fix(webui): sync gradio demo

Author: ChenZiHong-Gavin

.github/workflows/push-to-hf.yml

@@ -43,7 +43,7 @@ jobs:
         [[ -d hf-repo ]] && rm -rf hf-repo
         git clone https://huggingface.co/${HF_REPO_TYPE}/${HF_REPO_ID} hf-repo
 
-        rsync -a --delete --exclude='.git' ./ hf-repo/ || true
+        rsync -a --delete --exclude='.git' --exclude='hf-repo' ./ hf-repo/
 
         cd hf-repo
         git add .

graphgen/graphgen.py

@@ -23,8 +23,8 @@
     judge_statement,
     quiz,
     search_all,
-    traverse_graph_atomically,
-    traverse_graph_by_edge,
+    traverse_graph_for_aggregated,
+    traverse_graph_for_atomic,
     traverse_graph_for_multi_hop,
 )
 from .utils import (
@@ -69,6 +69,7 @@ def __post_init__(self):
         self.tokenizer_instance: Tokenizer = Tokenizer(
             model_name=self.config["tokenizer"]
         )
+        print(os.getenv("SYNTHESIZER_MODEL"), os.getenv("SYNTHESIZER_API_KEY"))
         self.synthesizer_llm_client: OpenAIModel = OpenAIModel(
             model_name=os.getenv("SYNTHESIZER_MODEL"),
             api_key=os.getenv("SYNTHESIZER_API_KEY"),
@@ -326,7 +327,7 @@ async def async_traverse(self):
         output_data_type = self.config["output_data_type"]
 
         if output_data_type == "atomic":
-            results = await traverse_graph_atomically(
+            results = await traverse_graph_for_atomic(
                 self.synthesizer_llm_client,
                 self.tokenizer_instance,
                 self.graph_storage,
@@ -344,7 +345,7 @@ async def async_traverse(self):
                 self.progress_bar,
             )
         elif output_data_type == "aggregated":
-            results = await traverse_graph_by_edge(
+            results = await traverse_graph_for_aggregated(
                 self.synthesizer_llm_client,
                 self.tokenizer_instance,
                 self.graph_storage,

graphgen/operators/__init__.py

@@ -5,8 +5,8 @@
 from .judge import judge_statement
 from .quiz import quiz
 from .traverse_graph import (
-    traverse_graph_atomically,
-    traverse_graph_by_edge,
+    traverse_graph_for_aggregated,
+    traverse_graph_for_atomic,
     traverse_graph_for_multi_hop,
 )
 
@@ -15,8 +15,8 @@
     "quiz",
     "judge_statement",
     "search_all",
-    "traverse_graph_by_edge",
-    "traverse_graph_atomically",
+    "traverse_graph_for_aggregated",
+    "traverse_graph_for_atomic",
     "traverse_graph_for_multi_hop",
     "generate_cot",
 ]

graphgen/operators/traverse_graph.py

@@ -158,7 +158,7 @@ def _post_process_synthetic_data(data):
     return qas
 
 
-async def traverse_graph_by_edge(
+async def traverse_graph_for_aggregated(
     llm_client: OpenAIModel,
     tokenizer: Tokenizer,
     graph_storage: NetworkXStorage,
@@ -251,7 +251,6 @@ async def _process_single_batch(
             qas = _post_process_synthetic_data(content)
 
             if len(qas) == 0:
-                print(content)
                 logger.error(
                     "Error occurred while processing batch, question or answer is None"
                 )
@@ -307,7 +306,8 @@ async def _process_single_batch(
     return results
 
 
-async def traverse_graph_atomically(
+# pylint: disable=too-many-branches, too-many-statements
+async def traverse_graph_for_atomic(
     llm_client: OpenAIModel,
     tokenizer: Tokenizer,
     graph_storage: NetworkXStorage,
@@ -328,17 +328,28 @@ async def traverse_graph_atomically(
     :param max_concurrent
     :return: question and answer
     """
-    assert traverse_strategy.qa_form == "atomic"
 
+    assert traverse_strategy.qa_form == "atomic"
     semaphore = asyncio.Semaphore(max_concurrent)
 
+    def _parse_qa(qa: str) -> tuple:
+        if "Question:" in qa and "Answer:" in qa:
+            question = qa.split("Question:")[1].split("Answer:")[0].strip()
+            answer = qa.split("Answer:")[1].strip()
+        elif "问题：" in qa and "答案：" in qa:
+            question = qa.split("问题：")[1].split("答案：")[0].strip()
+            answer = qa.split("答案：")[1].strip()
+        else:
+            return None, None
+        return question.strip('"'), answer.strip('"')
+
     async def _generate_question(node_or_edge: tuple):
         if len(node_or_edge) == 2:
             des = node_or_edge[0] + ": " + node_or_edge[1]["description"]
-            loss = node_or_edge[1]["loss"]
+            loss = node_or_edge[1]["loss"] if "loss" in node_or_edge[1] else -1.0
         else:
             des = node_or_edge[2]["description"]
-            loss = node_or_edge[2]["loss"]
+            loss = node_or_edge[2]["loss"] if "loss" in node_or_edge[2] else -1.0
 
         async with semaphore:
             try:
@@ -350,13 +361,8 @@ async def _generate_question(node_or_edge: tuple):
                     )
                 )
 
-                if "Question:" in qa and "Answer:" in qa:
-                    question = qa.split("Question:")[1].split("Answer:")[0].strip()
-                    answer = qa.split("Answer:")[1].strip()
-                elif "问题：" in qa and "答案：" in qa:
-                    question = qa.split("问题：")[1].split("答案：")[0].strip()
-                    answer = qa.split("答案：")[1].strip()
-                else:
+                question, answer = _parse_qa(qa)
+                if question is None or answer is None:
                     return {}
 
                 question = question.strip('"')
@@ -386,16 +392,18 @@ async def _generate_question(node_or_edge: tuple):
         if "<SEP>" in node[1]["description"]:
             description_list = node[1]["description"].split("<SEP>")
             for item in description_list:
-                tasks.append((node[0], {"description": item, "loss": node[1]["loss"]}))
+                tasks.append((node[0], {"description": item}))
+                if "loss" in node[1]:
+                    tasks[-1][1]["loss"] = node[1]["loss"]
         else:
             tasks.append((node[0], node[1]))
     for edge in edges:
         if "<SEP>" in edge[2]["description"]:
             description_list = edge[2]["description"].split("<SEP>")
             for item in description_list:
-                tasks.append(
-                    (edge[0], edge[1], {"description": item, "loss": edge[2]["loss"]})
-                )
+                tasks.append((edge[0], edge[1], {"description": item}))
+                if "loss" in edge[2]:
+                    tasks[-1][2]["loss"] = edge[2]["loss"]
         else:
             tasks.append((edge[0], edge[1], edge[2]))