phoenix-oss/llama-stack
1
0
Fork 0
forked from phoenix-oss/llama-stack-mirror
Code Pull requests Releases Packages 2 Activity Actions

Make embedding generation go through inference (#606)

Browse source 
This PR does the following:
1) adds the ability to generate embeddings in all supported inference
providers.
2) Moves all the memory providers to use the inference API and improved
the memory tests to setup the inference stack correctly and use the
embedding models

This is a merge from #589 and #598
This commit is contained in:
Dinesh Yeduguru 2024-12-12 11:47:50 -08:00 • committed by GitHub
parent a14785af46
commit 96e158eaac
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
37 changed files with 677 additions and 156 deletions
Download patch file Download diff file Expand all files Collapse all files

8
llama_stack/providers/remote/memory/pgvector/__init__.py
View file

@ -4,12 +4,16 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import Dict
from llama_stack.providers.datatypes import Api, ProviderSpec
from .config import PGVectorConfig
async def get_adapter_impl(config: PGVectorConfig, _deps):
async def get_adapter_impl(config: PGVectorConfig, deps: Dict[Api, ProviderSpec]):
from .pgvector import PGVectorMemoryAdapter
impl = PGVectorMemoryAdapter(config)
impl = PGVectorMemoryAdapter(config, deps[Api.inference])
await impl.initialize()
return impl

38
llama_stack/providers/remote/memory/pgvector/pgvector.py
View file

@ -16,9 +16,9 @@ from pydantic import BaseModel, parse_obj_as
from llama_stack.apis.memory import * # noqa: F403
from llama_stack.providers.datatypes import MemoryBanksProtocolPrivate
from llama_stack.providers.datatypes import Api, MemoryBanksProtocolPrivate
from llama_stack.providers.utils.memory.vector_store import (
ALL_MINILM_L6_V2_DIMENSION,
BankWithIndex,
EmbeddingIndex,
)
@ -120,8 +120,9 @@ class PGVectorIndex(EmbeddingIndex):
class PGVectorMemoryAdapter(Memory, MemoryBanksProtocolPrivate):
def __init__(self, config: PGVectorConfig) -> None:
def __init__(self, config: PGVectorConfig, inference_api: Api.inference) -> None:
self.config = config
self.inference_api = inference_api
self.cursor = None
self.conn = None
self.cache = {}
@ -160,27 +161,17 @@ class PGVectorMemoryAdapter(Memory, MemoryBanksProtocolPrivate):
async def shutdown(self) -> None:
pass
async def register_memory_bank(
self,
memory_bank: MemoryBank,
) -> None:
async def register_memory_bank(self, memory_bank: MemoryBank) -> None:
assert (
memory_bank.memory_bank_type == MemoryBankType.vector.value
), f"Only vector banks are supported {memory_bank.memory_bank_type}"
upsert_models(
self.cursor,
[
(memory_bank.identifier, memory_bank),
],
upsert_models(self.cursor, [(memory_bank.identifier, memory_bank)])
index = PGVectorIndex(memory_bank, memory_bank.embedding_dimension, self.cursor)
self.cache[memory_bank.identifier] = BankWithIndex(
memory_bank, index, self.inference_api
)
index = BankWithIndex(
bank=memory_bank,
index=PGVectorIndex(memory_bank, ALL_MINILM_L6_V2_DIMENSION, self.cursor),
)
self.cache[memory_bank.identifier] = index
async def unregister_memory_bank(self, memory_bank_id: str) -> None:
await self.cache[memory_bank_id].index.delete()
del self.cache[memory_bank_id]
@ -203,14 +194,13 @@ class PGVectorMemoryAdapter(Memory, MemoryBanksProtocolPrivate):
index = await self._get_and_cache_bank_index(bank_id)
return await index.query_documents(query, params)
self.inference_api = inference_api
async def _get_and_cache_bank_index(self, bank_id: str) -> BankWithIndex:
if bank_id in self.cache:
return self.cache[bank_id]
bank = await self.memory_bank_store.get_memory_bank(bank_id)
index = BankWithIndex(
bank=bank,
index=PGVectorIndex(bank, ALL_MINILM_L6_V2_DIMENSION, self.cursor),
)
self.cache[bank_id] = index
return index
index = PGVectorIndex(bank, bank.embedding_dimension, self.cursor)
self.cache[bank_id] = BankWithIndex(bank, index, self.inference_api)
return self.cache[bank_id]