blog only

.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
+document_store.json filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,93 @@
+import gradio as gr
+
+from haystack.document_stores.in_memory import InMemoryDocumentStore
+from haystack.components.retrievers.in_memory import InMemoryBM25Retriever, InMemoryEmbeddingRetriever
+from haystack.components.embedders import SentenceTransformersTextEmbedder
+from haystack.components.joiners import DocumentJoiner
+from haystack.components.rankers import TransformersSimilarityRanker
+from haystack import Pipeline,component,Document
+from haystack_integrations.components.generators.google_ai import GoogleAIGeminiGenerator
+from haystack.components.builders import PromptBuilder
+
+@component
+class AnswerBuilder :
+    @component.output_types(reply=str,documents=list[Document])
+    def run(self,replies:list[str],documents:list[Document]):
+        reply = replies[0]
+        return {"reply":reply,'documents':documents}
+
+document_store = InMemoryDocumentStore.load_from_disk(path='./document_store.json')
+print('document_store loaded' ,document_store.count_documents())
+
+class Niwa_rag :
+    def __init__(self):
+        self.createPipe()
+    def createPipe(self):
+        template = """
+            以下の情報に基づいて質問に答えて下さい。
+        
+            Context:
+            {% for document in documents %}
+                {{ document.content }}
+            {% endfor %}
+        
+            質問: {{question}}
+            回答:
+        """
+        text_embedder = SentenceTransformersTextEmbedder()
+        embedding_retriever = InMemoryEmbeddingRetriever(document_store)
+        bm25_retriever = InMemoryBM25Retriever(document_store)
+        document_joiner = DocumentJoiner()
+        ranker = TransformersSimilarityRanker(top_k=6)
+        prompt_builder = PromptBuilder(template=template)
+        gemini = GoogleAIGeminiGenerator(model="models/gemini-1.0-pro")
+        answer_builder = AnswerBuilder()
+        
+        pipe = Pipeline()
+        pipe.add_component("text_embedder", text_embedder)
+        pipe.add_component("embedding_retriever", embedding_retriever)
+        pipe.add_component("bm25_retriever", bm25_retriever)
+        pipe.add_component("document_joiner", document_joiner)
+        pipe.add_component("ranker", ranker)
+        pipe.add_component("prompt_builder", prompt_builder)
+        pipe.add_component("llm", gemini)
+        pipe.add_component("answer_builder", answer_builder)
+        
+        pipe.connect("text_embedder", "embedding_retriever")
+        pipe.connect("bm25_retriever", "document_joiner")
+        pipe.connect("embedding_retriever", "document_joiner")
+        pipe.connect("document_joiner", "ranker")
+        pipe.connect("ranker.documents", "prompt_builder.documents")
+        pipe.connect("prompt_builder.prompt", "llm")
+        pipe.connect("llm.replies", "answer_builder.replies")
+        pipe.connect("ranker.documents", "answer_builder.documents")
+        self.pipe = pipe
+    def run(self,q):
+        if not q :
+            return ['','']
+        result = self.pipe.run(
+            {"text_embedder": {"text": q}, "bm25_retriever": {"query": q}, "ranker": {"query": q},'prompt_builder':{'question':q}}
+        )
+        reply = result['answer_builder']['reply']
+        html = '<div>参考記事</div><div style="margin-left:30px">'
+        for doc in result['answer_builder']['documents'] :
+            title = doc.meta['title']
+            link = doc.meta['link']
+            row = f'<div><a href="{link}" target="_blank">{title}</a></div>'
+            html += row
+        html += '</div>'
+        return [reply,html]
+
+rag = Niwa_rag()
+
+with gr.Blocks() as app:
+    inputs=gr.Textbox(label='質問')
+    submit = gr.Button("送信",variant="primary")
+    reply =gr.Textbox(label='回答')
+    sources =gr.HTML(label='参考記事')
+    submit.click(lambda: gr.update(interactive=False),inputs=None, outputs=submit) \
+        .then(fn=rag.run, inputs=inputs, outputs=[reply,sources] ) \
+        .then(fn=lambda: gr.update(interactive=True),inputs=None, outputs=submit)
+
+if __name__ == "__main__":
+    app.launch(share=True)

document_store.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4c44c65ed78f60d71791e1d1dc28451984561427236f4c231c013e9590d37cd
+size 41714621

requirements.txt

@@ -0,0 +1,7 @@
+numpy
+pandas
+gradio
+haystack-ai
+google-ai-haystack
+accelerate
+sentence-transformers