']}"),
+        ],
+    )
+    logger.info(f"prose_content: {prose_content}")
+    return {"output": prose_content.content}

deepflow/src/prose/graph/prose_longer_node.py

@@ -0,0 +1,26 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_longer_node(state: ProseState):
+    logger.info("Generating prose longer content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_longer")),
+            HumanMessage(content=f"The existing text is: {state['content']}"),
+        ],
+    )
+    logger.info(f"prose_content: {prose_content}")
+    return {"output": prose_content.content}

deepflow/src/prose/graph/prose_shorter_node.py

@@ -0,0 +1,26 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_shorter_node(state: ProseState):
+    logger.info("Generating prose shorter content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_shorter")),
+            HumanMessage(content=f"The existing text is: {state['content']}"),
+        ],
+    )
+    logger.info(f"prose_content: {prose_content}")
+    return {"output": prose_content.content}

deepflow/src/prose/graph/prose_zap_node.py

@@ -0,0 +1,28 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+import logging
+
+from langchain.schema import HumanMessage, SystemMessage
+
+from src.config.agents import AGENT_LLM_MAP
+from src.llms.llm import get_llm_by_type
+from src.prompts.template import get_prompt_template
+from src.prose.graph.state import ProseState
+
+logger = logging.getLogger(__name__)
+
+
+def prose_zap_node(state: ProseState):
+    logger.info("Generating prose zap content...")
+    model = get_llm_by_type(AGENT_LLM_MAP["prose_writer"])
+    prose_content = model.invoke(
+        [
+            SystemMessage(content=get_prompt_template("prose/prose_zap")),
+            HumanMessage(
+                content=f"For this text: {state['content']}.\nYou have to respect the command: {state['command']}"
+            ),
+        ],
+    )
+    logger.info(f"prose_content: {prose_content}")
+    return {"output": prose_content.content}

deepflow/src/prose/graph/state.py

@@ -0,0 +1,20 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+# SPDX-License-Identifier: MIT
+
+from langgraph.graph import MessagesState
+
+
+class ProseState(MessagesState):
+    """State for the prose generation."""
+
+    # The content of the prose
+    content: str = ""
+
+    # Prose writer option: continue, improve, shorter, longer, fix, zap
+    option: str = ""
+
+    # The user custom command for the prose writer
+    command: str = ""
+
+    # Output
+    output: str = ""