feat: add bds baseline

baselines/BDS/README.md

@@ -0,0 +1,2 @@
+# Bottom-up Domain-specific Superintelligence: A Reliable Knowledge Graph is What We Need
+https://arxiv.org/pdf/2507.13966

baselines/BDS/bds.py

@@ -0,0 +1,161 @@
+import argparse
+import asyncio
+import json
+import os
+from dataclasses import dataclass
+from typing import List
+
+import networkx as nx
+from dotenv import load_dotenv
+from tqdm.asyncio import tqdm as tqdm_async
+
+from graphgen.models import NetworkXStorage, OpenAIModel, Tokenizer
+from graphgen.utils import create_event_loop
+
+QA_GENERATION_PROMPT = """
+Create an agriculture examination question for advanced agricultural students that tests the relationship between {src} and {tgt}. The relationship is: {path}. The question should:
+    1. Be in multiple choice format (4 options)
+    2. Require agriculture reasoning along the relationship
+    3. Include a brief farm or field scenario
+    4. Not directly mention the relationship in the question stem
+    5. Have one clearly correct answer
+Format:
+    <Question>
+        [Farm or Field Scenario]
+    </Question>
+    <Options>
+        A. [Option]
+        B. [Option]
+        C. [Option]
+        D. [Option]
+    </Options>
+    <Answer>:
+        [Correct Option Letter]
+    </Answer>
+"""
+
+
+def _post_process(text: str) -> dict:
+    try:
+        q = text.split("<Question>")[1].split("</Question>")[0].strip()
+        opts = text.split("<Options>")[1].split("</Options>")[0].strip().splitlines()
+        opts = [o.strip() for o in opts if o.strip()]
+        ans = text.split("<Answer>:")[1].strip()[0].upper()
+        return {
+            "question": q,
+            "options": opts,
+            "answer": ord(ans) - ord("A"),
+            "raw": text,
+        }
+    except Exception as e:  # pylint: disable=broad-except
+        print(f"Error in post-processing: {e}")
+        return {}
+
+
+@dataclass
+class BDS:
+    llm_client: OpenAIModel = None
+    max_concurrent: int = 1000
+
+    def generate(self, tasks: List[dict]) -> List[dict]:
+        loop = create_event_loop()
+        return loop.run_until_complete(self._async_generate(tasks))
+
+    async def _async_generate(self, tasks: List[dict]) -> List[dict]:
+        sem = asyncio.Semaphore(self.max_concurrent)
+
+        async def job(item):
+            async with sem:
+                path_str = " -> ".join([f"({h},{r},{t})" for h, r, t in item["path"]])
+                prompt = QA_GENERATION_PROMPT.format(
+                    src=item["src"], tgt=item["tgt"], path=path_str
+                )
+                resp = await self.llm_client.generate_answer(prompt)
+                return _post_process(resp)
+
+        tasks = [job(it) for it in tasks]
+        results = []
+        for coro in tqdm_async(asyncio.as_completed(tasks), total=len(tasks)):
+            try:
+                if r := await coro:
+                    results.append(r)
+            except Exception as e:  # pylint: disable=broad-except
+                print("Error:", e)
+        return results
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--input_file",
+        help="GraphML input file path.",
+        default="resources/input_examples/graphml_demo.graphml",
+        type=str,
+    )
+    parser.add_argument(
+        "--output_file",
+        help="Output file path.",
+        default="cache/data/bds_qa.jsonl",
+        type=str,
+    )
+    args = parser.parse_args()
+
+    load_dotenv()
+
+    tokenizer_instance: Tokenizer = Tokenizer(
+        model_name=os.getenv("TOKENIZER_MODEL", "cl100k_base")
+    )
+    llm_client = OpenAIModel(
+        model_name=os.getenv("SYNTHESIZER_MODEL"),
+        api_key=os.getenv("SYNTHESIZER_API_KEY"),
+        base_url=os.getenv("SYNTHESIZER_BASE_URL"),
+        tokenizer_instance=tokenizer_instance,
+    )
+    bds = BDS(llm_client=llm_client)
+
+    graph = NetworkXStorage.load_nx_graph(args.input_file)
+
+    MAX_PATH = 20000
+    all_paths = []
+
+    G = graph.to_directed() if not graph.is_directed() else graph
+    print(G)
+
+    source_nodes = [n for n in G.nodes if G.out_degree(n) > 0][:1000]
+
+    for src in source_nodes:
+        for path in nx.all_simple_paths(G, source=src, target=list(G.nodes), cutoff=3):
+            if len(path) == 4:
+                all_paths.append(path)
+                if len(all_paths) >= MAX_PATH:
+                    break
+            if len(all_paths) >= MAX_PATH:
+                break
+        if len(all_paths) >= MAX_PATH:
+            break
+
+    print(f"Found {len(all_paths)} 4-node paths")
+
+    items = []
+    for path in all_paths:
+        path_edges = []
+        for i in range(len(path) - 1):
+            edge_data = G.get_edge_data(path[i], path[i + 1])
+            if edge_data is None:
+                edge_data = G.get_edge_data(path[i + 1], path[i])
+            if edge_data is None:
+                print(f"Warning: No edge data between {path[i]} and {path[i+1]}")
+                relation = "related_to"
+            else:
+                relation = edge_data.get("relation", "related_to")
+            path_edges.append((path[i], relation, path[i + 1]))
+        items.append({"src": path[0], "tgt": path[-1], "path": path_edges})
+
+    print(f"Prepared {len(items)} items for question generation")
+
+    qa_pairs = bds.generate(items)
+    print(f"Generated {len(qa_pairs)} QA pairs")
+
+    # Save results
+    with open(args.output_file, "w", encoding="utf-8") as f:
+        json.dump(qa_pairs, f, indent=4, ensure_ascii=False)

baselines/EntiGraph/README.md

@@ -0,0 +1,3 @@
+# EntiGraph
+https://arxiv.org/abs/2409.07431
+https://github.com/zitongyang/synthetic_continued_pretraining

baselines/EntiGraph/entigraph.py

@@ -1,6 +1,3 @@
-# https://arxiv.org/abs/2409.07431
-# https://github.com/zitongyang/synthetic_continued_pretraining
-
 import argparse
 import asyncio
 import json

baselines/Genie/README.md

@@ -0,0 +1,2 @@
+# Genie
+https://arxiv.org/pdf/2401.14367

baselines/Genie/genie.py

@@ -1,5 +1,3 @@
-# https://arxiv.org/pdf/2401.14367
-
 import argparse
 import asyncio
 import json

baselines/LongForm/README.md

@@ -0,0 +1,3 @@
+# LongFrom
+https://arxiv.org/pdf/2304.08460
+https://github.com/akoksal/LongForm/tree/main

baselines/LongForm/longform.py

@@ -1,6 +1,3 @@
-# https://arxiv.org/pdf/2304.08460
-# https://github.com/akoksal/LongForm/tree/main
-
 import argparse
 import asyncio
 import json

baselines/SELF-QA/README.md

@@ -0,0 +1,2 @@
+# SELF-QA
+https://arxiv.org/abs/2305.11952

baselines/SELF-QA/self-qa.py

@@ -1,5 +1,3 @@
-# https://arxiv.org/abs/2305.11952
-
 import argparse
 import asyncio
 import json

baselines/Wrap/README.md

@@ -0,0 +1,2 @@
+# Wrap
+https://arxiv.org/abs/2401.16380

baselines/Wrap/wrap.py

@@ -1,5 +1,3 @@
-# https://arxiv.org/abs/2401.16380
-
 import argparse
 import asyncio
 import json