improve prompt

hugegraph-llm/src/hugegraph_llm/api/rag_api.py

@@ -59,10 +59,12 @@ def rag_answer_api(req: RAGRequest):
             gremlin_tmpl_num=req.gremlin_tmpl_num,
             gremlin_prompt=req.gremlin_prompt or prompt.gremlin_generate_prompt,
         )
+        # TODO: we need more info in the response for users to understand the query logic
         return {
-            key: value
-            for key, value in zip(["raw_answer", "vector_only", "graph_only", "graph_vector_answer"], result)
-            if getattr(req, key)
+            "query": req.query,
+            **{key: value
+               for key, value in zip(["raw_answer", "vector_only", "graph_only", "graph_vector_answer"], result)
+               if getattr(req, key)}
         }
 
     @router.post("/rag/graph", status_code=status.HTTP_200_OK)

hugegraph-llm/src/hugegraph_llm/config/prompt_config.py

@@ -153,16 +153,15 @@ class PromptConfig(BasePromptConfig):
 - 格式示例：
 KEYWORDS:关键词1,关键词2,...,关键词n
 
-MAX_KEYWORDS:
-{max_keywords}
+MAX_KEYWORDS: {max_keywords}
 文本：
 {question}
 """
     # pylint: disable=C0301
     # keywords_extract_prompt_EN = """
     # Instruction:
     # Please perform the following tasks on the text below:
-    # 1. Extract Keywords and Generate Synonyms from text:
+    # 1. Extract Keywords and Generate Synonyms from the text:
     #   - At least 0, at most {max_keywords} keywords.
     #   - For each keyword, generate its synonyms or possible variant forms.
     # Requirements:
@@ -174,29 +173,33 @@ class PromptConfig(BasePromptConfig):
     # Output Format:
     # - Output only one line, prefixed with KEYWORDS:, followed by all keywords and synonyms, separated by commas.No spaces or empty characters are allowed in the extracted keywords.
     # - Format example:
-    # KEYWORDS: keyword1, keyword2, ..., keywordn, synonym1, synonym2, ..., synonymn
+    # KEYWORDS: keyword1, keyword2, ..., keywordN, synonym1, synonym2, ..., synonymN
     # Text:
     # {question}
     # """
 
     gremlin_generate_prompt = """
-Given the graph schema:
+You are an expert in graph query language(Gremlin), your role is to understand the schema of the graph and generate 
+accurate Gremlin code based on the given instructions.
+
+# Graph Schema:
 ```json
 {schema}
 ```
+# Rule:
+The output format must be like:
+```gremlin
+g.V().limit(10)
+```
 
-Given the extracted vertex vid:
+# Extracted vertex vid:
 {vertices}
 
-Given the example query-gremlin pairs:
+# Given the example query-gremlin pairs:
 {example}
 
-Generate gremlin from the following user query.
+# Generate  gremlin from the following user query.
 {query}
 
-The output format must be like:
-```gremlin
-g.V().limit(10)
-```
 The generated gremlin is:
 """

hugegraph-llm/src/hugegraph_llm/demo/rag_demo/text2gremlin_block.py

@@ -25,18 +25,18 @@
 from hugegraph_llm.config import prompt, resource_path, huge_settings
 from hugegraph_llm.models.embeddings.init_embedding import Embeddings
 from hugegraph_llm.models.llms.init_llm import LLMs
-from hugegraph_llm.operators.gremlin_generate_task import GremlinGenerator
 from hugegraph_llm.operators.graph_rag_task import RAGPipeline
+from hugegraph_llm.operators.gremlin_generate_task import GremlinGenerator
 from hugegraph_llm.operators.hugegraph_op.schema_manager import SchemaManager
 from hugegraph_llm.utils.hugegraph_utils import run_gremlin_query
 from hugegraph_llm.utils.log import log
 
 
 def store_schema(schema, question, gremlin_prompt):
     if (
-        prompt.text2gql_graph_schema != schema
-        or prompt.default_question != question
-        or prompt.gremlin_generate_prompt != gremlin_prompt
+        prompt.text2gql_graph_schema != schema or
+        prompt.default_question != question or
+        prompt.gremlin_generate_prompt != gremlin_prompt
     ):
         prompt.text2gql_graph_schema = schema
         prompt.default_question = question
@@ -90,8 +90,7 @@ def gremlin_generate(
     updated_schema = sm.simple_schema(schema) if short_schema else schema
     store_schema(str(updated_schema), inp, gremlin_prompt)
     context = (
-        generator.example_index_query(example_num)
-        .gremlin_generate_synthesize(updated_schema, gremlin_prompt)
+        generator.example_index_query(example_num).gremlin_generate_synthesize(updated_schema, gremlin_prompt)
         .run(query=inp)
     )
     try: