quivr/backend/models/brains.py

207 lines
7.2 KiB
Python
Raw Normal View History

2023-06-28 20:39:27 +03:00
import os
from typing import Any, List, Optional
from uuid import UUID
from models.settings import CommonsDep, common_dependencies
2023-06-28 20:39:27 +03:00
from models.users import User
from pydantic import BaseModel
class Brain(BaseModel):
2023-06-28 20:39:27 +03:00
id: Optional[UUID] = None
name: Optional[str] = "Default brain"
status: Optional[str]= "public"
model: Optional[str] = "gpt-3.5-turbo-0613"
temperature: Optional[float] = 0.0
max_tokens: Optional[int] = 256
2023-06-28 20:39:27 +03:00
brain_size: Optional[float] = 0.0
max_brain_size: Optional[int] = int(os.getenv("MAX_BRAIN_SIZE", 0))
files: List[Any] = []
_commons: Optional[CommonsDep] = None
class Config:
arbitrary_types_allowed = True
@property
def commons(self) -> CommonsDep:
if not self._commons:
self.__class__._commons = common_dependencies()
return self._commons
2023-06-28 20:39:27 +03:00
@property
def brain_size(self):
self.get_unique_brain_files()
current_brain_size = sum(float(doc['size']) for doc in self.files)
print('current_brain_size', current_brain_size)
return current_brain_size
@property
def remaining_brain_size(self):
return float(self.max_brain_size) - self.brain_size
@classmethod
def create(cls, *args, **kwargs):
commons = common_dependencies()
return cls(commons=commons, *args, **kwargs)
def get_user_brains(self, user_id):
response = (
self.commons["supabase"]
.from_("brains_users")
.select("id:brain_id, brains (id: brain_id, name)")
.filter("user_id", "eq", user_id)
.execute()
)
return [item["brains"] for item in response.data]
def get_brain_details(self):
response = (
self.commons["supabase"]
.from_("brains")
.select("id:brain_id, name, *")
2023-06-28 20:39:27 +03:00
.filter("brain_id", "eq", self.id)
.execute()
)
return response.data
def delete_brain(self):
self.commons["supabase"].table("brains").delete().match(
2023-06-28 20:39:27 +03:00
{"brain_id": self.id}
).execute()
2023-06-28 20:39:27 +03:00
def create_brain(self):
commons = common_dependencies()
2023-06-28 20:39:27 +03:00
response = commons["supabase"].table("brains").insert({"name": self.name}).execute()
# set the brainId with response.data
self.id = response.data[0]['brain_id']
return response.data
2023-06-28 20:39:27 +03:00
def create_brain_user(self, user_id : UUID, rights, default_brain):
commons = common_dependencies()
response = commons["supabase"].table("brains_users").insert({"brain_id": str(self.id), "user_id":str( user_id), "rights": rights, "default_brain": default_brain}).execute()
return response.data
def create_brain_vector(self, vector_id):
response = (
self.commons["supabase"]
2023-06-28 20:39:27 +03:00
.table("brains_vectors")
.insert({"brain_id": str(self.id), "vector_id": str(vector_id)})
.execute()
)
return response.data
def get_vector_ids_from_file_sha1(self, file_sha1: str):
2023-06-28 20:39:27 +03:00
# move to vectors class
vectorsResponse = (
self.commons["supabase"]
.table("vectors")
.select("id")
.filter("metadata->>file_sha1", "eq", file_sha1)
.execute()
)
return vectorsResponse.data
def update_brain_fields(self):
self.commons["supabase"].table("brains").update({"name": self.name}).match(
2023-06-28 20:39:27 +03:00
{"brain_id": self.id}
).execute()
def update_brain_with_file(self, file_sha1: str):
2023-06-28 20:39:27 +03:00
# not used
vector_ids = self.get_vector_ids_from_file_sha1(file_sha1)
for vector_id in vector_ids:
self.create_brain_vector(vector_id)
2023-06-28 20:39:27 +03:00
def get_unique_brain_files(self):
"""
Retrieve unique brain data (i.e. uploaded files and crawled websites).
"""
response = (
self.commons["supabase"]
.from_("brains_vectors")
.select("vector_id")
.filter("brain_id", "eq", self.id)
.execute()
)
vector_ids = [item["vector_id"] for item in response.data]
print('vector_ids', vector_ids)
if len(vector_ids) == 0:
return []
self.files = self.get_unique_files_from_vector_ids(vector_ids)
print('unique_files', self.files)
return self.files
def get_unique_files_from_vector_ids(self, vectors_ids : List[int]):
# Move into Vectors class
"""
Retrieve unique user data vectors.
"""
vectors_response = self.commons['supabase'].table("vectors").select(
"name:metadata->>file_name, size:metadata->>file_size", count="exact") \
.filter("id", "in", tuple(vectors_ids))\
.execute()
documents = vectors_response.data # Access the data from the response
# Convert each dictionary to a tuple of items, then to a set to remove duplicates, and then back to a dictionary
unique_files = [dict(t) for t in set(tuple(d.items()) for d in documents)]
return unique_files
def delete_file_from_brain(self, file_name: str):
# First, get the vector_ids associated with the file_name
vector_response = self.commons["supabase"].table("vectors").select("id").filter("metadata->>file_name", "eq", file_name).execute()
vector_ids = [item["id"] for item in vector_response.data]
# For each vector_id, delete the corresponding entry from the 'brains_vectors' table
for vector_id in vector_ids:
self.commons["supabase"].table("brains_vectors").delete().filter("vector_id", "eq", vector_id).filter("brain_id", "eq", self.id).execute()
# Check if the vector is still associated with any other brains
associated_brains_response = self.commons["supabase"].table("brains_vectors").select("brain_id").filter("vector_id", "eq", vector_id).execute()
associated_brains = [item["brain_id"] for item in associated_brains_response.data]
# If the vector is not associated with any other brains, delete it from 'vectors' table
if not associated_brains:
self.commons["supabase"].table("vectors").delete().filter("id", "eq", vector_id).execute()
return {"message": f"File {file_name} in brain {self.id} has been deleted."}
def get_default_user_brain(user: User):
commons = common_dependencies()
response = (
commons["supabase"]
.from_("brains_users") # I'm assuming this is the correct table
.select("brain_id")
.filter("user_id", "eq", user.id)
.filter("default_brain", "eq", True) # Assuming 'default' is the correct column name
.execute()
)
default_brain_id = response.data[0]["brain_id"] if response.data else None
print(f"Default brain id: {default_brain_id}")
if default_brain_id:
brain_response = (
commons["supabase"]
.from_("brains")
.select("id:brain_id, name, *")
.filter("brain_id", "eq", default_brain_id)
.execute()
)
return brain_response.data[0] if brain_response.data else None
return None