Add application

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.sqlite3 filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,62 @@
+from mcp.server.fastmcp import FastMCP
+from tools.law_tool import LawTool
+from tools.law_rag_query import LawRAGQuery
+import gradio as gr
+
+law_tool = LawTool()
+law_rag_query = LawRAGQuery()
+
+# Create an MCP server
+mcp = FastMCP("Law Tool Service")
+
+
+# Tool implementation
+@mcp.tool()
+def get_law(category: str, number: int) -> str:
+    """
+    This is a tool that returns law content by input the category and number.
+
+    Args:
+        category: the law category (such as 民法, 中華民國刑法, 民事訴訟法, 刑事訴訟法, 律師法 etc).
+        number: the law number (such as 23).
+
+    Returns:
+        str: The content of the law.
+    """
+    result = law_tool(category, number)
+    return result
+
+# Tool implementation
+@mcp.tool()
+def rag_query(question: str) -> list:
+    """
+    This is a tool that returns law content by input a question. It will find the related law and return.
+
+    Args:
+        question: the question to query the law.
+
+    Returns:
+        list: A list of law content related to the question.
+    """
+    result = law_rag_query(question)
+    return result
+
+with gr.Blocks() as demo:
+    with gr.Tabs():
+        with gr.Tab("Law Tool"):
+            with gr.Row():
+                category = gr.Dropdown(label="Law Category", choices=["民法", "中華民國刑法", "民事訴訟法", "刑事訴訟法", "律師法"], info="選擇法律類別")
+                number = gr.Number(label="Law Number", info="ex：23")
+            query_btn = gr.Button("Submit")
+            result = gr.Textbox(label="Result")
+            query_btn.click(fn=get_law, inputs=[category, number], outputs=result)
+
+        with gr.Tab("Law RAG Query"):
+            with gr.Row():
+                text_input = gr.Textbox(label="Question")
+            rag_btn = gr.Button("Submit")
+            text_output = gr.List(headers=["content", "score"], value=[], label="Result", col_count=2)
+            rag_btn.click(fn=rag_query, inputs=text_input, outputs=text_output)
+
+if __name__ == "__main__":
+    demo.launch(mcp_server=True,server_name="0.0.0.0",allowed_paths=["/"])

chroma_db/bf252843-3061-49cb-984b-ed9693b00042/data_level0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88220a5c7fbc7492735982b90a409aa09054a1b7870c5b04f9c4ea1aa5457a1e
+size 8472000

chroma_db/bf252843-3061-49cb-984b-ed9693b00042/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0b959477622bd3d3a03a215f7b27bc5de1b741bf91bf267fa2b94518e51d0b6
+size 100

chroma_db/bf252843-3061-49cb-984b-ed9693b00042/index_metadata.pickle

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fce4cd4579cfb12481086209613d5b102a648159609a3bc71b4d66a04ab6eaf
+size 113967

chroma_db/bf252843-3061-49cb-984b-ed9693b00042/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f3e84c4f180dae5c9066041c4050236ae8957e1f2c62ed7df0cd8c26c86d922
+size 8000

chroma_db/bf252843-3061-49cb-984b-ed9693b00042/link_lists.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c393c07422eede9e16c4ce4b1395b78bcebb6ce004ca5df80248f241e829f8e4
+size 16976

chroma_db/chroma.sqlite3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18a18286966e3457e92e88e042446f0ee8047a9590a4172439162347d3f95563
+size 21917696

requirements.txt

@@ -0,0 +1,9 @@
+duckduckgo_search
+smolagents
+gradio[mcp]
+datasets
+langchain
+langchain-chroma
+langchain-text-splitters
+langchain-community
+sentence-transformers

tools/final_answer.py

@@ -0,0 +1,14 @@
+from typing import Any, Optional
+from smolagents.tools import Tool
+
+class FinalAnswerTool(Tool):
+    name = "final_answer"
+    description = "Provides a final answer to the given problem."
+    inputs = {'answer': {'type': 'any', 'description': 'The final answer to the problem'}}
+    output_type = "any"
+
+    def forward(self, answer: Any) -> Any:
+        return answer
+
+    def __init__(self, *args, **kwargs):
+        self.is_initialized = False

tools/law_rag_query.py

@@ -0,0 +1,54 @@
+from smolagents import Tool
+from langchain_chroma import Chroma
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from datasets import load_dataset
+import os
+
+class LawRAGQuery(Tool):
+    name = "law_rag_query"
+    description = """
+    This is a tool that returns law content by input a question. It will find the related law and return."""
+    inputs = {
+        "question": {
+            "type": "string",
+            "description": "the question",
+        }
+    }
+    output_type = "array"
+    vectorstore = None
+
+    def __init__(self):
+        dataset = load_dataset("robin0307/law", split='train')
+        law = dataset.to_pandas()
+        self.vectorstore = self.get_vectorstore("thenlper/gte-large-zh", list(law['content']))
+        super().__init__()
+
+    def get_vectorstore(self, model_path, data_list, path="chroma_db"):
+        embeddings = HuggingFaceEmbeddings(model_name=model_path)
+        text_splitter = RecursiveCharacterTextSplitter(chunk_size=2048, chunk_overlap=50)
+        chunks = [text_splitter.split_text(text) for text in data_list]
+
+        # Flatten the list
+        if os.path.isdir(path):
+            vectorstore = Chroma(embedding_function=embeddings, persist_directory=path)
+        else:
+            splits = [chunk for sublist in chunks for chunk in sublist]
+            vectorstore = Chroma.from_texts(texts=splits, embedding=embeddings, persist_directory=path)
+        print("count:", vectorstore._collection.count())
+        return vectorstore
+
+    def get_docs(self, input, k=10):
+        retrieved_documents = self.vectorstore.similarity_search_with_score(input, k=50)
+
+        results = []
+        for i, (doc, score) in enumerate(retrieved_documents):
+            results.append((doc.page_content, score))
+            if i >= k:
+                break
+        return results
+
+    def forward(self, question: str):
+        docs = self.get_docs(question)
+        return docs
+    

tools/law_tool.py

@@ -0,0 +1,31 @@
+from smolagents import Tool
+from datasets import load_dataset
+
+class LawTool(Tool):
+    name = "law_tool"
+    description = """
+    This is a tool that returns law content by input the category and number."""
+    inputs = {
+        "category": {
+            "type": "string",
+            "description": "the law category (such as 民法, 中華民國刑法, 民事訴訟法, 刑事訴訟法, 律師法 etc)",
+        },
+        "number": {
+            "type": "integer",
+            "description": "the law number (such as 23)"
+        }
+    }
+    output_type = "string"
+    law = None
+
+    def __init__(self):
+        dataset = load_dataset("robin0307/law", split='train')
+        self.law = dataset.to_pandas()
+        super().__init__()
+
+    def forward(self, category: str, number: int):
+        if category == "刑法":
+            category = "中華民國刑法"
+        
+        data = self.law.loc[(self.law["category"]==category) & (self.law["number"]==number), "content"].values[0]
+        return data