llama_toolchain -> llama_stack

llama_stack/memory/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.

llama_stack/memory/adapters/chroma/__init__.py

@@ -0,0 +1,15 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from llama_stack.core.datatypes import RemoteProviderConfig
+
+
+async def get_adapter_impl(config: RemoteProviderConfig, _deps):
+    from .chroma import ChromaMemoryAdapter
+
+    impl = ChromaMemoryAdapter(config.url)
+    await impl.initialize()
+    return impl

llama_stack/memory/adapters/chroma/chroma.py

@@ -0,0 +1,165 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import json
+import uuid
+from typing import List
+from urllib.parse import urlparse
+
+import chromadb
+from numpy.typing import NDArray
+
+from llama_stack.memory.api import *  # noqa: F403
+
+
+from llama_stack.memory.common.vector_store import BankWithIndex, EmbeddingIndex
+
+
+class ChromaIndex(EmbeddingIndex):
+    def __init__(self, client: chromadb.AsyncHttpClient, collection):
+        self.client = client
+        self.collection = collection
+
+    async def add_chunks(self, chunks: List[Chunk], embeddings: NDArray):
+        assert len(chunks) == len(
+            embeddings
+        ), f"Chunk length {len(chunks)} does not match embedding length {len(embeddings)}"
+
+        for i, chunk in enumerate(chunks):
+            print(f"Adding chunk #{i} tokens={chunk.token_count}")
+
+        await self.collection.add(
+            documents=[chunk.json() for chunk in chunks],
+            embeddings=embeddings,
+            ids=[f"{c.document_id}:chunk-{i}" for i, c in enumerate(chunks)],
+        )
+
+    async def query(self, embedding: NDArray, k: int) -> QueryDocumentsResponse:
+        results = await self.collection.query(
+            query_embeddings=[embedding.tolist()],
+            n_results=k,
+            include=["documents", "distances"],
+        )
+        distances = results["distances"][0]
+        documents = results["documents"][0]
+
+        chunks = []
+        scores = []
+        for dist, doc in zip(distances, documents):
+            try:
+                doc = json.loads(doc)
+                chunk = Chunk(**doc)
+            except Exception:
+                import traceback
+
+                traceback.print_exc()
+                print(f"Failed to parse document: {doc}")
+                continue
+
+            chunks.append(chunk)
+            scores.append(1.0 / float(dist))
+
+        return QueryDocumentsResponse(chunks=chunks, scores=scores)
+
+
+class ChromaMemoryAdapter(Memory):
+    def __init__(self, url: str) -> None:
+        print(f"Initializing ChromaMemoryAdapter with url: {url}")
+        url = url.rstrip("/")
+        parsed = urlparse(url)
+
+        if parsed.path and parsed.path != "/":
+            raise ValueError("URL should not contain a path")
+
+        self.host = parsed.hostname
+        self.port = parsed.port
+
+        self.client = None
+        self.cache = {}
+
+    async def initialize(self) -> None:
+        try:
+            print(f"Connecting to Chroma server at: {self.host}:{self.port}")
+            self.client = await chromadb.AsyncHttpClient(host=self.host, port=self.port)
+        except Exception as e:
+            import traceback
+
+            traceback.print_exc()
+            raise RuntimeError("Could not connect to Chroma server") from e
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank:
+        bank_id = str(uuid.uuid4())
+        bank = MemoryBank(
+            bank_id=bank_id,
+            name=name,
+            config=config,
+            url=url,
+        )
+        collection = await self.client.create_collection(
+            name=bank_id,
+            metadata={"bank": bank.json()},
+        )
+        bank_index = BankWithIndex(
+            bank=bank, index=ChromaIndex(self.client, collection)
+        )
+        self.cache[bank_id] = bank_index
+        return bank
+
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
+        bank_index = await self._get_and_cache_bank_index(bank_id)
+        if bank_index is None:
+            return None
+        return bank_index.bank
+
+    async def _get_and_cache_bank_index(self, bank_id: str) -> Optional[BankWithIndex]:
+        if bank_id in self.cache:
+            return self.cache[bank_id]
+
+        collections = await self.client.list_collections()
+        for collection in collections:
+            if collection.name == bank_id:
+                print(collection.metadata)
+                bank = MemoryBank(**json.loads(collection.metadata["bank"]))
+                index = BankWithIndex(
+                    bank=bank,
+                    index=ChromaIndex(self.client, collection),
+                )
+                self.cache[bank_id] = index
+                return index
+
+        return None
+
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+        ttl_seconds: Optional[int] = None,
+    ) -> None:
+        index = await self._get_and_cache_bank_index(bank_id)
+        if not index:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        await index.insert_documents(documents)
+
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        index = await self._get_and_cache_bank_index(bank_id)
+        if not index:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        return await index.query_documents(query, params)

llama_stack/memory/adapters/pgvector/__init__.py

@@ -0,0 +1,15 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .config import PGVectorConfig
+
+
+async def get_adapter_impl(config: PGVectorConfig, _deps):
+    from .pgvector import PGVectorMemoryAdapter
+
+    impl = PGVectorMemoryAdapter(config)
+    await impl.initialize()
+    return impl

llama_stack/memory/adapters/pgvector/config.py

@@ -0,0 +1,17 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from llama_models.schema_utils import json_schema_type
+from pydantic import BaseModel, Field
+
+
+@json_schema_type
+class PGVectorConfig(BaseModel):
+    host: str = Field(default="localhost")
+    port: int = Field(default=5432)
+    db: str
+    user: str
+    password: str

llama_stack/memory/adapters/pgvector/pgvector.py

@@ -0,0 +1,234 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import uuid
+
+from typing import List, Tuple
+
+import psycopg2
+from numpy.typing import NDArray
+from psycopg2 import sql
+from psycopg2.extras import execute_values, Json
+from pydantic import BaseModel
+from llama_stack.memory.api import *  # noqa: F403
+
+
+from llama_stack.memory.common.vector_store import (
+    ALL_MINILM_L6_V2_DIMENSION,
+    BankWithIndex,
+    EmbeddingIndex,
+)
+
+from .config import PGVectorConfig
+
+
+def check_extension_version(cur):
+    cur.execute("SELECT extversion FROM pg_extension WHERE extname = 'vector'")
+    result = cur.fetchone()
+    return result[0] if result else None
+
+
+def upsert_models(cur, keys_models: List[Tuple[str, BaseModel]]):
+    query = sql.SQL(
+        """
+        INSERT INTO metadata_store (key, data)
+        VALUES %s
+        ON CONFLICT (key) DO UPDATE
+        SET data = EXCLUDED.data
+    """
+    )
+
+    values = [(key, Json(model.dict())) for key, model in keys_models]
+    execute_values(cur, query, values, template="(%s, %s)")
+
+
+def load_models(cur, keys: List[str], cls):
+    query = "SELECT key, data FROM metadata_store"
+    if keys:
+        placeholders = ",".join(["%s"] * len(keys))
+        query += f" WHERE key IN ({placeholders})"
+        cur.execute(query, keys)
+    else:
+        cur.execute(query)
+
+    rows = cur.fetchall()
+    return [cls(**row["data"]) for row in rows]
+
+
+class PGVectorIndex(EmbeddingIndex):
+    def __init__(self, bank: MemoryBank, dimension: int, cursor):
+        self.cursor = cursor
+        self.table_name = f"vector_store_{bank.name}"
+
+        self.cursor.execute(
+            f"""
+            CREATE TABLE IF NOT EXISTS {self.table_name} (
+                id TEXT PRIMARY KEY,
+                document JSONB,
+                embedding vector({dimension})
+            )
+        """
+        )
+
+    async def add_chunks(self, chunks: List[Chunk], embeddings: NDArray):
+        assert len(chunks) == len(
+            embeddings
+        ), f"Chunk length {len(chunks)} does not match embedding length {len(embeddings)}"
+
+        values = []
+        for i, chunk in enumerate(chunks):
+            print(f"Adding chunk #{i} tokens={chunk.token_count}")
+            values.append(
+                (
+                    f"{chunk.document_id}:chunk-{i}",
+                    Json(chunk.dict()),
+                    embeddings[i].tolist(),
+                )
+            )
+
+        query = sql.SQL(
+            f"""
+        INSERT INTO {self.table_name} (id, document, embedding)
+        VALUES %s
+        ON CONFLICT (id) DO UPDATE SET embedding = EXCLUDED.embedding, document = EXCLUDED.document
+    """
+        )
+        execute_values(self.cursor, query, values, template="(%s, %s, %s::vector)")
+
+    async def query(self, embedding: NDArray, k: int) -> QueryDocumentsResponse:
+        self.cursor.execute(
+            f"""
+        SELECT document, embedding <-> %s::vector AS distance
+        FROM {self.table_name}
+        ORDER BY distance
+        LIMIT %s
+    """,
+            (embedding.tolist(), k),
+        )
+        results = self.cursor.fetchall()
+
+        chunks = []
+        scores = []
+        for doc, dist in results:
+            chunks.append(Chunk(**doc))
+            scores.append(1.0 / float(dist))
+
+        return QueryDocumentsResponse(chunks=chunks, scores=scores)
+
+
+class PGVectorMemoryAdapter(Memory):
+    def __init__(self, config: PGVectorConfig) -> None:
+        print(f"Initializing PGVectorMemoryAdapter -> {config.host}:{config.port}")
+        self.config = config
+        self.cursor = None
+        self.conn = None
+        self.cache = {}
+
+    async def initialize(self) -> None:
+        try:
+            self.conn = psycopg2.connect(
+                host=self.config.host,
+                port=self.config.port,
+                database=self.config.db,
+                user=self.config.user,
+                password=self.config.password,
+            )
+            self.cursor = self.conn.cursor()
+
+            version = check_extension_version(self.cursor)
+            if version:
+                print(f"Vector extension version: {version}")
+            else:
+                raise RuntimeError("Vector extension is not installed.")
+
+            self.cursor.execute(
+                """
+                CREATE TABLE IF NOT EXISTS metadata_store (
+                    key TEXT PRIMARY KEY,
+                    data JSONB
+                )
+            """
+            )
+        except Exception as e:
+            import traceback
+
+            traceback.print_exc()
+            raise RuntimeError("Could not connect to PGVector database server") from e
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank:
+        bank_id = str(uuid.uuid4())
+        bank = MemoryBank(
+            bank_id=bank_id,
+            name=name,
+            config=config,
+            url=url,
+        )
+        upsert_models(
+            self.cursor,
+            [
+                (bank.bank_id, bank),
+            ],
+        )
+        index = BankWithIndex(
+            bank=bank,
+            index=PGVectorIndex(bank, ALL_MINILM_L6_V2_DIMENSION, self.cursor),
+        )
+        self.cache[bank_id] = index
+        return bank
+
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
+        bank_index = await self._get_and_cache_bank_index(bank_id)
+        if bank_index is None:
+            return None
+        return bank_index.bank
+
+    async def _get_and_cache_bank_index(self, bank_id: str) -> Optional[BankWithIndex]:
+        if bank_id in self.cache:
+            return self.cache[bank_id]
+
+        banks = load_models(self.cursor, [bank_id], MemoryBank)
+        if not banks:
+            return None
+
+        bank = banks[0]
+        index = BankWithIndex(
+            bank=bank,
+            index=PGVectorIndex(bank, ALL_MINILM_L6_V2_DIMENSION, self.cursor),
+        )
+        self.cache[bank_id] = index
+        return index
+
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+        ttl_seconds: Optional[int] = None,
+    ) -> None:
+        index = await self._get_and_cache_bank_index(bank_id)
+        if not index:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        await index.insert_documents(documents)
+
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        index = await self._get_and_cache_bank_index(bank_id)
+        if not index:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        return await index.query_documents(query, params)

llama_stack/memory/api/__init__.py

@@ -0,0 +1,7 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .api import *  # noqa: F401 F403

llama_stack/memory/api/api.py

@@ -0,0 +1,156 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+from typing import List, Optional, Protocol
+
+from llama_models.schema_utils import json_schema_type, webmethod
+
+from pydantic import BaseModel, Field
+from typing_extensions import Annotated
+
+from llama_models.llama3.api.datatypes import *  # noqa: F403
+
+
+@json_schema_type
+class MemoryBankDocument(BaseModel):
+    document_id: str
+    content: InterleavedTextMedia | URL
+    mime_type: str | None = None
+    metadata: Dict[str, Any] = Field(default_factory=dict)
+
+
+@json_schema_type
+class MemoryBankType(Enum):
+    vector = "vector"
+    keyvalue = "keyvalue"
+    keyword = "keyword"
+    graph = "graph"
+
+
+class VectorMemoryBankConfig(BaseModel):
+    type: Literal[MemoryBankType.vector.value] = MemoryBankType.vector.value
+    embedding_model: str
+    chunk_size_in_tokens: int
+    overlap_size_in_tokens: Optional[int] = None
+
+
+class KeyValueMemoryBankConfig(BaseModel):
+    type: Literal[MemoryBankType.keyvalue.value] = MemoryBankType.keyvalue.value
+
+
+class KeywordMemoryBankConfig(BaseModel):
+    type: Literal[MemoryBankType.keyword.value] = MemoryBankType.keyword.value
+
+
+class GraphMemoryBankConfig(BaseModel):
+    type: Literal[MemoryBankType.graph.value] = MemoryBankType.graph.value
+
+
+MemoryBankConfig = Annotated[
+    Union[
+        VectorMemoryBankConfig,
+        KeyValueMemoryBankConfig,
+        KeywordMemoryBankConfig,
+        GraphMemoryBankConfig,
+    ],
+    Field(discriminator="type"),
+]
+
+
+class Chunk(BaseModel):
+    content: InterleavedTextMedia
+    token_count: int
+    document_id: str
+
+
+@json_schema_type
+class QueryDocumentsResponse(BaseModel):
+    chunks: List[Chunk]
+    scores: List[float]
+
+
+@json_schema_type
+class QueryAPI(Protocol):
+    @webmethod(route="/query_documents")
+    def query_documents(
+        self,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse: ...
+
+
+@json_schema_type
+class MemoryBank(BaseModel):
+    bank_id: str
+    name: str
+    config: MemoryBankConfig
+    # if there's a pre-existing (reachable-from-distribution) store which supports QueryAPI
+    url: Optional[URL] = None
+
+
+class Memory(Protocol):
+    @webmethod(route="/memory_banks/create")
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank: ...
+
+    @webmethod(route="/memory_banks/list", method="GET")
+    async def list_memory_banks(self) -> List[MemoryBank]: ...
+
+    @webmethod(route="/memory_banks/get", method="GET")
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]: ...
+
+    @webmethod(route="/memory_banks/drop", method="DELETE")
+    async def drop_memory_bank(
+        self,
+        bank_id: str,
+    ) -> str: ...
+
+    # this will just block now until documents are inserted, but it should
+    # probably return a Job instance which can be polled for completion
+    @webmethod(route="/memory_bank/insert")
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+        ttl_seconds: Optional[int] = None,
+    ) -> None: ...
+
+    @webmethod(route="/memory_bank/update")
+    async def update_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+    ) -> None: ...
+
+    @webmethod(route="/memory_bank/query")
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse: ...
+
+    @webmethod(route="/memory_bank/documents/get", method="GET")
+    async def get_documents(
+        self,
+        bank_id: str,
+        document_ids: List[str],
+    ) -> List[MemoryBankDocument]: ...
+
+    @webmethod(route="/memory_bank/documents/delete", method="DELETE")
+    async def delete_documents(
+        self,
+        bank_id: str,
+        document_ids: List[str],
+    ) -> None: ...

llama_stack/memory/client.py

@@ -0,0 +1,196 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import asyncio
+import json
+import os
+from pathlib import Path
+
+from typing import Any, Dict, List, Optional
+
+import fire
+import httpx
+from termcolor import cprint
+
+from llama_stack.core.datatypes import RemoteProviderConfig
+
+from .api import *  # noqa: F403
+from .common.file_utils import data_url_from_file
+
+
+async def get_client_impl(config: RemoteProviderConfig, _deps: Any) -> Memory:
+    return MemoryClient(config.url)
+
+
+class MemoryClient(Memory):
+    def __init__(self, base_url: str):
+        self.base_url = base_url
+
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
+        async with httpx.AsyncClient() as client:
+            r = await client.get(
+                f"{self.base_url}/memory_banks/get",
+                params={
+                    "bank_id": bank_id,
+                },
+                headers={"Content-Type": "application/json"},
+                timeout=20,
+            )
+            r.raise_for_status()
+            d = r.json()
+            if not d:
+                return None
+            return MemoryBank(**d)
+
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank:
+        async with httpx.AsyncClient() as client:
+            r = await client.post(
+                f"{self.base_url}/memory_banks/create",
+                json={
+                    "name": name,
+                    "config": config.dict(),
+                    "url": url,
+                },
+                headers={"Content-Type": "application/json"},
+                timeout=20,
+            )
+            r.raise_for_status()
+            d = r.json()
+            if not d:
+                return None
+            return MemoryBank(**d)
+
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+    ) -> None:
+        async with httpx.AsyncClient() as client:
+            r = await client.post(
+                f"{self.base_url}/memory_bank/insert",
+                json={
+                    "bank_id": bank_id,
+                    "documents": [d.dict() for d in documents],
+                },
+                headers={"Content-Type": "application/json"},
+                timeout=20,
+            )
+            r.raise_for_status()
+
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        async with httpx.AsyncClient() as client:
+            r = await client.post(
+                f"{self.base_url}/memory_bank/query",
+                json={
+                    "bank_id": bank_id,
+                    "query": query,
+                    "params": params,
+                },
+                headers={"Content-Type": "application/json"},
+                timeout=20,
+            )
+            r.raise_for_status()
+            return QueryDocumentsResponse(**r.json())
+
+
+async def run_main(host: str, port: int, stream: bool):
+    client = MemoryClient(f"http://{host}:{port}")
+
+    # create a memory bank
+    bank = await client.create_memory_bank(
+        name="test_bank",
+        config=VectorMemoryBankConfig(
+            bank_id="test_bank",
+            embedding_model="dragon-roberta-query-2",
+            chunk_size_in_tokens=512,
+            overlap_size_in_tokens=64,
+        ),
+    )
+    cprint(json.dumps(bank.dict(), indent=4), "green")
+
+    retrieved_bank = await client.get_memory_bank(bank.bank_id)
+    assert retrieved_bank is not None
+    assert retrieved_bank.config.embedding_model == "dragon-roberta-query-2"
+
+    urls = [
+        "memory_optimizations.rst",
+        "chat.rst",
+        "llama3.rst",
+        "datasets.rst",
+        "qat_finetune.rst",
+        "lora_finetune.rst",
+    ]
+    documents = [
+        MemoryBankDocument(
+            document_id=f"num-{i}",
+            content=URL(
+                uri=f"https://raw.githubusercontent.com/pytorch/torchtune/main/docs/source/tutorials/{url}"
+            ),
+            mime_type="text/plain",
+        )
+        for i, url in enumerate(urls)
+    ]
+
+    this_dir = os.path.dirname(__file__)
+    files = [Path(this_dir).parent.parent / "CONTRIBUTING.md"]
+    documents += [
+        MemoryBankDocument(
+            document_id=f"num-{i}",
+            content=data_url_from_file(path),
+        )
+        for i, path in enumerate(files)
+    ]
+
+    # insert some documents
+    await client.insert_documents(
+        bank_id=bank.bank_id,
+        documents=documents,
+    )
+
+    # query the documents
+    response = await client.query_documents(
+        bank_id=bank.bank_id,
+        query=[
+            "How do I use Lora?",
+        ],
+    )
+    for chunk, score in zip(response.chunks, response.scores):
+        print(f"Score: {score}")
+        print(f"Chunk:\n========\n{chunk}\n========\n")
+
+    response = await client.query_documents(
+        bank_id=bank.bank_id,
+        query=[
+            "Tell me more about llama3 and torchtune",
+        ],
+    )
+    for chunk, score in zip(response.chunks, response.scores):
+        print(f"Score: {score}")
+        print(f"Chunk:\n========\n{chunk}\n========\n")
+
+
+def main(host: str, port: int, stream: bool = True):
+    asyncio.run(run_main(host, port, stream))
+
+
+if __name__ == "__main__":
+    fire.Fire(main)

llama_stack/memory/common/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.

llama_stack/memory/common/file_utils.py

@@ -0,0 +1,26 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import base64
+import mimetypes
+import os
+
+from llama_models.llama3.api.datatypes import URL
+
+
+def data_url_from_file(file_path: str) -> URL:
+    if not os.path.exists(file_path):
+        raise FileNotFoundError(f"File not found: {file_path}")
+
+    with open(file_path, "rb") as file:
+        file_content = file.read()
+
+    base64_content = base64.b64encode(file_content).decode("utf-8")
+    mime_type, _ = mimetypes.guess_type(file_path)
+
+    data_url = f"data:{mime_type};base64,{base64_content}"
+
+    return URL(uri=data_url)

llama_stack/memory/common/vector_store.py

@@ -0,0 +1,180 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+import base64
+import io
+import re
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional
+from urllib.parse import unquote
+
+import chardet
+import httpx
+import numpy as np
+from numpy.typing import NDArray
+from pypdf import PdfReader
+
+from llama_models.llama3.api.datatypes import *  # noqa: F403
+from llama_models.llama3.api.tokenizer import Tokenizer
+
+from llama_stack.memory.api import *  # noqa: F403
+
+ALL_MINILM_L6_V2_DIMENSION = 384
+
+EMBEDDING_MODEL = None
+
+
+def get_embedding_model() -> "SentenceTransformer":
+    global EMBEDDING_MODEL
+
+    if EMBEDDING_MODEL is None:
+        print("Loading sentence transformer")
+
+        from sentence_transformers import SentenceTransformer
+
+        EMBEDDING_MODEL = SentenceTransformer("all-MiniLM-L6-v2")
+
+    return EMBEDDING_MODEL
+
+
+def parse_data_url(data_url: str):
+    data_url_pattern = re.compile(
+        r"^"
+        r"data:"
+        r"(?P<mimetype>[\w/\-+.]+)"
+        r"(?P<charset>;charset=(?P<encoding>[\w-]+))?"
+        r"(?P<base64>;base64)?"
+        r",(?P<data>.*)"
+        r"$",
+        re.DOTALL,
+    )
+    match = data_url_pattern.match(data_url)
+    if not match:
+        raise ValueError("Invalid Data URL format")
+
+    parts = match.groupdict()
+    parts["is_base64"] = bool(parts["base64"])
+    return parts
+
+
+def content_from_data(data_url: str) -> str:
+    parts = parse_data_url(data_url)
+    data = parts["data"]
+
+    if parts["is_base64"]:
+        data = base64.b64decode(data)
+    else:
+        data = unquote(data)
+        encoding = parts["encoding"] or "utf-8"
+        data = data.encode(encoding)
+
+    encoding = parts["encoding"]
+    if not encoding:
+        detected = chardet.detect(data)
+        encoding = detected["encoding"]
+
+    mime_type = parts["mimetype"]
+    mime_category = mime_type.split("/")[0]
+    if mime_category == "text":
+        # For text-based files (including CSV, MD)
+        return data.decode(encoding)
+
+    elif mime_type == "application/pdf":
+        # For PDF and DOC/DOCX files, we can't reliably convert to string)
+        pdf_bytes = io.BytesIO(data)
+        pdf_reader = PdfReader(pdf_bytes)
+        return "\n".join([page.extract_text() for page in pdf_reader.pages])
+
+    else:
+        cprint("Could not extract content from data_url properly.", color="red")
+        return ""
+
+
+async def content_from_doc(doc: MemoryBankDocument) -> str:
+    if isinstance(doc.content, URL):
+        if doc.content.uri.startswith("data:"):
+            return content_from_data(doc.content.uri)
+        else:
+            async with httpx.AsyncClient() as client:
+                r = await client.get(doc.content.uri)
+                return r.text
+
+    return interleaved_text_media_as_str(doc.content)
+
+
+def make_overlapped_chunks(
+    document_id: str, text: str, window_len: int, overlap_len: int
+) -> List[Chunk]:
+    tokenizer = Tokenizer.get_instance()
+    tokens = tokenizer.encode(text, bos=False, eos=False)
+
+    chunks = []
+    for i in range(0, len(tokens), window_len - overlap_len):
+        toks = tokens[i : i + window_len]
+        chunk = tokenizer.decode(toks)
+        chunks.append(
+            Chunk(content=chunk, token_count=len(toks), document_id=document_id)
+        )
+
+    return chunks
+
+
+class EmbeddingIndex(ABC):
+    @abstractmethod
+    async def add_chunks(self, chunks: List[Chunk], embeddings: NDArray):
+        raise NotImplementedError()
+
+    @abstractmethod
+    async def query(self, embedding: NDArray, k: int) -> QueryDocumentsResponse:
+        raise NotImplementedError()
+
+
+@dataclass
+class BankWithIndex:
+    bank: MemoryBank
+    index: EmbeddingIndex
+
+    async def insert_documents(
+        self,
+        documents: List[MemoryBankDocument],
+    ) -> None:
+        model = get_embedding_model()
+        for doc in documents:
+            content = await content_from_doc(doc)
+            chunks = make_overlapped_chunks(
+                doc.document_id,
+                content,
+                self.bank.config.chunk_size_in_tokens,
+                self.bank.config.overlap_size_in_tokens
+                or (self.bank.config.chunk_size_in_tokens // 4),
+            )
+            embeddings = model.encode([x.content for x in chunks]).astype(np.float32)
+
+            await self.index.add_chunks(chunks, embeddings)
+
+    async def query_documents(
+        self,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        if params is None:
+            params = {}
+        k = params.get("max_chunks", 3)
+
+        def _process(c) -> str:
+            if isinstance(c, str):
+                return c
+            else:
+                return "<media>"
+
+        if isinstance(query, list):
+            query_str = " ".join([_process(c) for c in query])
+        else:
+            query_str = _process(query)
+
+        model = get_embedding_model()
+        query_vector = model.encode([query_str])[0].astype(np.float32)
+        return await self.index.query(query_vector, k)

llama_stack/memory/meta_reference/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.

llama_stack/memory/meta_reference/faiss/__init__.py

@@ -0,0 +1,19 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .config import FaissImplConfig
+
+
+async def get_provider_impl(config: FaissImplConfig, _deps):
+    from .faiss import FaissMemoryImpl
+
+    assert isinstance(
+        config, FaissImplConfig
+    ), f"Unexpected config type: {type(config)}"
+
+    impl = FaissMemoryImpl(config)
+    await impl.initialize()
+    return impl

llama_stack/memory/meta_reference/faiss/config.py

@@ -0,0 +1,13 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from llama_models.schema_utils import json_schema_type
+
+from pydantic import BaseModel
+
+
+@json_schema_type
+class FaissImplConfig(BaseModel): ...

llama_stack/memory/meta_reference/faiss/faiss.py

@@ -0,0 +1,124 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import logging
+import uuid
+
+from typing import Any, Dict, List, Optional
+
+import faiss
+import numpy as np
+from numpy.typing import NDArray
+
+from llama_models.llama3.api.datatypes import *  # noqa: F403
+
+from llama_stack.memory.api import *  # noqa: F403
+from llama_stack.memory.common.vector_store import (
+    ALL_MINILM_L6_V2_DIMENSION,
+    BankWithIndex,
+    EmbeddingIndex,
+)
+from llama_stack.telemetry import tracing
+from .config import FaissImplConfig
+
+logger = logging.getLogger(__name__)
+
+
+class FaissIndex(EmbeddingIndex):
+    id_by_index: Dict[int, str]
+    chunk_by_index: Dict[int, str]
+
+    def __init__(self, dimension: int):
+        self.index = faiss.IndexFlatL2(dimension)
+        self.id_by_index = {}
+        self.chunk_by_index = {}
+
+    @tracing.span(name="add_chunks")
+    async def add_chunks(self, chunks: List[Chunk], embeddings: NDArray):
+        indexlen = len(self.id_by_index)
+        for i, chunk in enumerate(chunks):
+            self.chunk_by_index[indexlen + i] = chunk
+            logger.info(f"Adding chunk #{indexlen + i} tokens={chunk.token_count}")
+            self.id_by_index[indexlen + i] = chunk.document_id
+
+        self.index.add(np.array(embeddings).astype(np.float32))
+
+    async def query(self, embedding: NDArray, k: int) -> QueryDocumentsResponse:
+        distances, indices = self.index.search(
+            embedding.reshape(1, -1).astype(np.float32), k
+        )
+
+        chunks = []
+        scores = []
+        for d, i in zip(distances[0], indices[0]):
+            if i < 0:
+                continue
+            chunks.append(self.chunk_by_index[int(i)])
+            scores.append(1.0 / float(d))
+
+        return QueryDocumentsResponse(chunks=chunks, scores=scores)
+
+
+class FaissMemoryImpl(Memory):
+    def __init__(self, config: FaissImplConfig) -> None:
+        self.config = config
+        self.cache = {}
+
+    async def initialize(self) -> None: ...
+
+    async def shutdown(self) -> None: ...
+
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank:
+        assert url is None, "URL is not supported for this implementation"
+        assert (
+            config.type == MemoryBankType.vector.value
+        ), f"Only vector banks are supported {config.type}"
+
+        bank_id = str(uuid.uuid4())
+        bank = MemoryBank(
+            bank_id=bank_id,
+            name=name,
+            config=config,
+            url=url,
+        )
+        index = BankWithIndex(bank=bank, index=FaissIndex(ALL_MINILM_L6_V2_DIMENSION))
+        self.cache[bank_id] = index
+        return bank
+
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
+        index = self.cache.get(bank_id)
+        if index is None:
+            return None
+        return index.bank
+
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+        ttl_seconds: Optional[int] = None,
+    ) -> None:
+        index = self.cache.get(bank_id)
+        if index is None:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        await index.insert_documents(documents)
+
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        index = self.cache.get(bank_id)
+        if index is None:
+            raise ValueError(f"Bank {bank_id} not found")
+
+        return await index.query_documents(query, params)

llama_stack/memory/providers.py

@@ -0,0 +1,45 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from typing import List
+
+from llama_stack.core.datatypes import *  # noqa: F403
+
+EMBEDDING_DEPS = [
+    "blobfile",
+    "chardet",
+    "pypdf",
+    "sentence-transformers",
+]
+
+
+def available_providers() -> List[ProviderSpec]:
+    return [
+        InlineProviderSpec(
+            api=Api.memory,
+            provider_id="meta-reference-faiss",
+            pip_packages=EMBEDDING_DEPS + ["faiss-cpu"],
+            module="llama_stack.memory.meta_reference.faiss",
+            config_class="llama_stack.memory.meta_reference.faiss.FaissImplConfig",
+        ),
+        remote_provider_spec(
+            Api.memory,
+            AdapterSpec(
+                adapter_id="chromadb",
+                pip_packages=EMBEDDING_DEPS + ["chromadb-client"],
+                module="llama_stack.memory.adapters.chroma",
+            ),
+        ),
+        remote_provider_spec(
+            Api.memory,
+            AdapterSpec(
+                adapter_id="pgvector",
+                pip_packages=EMBEDDING_DEPS + ["psycopg2-binary"],
+                module="llama_stack.memory.adapters.pgvector",
+                config_class="llama_stack.memory.adapters.pgvector.PGVectorConfig",
+            ),
+        ),
+    ]

llama_stack/memory/router/__init__.py

@@ -0,0 +1,17 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from typing import Any, List, Tuple
+
+from llama_stack.core.datatypes import Api
+
+
+async def get_router_impl(inner_impls: List[Tuple[str, Any]], deps: List[Api]):
+    from .router import MemoryRouterImpl
+
+    impl = MemoryRouterImpl(inner_impls, deps)
+    await impl.initialize()
+    return impl

llama_stack/memory/router/router.py

@@ -0,0 +1,91 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from typing import Any, Dict, List, Tuple
+
+from llama_stack.core.datatypes import Api
+from llama_stack.memory.api import *  # noqa: F403
+
+
+class MemoryRouterImpl(Memory):
+    """Routes to an provider based on the memory bank type"""
+
+    def __init__(
+        self,
+        inner_impls: List[Tuple[str, Any]],
+        deps: List[Api],
+    ) -> None:
+        self.deps = deps
+
+        bank_types = [v.value for v in MemoryBankType]
+
+        self.providers = {}
+        for routing_key, provider_impl in inner_impls:
+            if routing_key not in bank_types:
+                raise ValueError(
+                    f"Unknown routing key `{routing_key}` for memory bank type"
+                )
+            self.providers[routing_key] = provider_impl
+
+        self.bank_id_to_type = {}
+
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        for p in self.providers.values():
+            await p.shutdown()
+
+    def get_provider(self, bank_type):
+        if bank_type not in self.providers:
+            raise ValueError(f"Memory bank type {bank_type} not supported")
+
+        return self.providers[bank_type]
+
+    async def create_memory_bank(
+        self,
+        name: str,
+        config: MemoryBankConfig,
+        url: Optional[URL] = None,
+    ) -> MemoryBank:
+        provider = self.get_provider(config.type)
+        bank = await provider.create_memory_bank(name, config, url)
+        self.bank_id_to_type[bank.bank_id] = config.type
+        return bank
+
+    async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
+        bank_type = self.bank_id_to_type.get(bank_id)
+        if not bank_type:
+            raise ValueError(f"Could not find bank type for {bank_id}")
+
+        provider = self.get_provider(bank_type)
+        return await provider.get_memory_bank(bank_id)
+
+    async def insert_documents(
+        self,
+        bank_id: str,
+        documents: List[MemoryBankDocument],
+        ttl_seconds: Optional[int] = None,
+    ) -> None:
+        bank_type = self.bank_id_to_type.get(bank_id)
+        if not bank_type:
+            raise ValueError(f"Could not find bank type for {bank_id}")
+
+        provider = self.get_provider(bank_type)
+        return await provider.insert_documents(bank_id, documents, ttl_seconds)
+
+    async def query_documents(
+        self,
+        bank_id: str,
+        query: InterleavedTextMedia,
+        params: Optional[Dict[str, Any]] = None,
+    ) -> QueryDocumentsResponse:
+        bank_type = self.bank_id_to_type.get(bank_id)
+        if not bank_type:
+            raise ValueError(f"Could not find bank type for {bank_id}")
+
+        provider = self.get_provider(bank_type)
+        return await provider.query_documents(bank_id, query, params)