feat: use output config instead of relying on leaf node type to save output

Author: ChenZiHong-Gavin

examples/extract/extract_schema_guided/schema_guided_extraction_config.yaml

@@ -30,6 +30,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true
     params:
       method: schema_guided
       schema_path: graphgen/templates/extraction/schemas/legal_contract.json

examples/generate/generate_aggregated_qa/aggregated_config.yaml

@@ -74,6 +74,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true # save output
     params:
       method: aggregated # atomic, aggregated, multi_hop, cot, vqa
       data_format: ChatML # Alpaca, Sharegpt, ChatML

examples/generate/generate_atomic_qa/atomic_config.yaml

@@ -50,6 +50,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true
     params:
       method: atomic
       data_format: Alpaca

examples/generate/generate_cot_qa/cot_config.yaml

@@ -52,6 +52,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true
     params:
       method: cot
       data_format: Sharegpt

examples/generate/generate_multi_hop_qa/multi_hop_config.yaml

@@ -53,6 +53,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true
     params:
       method: multi_hop
       data_format: ChatML

examples/generate/generate_vqa/vqa_config.yaml

@@ -54,6 +54,7 @@ nodes:
     execution_params:
       replicas: 1
       batch_size: 128
+    save_output: true
     params:
       method: vqa
       data_format: ChatML

graphgen/bases/datatypes.py

@@ -63,7 +63,11 @@ class Node(BaseModel):
         default_factory=list, description="list of dependent node ids"
     )
     execution_params: dict = Field(
-        default_factory=dict, description="execution parameters like replicas, batch_size"
+        default_factory=dict,
+        description="execution parameters like replicas, batch_size",
+    )
+    save_output: bool = Field(
+        default=False, description="whether to save the output of this node"
     )
 
     @classmethod

graphgen/engine.py

@@ -1,21 +1,22 @@
-import os
 import inspect
 import logging
+import os
 from collections import defaultdict, deque
 from functools import wraps
 from typing import Any, Callable, Dict, List, Set
-from dotenv import load_dotenv
 
 import ray
 import ray.data
+from dotenv import load_dotenv
 from ray.data import DataContext
 
 from graphgen.bases import Config, Node
-from graphgen.utils import logger
 from graphgen.common import init_llm, init_storage
+from graphgen.utils import logger
 
 load_dotenv()
 
+
 class Engine:
     def __init__(
         self, config: Dict[str, Any], functions: Dict[str, Callable], **ray_init_kwargs
@@ -42,7 +43,7 @@ def __init__(
         existing_env_vars = ray_init_kwargs["runtime_env"].get("env_vars", {})
         ray_init_kwargs["runtime_env"]["env_vars"] = {
             **all_env_vars,
-            **existing_env_vars
+            **existing_env_vars,
         }
 
         if not ray.is_initialized():
@@ -265,24 +266,11 @@ def func_wrapper(row_or_batch: Dict[str, Any]) -> Dict[str, Any]:
                     f"Unsupported node type {node.type} for node {node.id}"
                 )
 
-    @staticmethod
-    def _find_leaf_nodes(nodes: List[Node]) -> Set[str]:
-        all_ids = {n.id for n in nodes}
-        deps_set = set()
-        for n in nodes:
-            deps_set.update(n.dependencies)
-        return all_ids - deps_set
-
     def execute(self, initial_ds: ray.data.Dataset) -> Dict[str, ray.data.Dataset]:
         sorted_nodes = self._topo_sort(self.config.nodes)
 
         for node in sorted_nodes:
             self._execute_node(node, initial_ds)
 
-        leaf_nodes = self._find_leaf_nodes(sorted_nodes)
-
-        @ray.remote
-        def _fetch_result(ds: ray.data.Dataset) -> List[Any]:
-            return ds.take_all()
-
-        return {node_id: self.datasets[node_id] for node_id in leaf_nodes}
+        output_nodes = [n for n in sorted_nodes if getattr(n, "save_output", False)]
+        return {node.id: self.datasets[node.id] for node in output_nodes}