Dev rebased (#332)

* update the default embd. (#310) * Feature/add jina reranker rebased (#312) * Add jina reranker * fix oai * Revampt client & server approach. (#316) * Revampt client & server approach. * cleanups * tweak hyde prompt * Feature/add agent provider (#317) * update pipeline (#315) * Update CONTRIBUTING.md * Delete CONTRIBUTOR.md * Adding agent provider * Feature/add agent provider rebased v2 (#319) * modify prompt provider workflow, add agent * fix run qna client * add provider abstraction * tweaks and cleans * move text splitter config locale * Feature/modify get all uniq values (#325) * refine * update * format * fix * Feature/dev merge rebased (#329) * Update local_rag.mdx * Update llms.mdx (#322) * update the default embd. (#310) * Feature/add agent provider (#317) * update pipeline (#315) * Update CONTRIBUTING.md * Delete CONTRIBUTOR.md * Adding agent provider * Feature/modify get all uniq values (#325) * refine * update * format * fix * dev merge * Feature/fix dev merge mistakes rebased (#330) * Feature/add agent provider (#317) * update pipeline (#315) * Update CONTRIBUTING.md * Delete CONTRIBUTOR.md * Adding agent provider * Feature/modify get all uniq values (#325) * refine * update * format * fix * dev merge * cleanup * Feature/dev cleanup (#331) * final pub * final pub * json clean * fix sentence transformer issue * include rerank * fix llama cpp * rebase * fix rerank * small tweaks * rollbk config * cleanup
SciPhi-AI · Apr 24, 2024 · 510224f · 510224f
1 parent 511367e
commit 510224f
Show file tree

Hide file tree

Showing 74 changed files with 2,692 additions and 1,772 deletions.
diff --git a/.gitignore b/.gitignore
@@ -11,7 +11,8 @@ env/
 node_modules
 
 
-**/*.sqlite
-**/*.sqlite3
+**/*.sqlite*
+**/*.sqlite3*
 
-qdrant_storage/
+qdrant_storage/
+r2r/examples/data/*
diff --git a/README.md b/README.md
@@ -28,6 +28,17 @@ R2R was conceived to bridge the gap between local LLM experimentation and scalab
 - **⚖️ Autoscale**: Scale your pipeline effortlessly in the cloud using [SciPhi](https://sciphi.ai/).
 - **🤖 OSS**: Benefit from a framework developed by the open-source community, designed to simplify RAG deployment.
 
+# Table of Contents
+1. [Demo(s)](#demos)
+2. [Links](#links)
+3. [Quick Install](#quick-install)
+4. [Docker](#docker)
+5. [Q&A Example](#q&a-example)
+6. [HyDE Example](#hyde-example)
+7. [Running Local RAG](#running-local-rag)
+8. [Core Abstractions](#core-abstractions)
+
+
 ## Demo(s)
 
 Using the cloud application to deploy the pre-built basic pipeline:
@@ -73,13 +84,13 @@ docker pull emrgntcmplxty/r2r:latest
 docker run -d --name r2r_container -p 8000:8000 -e CONFIG_OPTION=local_ollama  emrgntcmplxty/r2r:latest
 ```
 
-## Basic Example
+## Q&A Example
 
 [`Configurable Pipeline`](r2r/examples/servers/config_pipeline.py): Execute this script to select and serve a **Q&A RAG**, **Web RAG**, or **Agent RAG** pipeline. This starter pipeline supports ingestion, embedding, and question and the specified RAG, all accessible via a REST API.
    ```bash
    # launch the server
-   # For ex., do `export CONFIG_OPTION=local_ollama` or ``--config=local_ollama` to run fully locally
-   # For ex., do `export PIPELINE_OPTION=web` or ``--pipeline=web` to run WebRAG pipeline
+   # For ex., do `export CONFIG_OPTION=local_ollama` or `--config=local_ollama` to run fully locally
+   # For ex., do `export PIPELINE_OPTION=web` or `--pipeline=web` to run WebRAG pipeline
    python -m r2r.examples.servers.config_pipeline --config=default --pipeline=qna
    ```
 
@@ -113,39 +124,34 @@ docker run -d --name r2r_container -p 8000:8000 -e CONFIG_OPTION=local_ollama  e
    python -m r2r.examples.clients.run_qna_client rag_completion_streaming --query="What was lyfts profit in 2020?"
 
    # <search>[{"id": "a0f6b427-9083-5ef2-aaa1-024b6cebbaee", "score": 0.6862949051074227, "metadata": {"user_id": "df7021ed-6e66-5581-bd69-d4e9ac1e5ada", "pipeline_run_id": "0c2c9a81-0720-4e34-8736-b66189956013", "text": "Title: Lyft 10k 2021\nNet loss was $ ... </search>
-
+   #
    # <context> Title: Lyft 10k 2021 ... </context>
-
+   #
    # <completion>Lyft's net loss in 2020 was $1.8 billion.</completion>
    ```
-### Running Local RAG
 
-[Refer here](https://r2r-docs.sciphi.ai/tutorials/local_rag) for a tutorial on how to modify the commands above to use local providers.
-
-## Synthetic Queries Example
+## HyDE Example
 
-[`Synthetic Query Pipeline`](r2r/examples/servers/synthetic_query_pipeline.py): Execute this script to start a backend server equipped with more advanced synthetic query pipeline. This pipeline is designed to create synthetic queries, enhancing the RAG system's learning and performance.
+[`HyDE Pipeline`](r2r/examples/servers/hyde_pipeline.py): Execute this script to start a backend server equipped with more advanced synthetic query pipeline. This pipeline is designed to create synthetic queries, enhancing the RAG system's learning and performance.
 
    ```bash
    # launch the server
-   python -m r2r.examples.servers.synthetic_query_pipeline
+   python -m r2r.examples.servers.config_pipeline --config=default --pipeline=hyde
    ```
 
-[`Synthetic Query Client`](r2r/examples/clients/run_synthetic_query_client.py): Use this client script after the synthetic query pipeline is running. It's tailored for use with the synthetic query pipeline, demonstrating the improved features of the RAG system.
-
    ```bash
+   # ingests Lyft 10K, Uber 10K, and others
+   python -m r2r.examples.clients.run_qna_client ingest --document_filter=all
+
    # run the client
-   python -m r2r.examples.clients.run_synthetic_query_client
-   ```
+   python -m r2r.examples.clients.run_qna_client search --query="What was lyft and ubers profit in 2020?"
 
-## Extra Examples
+   # {... 'message': {'content': 'In 2020, Lyft reported a net loss of $1.7529 billion [8]. Uber also reported a significant loss for the year 2020, with its net loss improving by $1.8 billion from 2020, indicating a substantial loss for the year as well [38]. Neither company achieved a profit in 2020; instead, they both experienced considerable losses.' ...}
+   ```
 
-[`Reducto Pipeline`](r2r/examples/servers/reducto_pipeline.py): Launch this script to activate a backend server that integrates a Reducto adapter for enhanced PDF ingestion.
+## Running Local RAG
 
-   ```bash
-   # launch the server
-   python -m r2r.examples.servers.reducto_pipeline
-   ```
+[Refer here](https://r2r-docs.sciphi.ai/tutorials/local_rag) for a tutorial on how to modify the commands above to use local providers.
 
 ## Core Abstractions
 

diff --git a/config.json b/config.json
@@ -1,36 +1,36 @@
 {
-  "embedding": {
-    "provider": "openai",
-    "model": "text-embedding-3-small",
-    "dimension": 1536,
-    "batch_size": 32
-  },
-  "evals": {
-    "provider": "parea",
-    "frequency": 1.0
-  },
   "language_model": {
     "provider": "litellm"
   },
-  "logging_database": {
+  "vector_database": {
     "provider": "local",
-    "collection_name": "demo_logs",
-    "level": "INFO"
+    "collection_name": "demo_vecs"
   },
   "ingestion":{
-    "provider": "local",
+    "provider": "local"
+  },
+  "embedding": {
+    "provider": "openai",
+    "search_model": "text-embedding-3-small",
+    "search_dimension": 512,
+    "batch_size": 32,
     "text_splitter": {
       "type": "recursive_character",
       "chunk_size": 512,
       "chunk_overlap": 20
     }
   },
-  "vector_database": {
-    "provider": "local",
-    "collection_name": "demo_vecs"
+  "evals": {
+    "provider": "parea",
+    "frequency": 0.0
   },
   "app": {
     "max_logs": 100,
     "max_file_size_in_mb": 100
+  },
+  "logging_database": {
+    "provider": "local",
+    "collection_name": "demo_logs",
+    "level": "INFO"
   }
 }
diff --git a/docs/pages/deep-dive/app.mdx b/docs/pages/deep-dive/app.mdx
@@ -83,7 +83,12 @@ The application uses a configuration file (`config.json`) to set various setting
     "provider": "openai",
     "model": "text-embedding-3-small",
     "dimension": 1536,
-    "batch_size": 32
+    "batch_size": 32,
+    "text_splitter": {
+      "type": "recursive_character",
+      "chunk_size": 512,
+      "chunk_overlap": 20
+    }
   },
   "evals": {
     "provider": "deepeval",
@@ -99,11 +104,6 @@ The application uses a configuration file (`config.json`) to set various setting
   },
   "ingestion": {
     "provider": "local",
-    "text_splitter": {
-      "type": "recursive_character",
-      "chunk_size": 512,
-      "chunk_overlap": 20
-    }
   },
   "vector_database": {
     "provider": "local",

diff --git a/docs/pages/deep-dive/config.mdx b/docs/pages/deep-dive/config.mdx
@@ -17,7 +17,12 @@ The default values for the config are shown below:
     "provider": "openai",
     "model": "text-embedding-ada-002",
     "dimension": 1536,
-    "batch_size": 32
+    "batch_size": 32,
+    "text_splitter": {
+      "type": "recursive_character",
+      "chunk_size": 512,
+      "chunk_overlap": 20
+    }
   },
   "evals": {
     "provider": "deepeval",
@@ -32,12 +37,7 @@ The default values for the config are shown below:
     "level": "INFO"
   },
   "ingestion": {
-    "provider": "local",
-    "text_splitter": {
-      "type": "recursive_character",
-      "chunk_size": 512,
-      "chunk_overlap": 20
-    }
+    "provider": "local"
   },
   "vector_database": {
     "provider": "local",

diff --git a/docs/pages/deep-dive/embedding.mdx b/docs/pages/deep-dive/embedding.mdx
@@ -8,7 +8,7 @@ The `BasicEmbeddingPipeline` is a simple implementation of the `EmbeddingPipelin
 
 ### Initialization
 
-The `BasicEmbeddingPipeline` is initialized with the `embedding_model`, `embeddings_provider`, `db`, `text_splitter`, an optional `logging_connection`, `embedding_batch_size`, and `id_prefix`.
+The `BasicEmbeddingPipeline` is initialized with an `embedding_provider`, `vector_db_provider`, `text_splitter`, an optional `logging_connection`, `embedding_batch_size`, and `id_prefix`.
 
 ### Text Extraction
 
@@ -70,18 +70,16 @@ import textstat
 class CustomEmbeddingPipeline(BasicEmbeddingPipeline):
     def __init__(
         self,
-        embedding_model: str,
-        embeddings_provider: OpenAIEmbeddingProvider,
-        db: VectorDBProvider,
+        embedding_provider: OpenAIEmbeddingProvider,
+        vector_db_provider: VectorDBProvider,
         text_splitter: TextSplitter,
         logging_connection: Optional[LoggingDatabaseConnection] = None,
         embedding_batch_size: int = 1,
         id_prefix: str = "demo",
     ):
         super().__init__(
-            embedding_model,
-            embeddings_provider,
-            db,
+            embedding_provider,
+            vector_db_provider,
             text_splitter,
             logging_connection,
             embedding_batch_size,

diff --git a/docs/pages/deep-dive/factory.mdx b/docs/pages/deep-dive/factory.mdx
@@ -11,10 +11,9 @@ from r2r.main import E2EPipelineFactory, R2RConfig
 
 app = E2EPipelineFactory.create_pipeline(
     config=R2RConfig.load_config(),
-    db=None,
-    embeddings_provider=None,
-    llm=None,
-    text_splitter=None,
+    vector_db_provider=None,
+    embedding_provider=None,
+    llm_provider=None,
     adapters=None,
     ingestion_pipeline_impl=BasicIngestionPipeline,
     embedding_pipeline_impl=BasicEmbeddingPipeline,
@@ -37,7 +36,7 @@ This method retrieves the appropriate vector database based on the `database_con
 - `pgvector`: Returns an instance of `PGVectorDB`.
 - `local`: Returns an instance of `LocalVectorDB`.
 
-### `get_embeddings_provider(embedding_config: dict[str, Any])`
+### `get_embedding_provider(embedding_config: dict[str, Any])`
 
 This method retrieves the appropriate embeddings provider based on the `embedding_config`. It supports the following embeddings providers:
 - `openai`: Returns an instance of `OpenAIEmbeddingProvider`.
@@ -46,8 +45,8 @@ This method retrieves the appropriate embeddings provider based on the `embeddin
 ### `get_llm(llm_config: dict[str, Any])`
 
 This method retrieves the appropriate language model based on the `llm_config`. It supports the following language models:
-- `openai`: Returns an instance of `OpenAILLM` with the `OpenAIConfig`.
-- `litellm`: Returns an instance of `LiteLLM` with the `LiteLLMConfig`.
+- `openai`: Returns an instance of `OpenAILLM`.
+- `litellm`: Returns an instance of `LiteLLM`.
 
 ### `get_text_splitter(text_splitter_config: dict[str, Any])`
 
@@ -58,13 +57,12 @@ This method retrieves the appropriate text splitter based on the `text_splitter_
 This method creates the end-to-end pipeline by assembling various components based on the provided `config` and optional parameters. It performs the following steps:
 
 1. Sets up logging based on the `logging_database` configuration.
-2. Retrieves the embeddings provider using `get_embeddings_provider` or uses the provided `embeddings_provider`.
-3. Retrieves the vector database using `get_vector_db` or uses the provided `db`.
+2. Retrieves the embeddings provider using `get_embedding_provider` or uses the provided `embedding_provider`.
+3. Retrieves the vector database using `get_vector_db_provider` or uses the provided `vector_db_provider`.
 4. Initializes the vector database collection with the specified `collection_name` and `embedding_dimension`.
-5. Retrieves the language model using `get_llm` or uses the provided `llm`.
+5. Retrieves the language model using `get_llm_provider` or uses the provided `llm`.
 6. Creates a `LoggingDatabaseConnection` instance for logging purposes.
 7. Creates an instance of the `rag_pipeline_impl` (default: `QnARAGPipeline`) with the language model, vector database, embedding model, embeddings provider, and logging connection.
-8. Retrieves the text splitter using `get_text_splitter` or uses the provided `text_splitter`.
 9. Creates an instance of the `embedding_pipeline_impl` (default: `BasicEmbeddingPipeline`) with the embedding model, embeddings provider, vector database, logging connection, text splitter, and embedding batch size.
 10. Creates an instance of the `eval_pipeline_impl` (default: `BasicEvalPipeline`) with the evaluation configuration and logging connection.
 11. Creates an instance of the `ingestion_pipeline_impl` (default: `BasicIngestionPipeline`) with the specified adapters.

diff --git a/docs/pages/deep-dive/rag.mdx b/docs/pages/deep-dive/rag.mdx
@@ -9,10 +9,9 @@ The `QnARAGPipeline` is a simple implementation of the `RAGPipeline` abstract ba
 ### Initialization
 
 The `QnARAGPipeline` is initialized with the following parameters:
-- `llm`: An instance of `LLMProvider` for generating completions.
-- `db`: An instance of `VectorDBProvider` for searching and retrieving documents.
-- `embedding_model`: The name of the embedding model to use.
-- `embeddings_provider`: An instance of `OpenAIEmbeddingProvider` for generating embeddings.
+- `llm_provider`: An instance of `LLMProvider` for generating completions.
+- `vector_db_provider`: An instance of `VectorDBProvider` for searching and retrieving documents.
+- `embedding_provider`: An instance of `OpenAIEmbeddingProvider` for generating embeddings.
 - `prompt_provider` (optional): An instance of `PromptProvider` for providing prompts (default is `BasicPromptProvider`).
 - `logging_connection` (optional): An instance of `LoggingDatabaseConnection` for logging.
 
@@ -40,10 +39,9 @@ To create a custom RAG pipeline, you can subclass the `RAGPipeline` abstract bas
 class CustomRAGPipeline(RAGPipeline):
     def __init__(
         self,
-        llm: LLMProvider,
-        db: VectorDBProvider,
-        embedding_model: str,
-        embeddings_provider: OpenAIEmbeddingProvider,
+        llm_provider: LLMProvider,
+        vector_db_provider: VectorDBProvider,
+        embedding_provider: OpenAIEmbeddingProvider,
         prompt_provider: Optional[PromptProvider] = None,
         logging_connection: Optional[LoggingDatabaseConnection] = None,
     ) -> None:
@@ -52,13 +50,12 @@ class CustomRAGPipeline(RAGPipeline):
         self.prompt_provider = prompt_provider
 
         super().__init__(
-            llm,
+            llm_provider=llm_provider,
             prompt_provider=prompt_provider,
+            embedding_provider=embedding_provider,
+            vector_db_provider=vector_db_provider,
             logging_connection=logging_connection,
         )
-        self.embedding_model = embedding_model
-        self.embeddings_provider = embeddings_provider
-        self.db = db
         self.pipeline_run_info = None
 
     def transform_query(self, query: str) -> str:
@@ -75,21 +72,20 @@ class CustomRAGPipeline(RAGPipeline):
         **kwargs,
     ) -> list[VectorSearchResult]:
         # Custom document retrieval logic
-        results = self.db.search(
-            query_vector=self.embeddings_provider.get_embedding(
+        results = self.vector_db_provider.search(
+            query_vector=self.embedding_provider.get_embedding(
                 transformed_query,
-                self.embedding_model,
             ),
             filters=filters,
             limit=limit,
         )
         return results
 
     def rerank_results(
-        self, results: list[VectorSearchResult]
+        self, transformed_query: str, results: list[VectorSearchResult], limit
     ) -> list[VectorSearchResult]:
         # Custom result reranking logic
-        return list(reversed(results))
+        return list(reversed(results))[0:limit]
 
     def _format_results(self, results: list[VectorSearchResult]) -> str:
         # Custom result formatting logic

diff --git a/docs/pages/providers/embeddings.mdx b/docs/pages/providers/embeddings.mdx
@@ -35,4 +35,4 @@ Anything supported by OpenAI, such as:
 - **Pricing**: Approximately 12,500 pages per dollar. Balances cost and performance effectively.
 - **More**: [Embeddings Guide](https://platform.openai.com/docs/guides/embeddings)
 
-Lastly, the `sentence_transformer` package from HuggingFace is also supported as a provider. For example, one such popular model is [`all-MiniLM-L6-v2`](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2).
+Lastly, the `sentence_transformer` package from HuggingFace is also supported as a provider. For example, one such popular model is [`mixedbread-ai/mxbai-embed-large-v1`](https://huggingface.co/sentence-transformers/mixedbread-ai/mxbai-embed-large-v1).
diff --git a/docs/pages/providers/llms.mdx b/docs/pages/providers/llms.mdx
@@ -93,7 +93,7 @@ pip install llama-cpp-python # or poetry install ... -E local-llm
 
 To configure the LlamaCpp provider, you need to set the following options in the `language_model` section of the configuration file (`config.json`):
 
-- `provider_name`: Set it to `"llama-cpp"` to use the LlamaCpp provider.
+- `provider`: Set it to `"llama-cpp"` to use the LlamaCpp provider.
 - `model_path`: Specify the path where the LlamaCpp models are located. If not provided, it defaults to `~/.cache/models`.
 - `model_name`: Specify the name of the LlamaCpp model to use. If not provided, it defaults to `"tinyllama-1.1b-chat-v1.0.Q2_K.gguf"`.
 

diff --git a/docs/pages/tutorials/configuring_your_rag_pipeline.mdx b/docs/pages/tutorials/configuring_your_rag_pipeline.mdx
@@ -226,7 +226,7 @@ Set the `provider` field under `vector_database` in `config.json` to specify you
 #### Embedding Provider
 R2R supports OpenAI and local inference embedding providers:
 - `openai`: OpenAI models like `text-embedding-3-small`
-- `sentence-transformers`: HuggingFace models like `all-MiniLM-L6-v2`
+- `sentence-transformers`: HuggingFace models like `mixedbread-ai/mxbai-embed-large-v1`
 
 Configure the `embedding` section to set your desired embedding model, dimension, and batch size.