sbert gradio interface

.gitignore

@@ -1 +1,3 @@
 venv/
+__pycache__/
+

README.md

@@ -1,5 +1,13 @@
 # Document Semantic Search
 
+## Run
+
+Run the app in reload mode with this command. This will let the app reload automatically when changes are made to the python script.
+
+```shell
+$ gradio main.py
+```
+
 ## Setup
 
 [Link to venv docs](https://docs.python.org/3/library/venv.html)

main.py

@@ -1,6 +1,6 @@
-from haystack.nodes import PreProcessor, PDFToTextConverter, EmbeddingRetriever
+from haystack.nodes import PreProcessor, PDFToTextConverter, EmbeddingRetriever, TransformersReader
 from haystack.document_stores import InMemoryDocumentStore
-from haystack.pipelines import DocumentSearchPipeline
+from haystack.pipelines import DocumentSearchPipeline, ExtractiveQAPipeline
 import gradio as gr
 
 preprocessor = PreProcessor(
@@ -12,9 +12,11 @@ preprocessor = PreProcessor(
     split_respect_sentence_boundary=True,
     split_overlap=3
 )
-document_store = InMemoryDocumentStore()
+document_store = InMemoryDocumentStore(embedding_dim=384)
+reader = TransformersReader("sentence-transformers/all-MiniLM-L6-v2")
 retriever = EmbeddingRetriever(document_store=document_store, embedding_model="sentence-transformers/all-MiniLM-L6-v2")
-pipeline = DocumentSearchPipeline(retriever)
+pipeline = ExtractiveQAPipeline(reader, retriever)
+converter = PDFToTextConverter(remove_numeric_tables=True)
 
 def print_answers(results):
     fields = ["answer", "score"] # "context"
@@ -28,27 +30,26 @@ def print_answers(results):
     return filtered_answers
 
 def write_pdf(pdf_file):
-    converter = PDFToTextConverter(remove_numeric_tables=True, valid_languages=["en"])
     document = converter.convert(file_path=pdf_file.name, meta=None)[0]
     preprocessed_docs = preprocessor.process(document)
     document_store.write_documents(preprocessed_docs)
+    document_store.update_embeddings(retriever)
 
 def predict(question, pdf_file):
-    print("Start processing pdf")
     write_pdf(pdf_file)
-    print("Processing done.")
     result = pipeline.run(query=question, params={"Retriever": { "top_k": 2 }})
     answers = print_answers(result)
     return answers
 
-title = "Search"
 interface = gr.Interface(
     fn=predict,
-    inputs=[gr.components.Textbox(lines = 3, label="Ask an open question!"),gr.components.File(file_count="single", type="file", label="Upload a pdf")],
+    inputs=[
+        gr.components.Textbox(lines = 1, label="Enter your search query here..."),
+        gr.components.File(file_count="single", type="file", label="Upload a file here.")
+    ],
     outputs="text",
-    title=title, 
-    flagging_options=["top", "medium", "bad"],
-    interpretation="default",
+    title="Search",
+    interpretation=None,
     theme="default" # “default", “huggingface", “dark-grass", “peach"
 )