mirror of
https://github.com/khoj-ai/khoj.git
synced 2024-11-29 10:23:02 +01:00
1168244c92
- Improve search speed by ~10x
Tested on corpus of 125K lines, 12.5K entries
- Allow cross-encoder to re-rank results by settings &?r=true when querying /search API
- It's an optional param that default to False
- Earlier all results were re-ranked by cross-encoder
- Making this configurable allows for much faster results, if desired
but for lower accuracy
309 lines
12 KiB
Python
309 lines
12 KiB
Python
# Standard Packages
|
|
import sys, json, yaml, os
|
|
import time
|
|
from typing import Optional
|
|
|
|
# External Packages
|
|
import uvicorn
|
|
import torch
|
|
from fastapi import FastAPI, Request
|
|
from fastapi.responses import HTMLResponse, FileResponse
|
|
from fastapi.staticfiles import StaticFiles
|
|
from fastapi.templating import Jinja2Templates
|
|
|
|
# Internal Packages
|
|
from src.search_type import image_search, text_search
|
|
from src.processor.org_mode.org_to_jsonl import org_to_jsonl
|
|
from src.processor.ledger.beancount_to_jsonl import beancount_to_jsonl
|
|
from src.processor.markdown.markdown_to_jsonl import markdown_to_jsonl
|
|
from src.utils.helpers import get_absolute_path, get_from_dict
|
|
from src.utils.cli import cli
|
|
from src.utils.config import SearchType, SearchModels, ProcessorConfigModel, ConversationProcessorConfigModel
|
|
from src.utils.rawconfig import FullConfig
|
|
from src.processor.conversation.gpt import converse, extract_search_type, message_to_log, message_to_prompt, understand, summarize
|
|
from src.search_filter.explicit_filter import ExplicitFilter
|
|
from src.search_filter.date_filter import DateFilter
|
|
|
|
# Application Global State
|
|
config = FullConfig()
|
|
model = SearchModels()
|
|
processor_config = ProcessorConfigModel()
|
|
config_file = ""
|
|
verbose = 0
|
|
app = FastAPI()
|
|
web_directory = f'src/interface/web/'
|
|
|
|
app.mount("/static", StaticFiles(directory=web_directory), name="static")
|
|
app.mount("/views", StaticFiles(directory="views"), name="views")
|
|
templates = Jinja2Templates(directory="views/")
|
|
|
|
@app.get("/", response_class=FileResponse)
|
|
def index():
|
|
return FileResponse(web_directory + "index.html")
|
|
|
|
@app.get('/ui', response_class=HTMLResponse)
|
|
def ui(request: Request):
|
|
return templates.TemplateResponse("config.html", context={'request': request})
|
|
|
|
@app.get('/config', response_model=FullConfig)
|
|
def config_data():
|
|
return config
|
|
|
|
@app.post('/config')
|
|
async def config_data(updated_config: FullConfig):
|
|
global config
|
|
config = updated_config
|
|
with open(config_file, 'w') as outfile:
|
|
yaml.dump(yaml.safe_load(config.json(by_alias=True)), outfile)
|
|
outfile.close()
|
|
return config
|
|
|
|
@app.get('/search')
|
|
def search(q: str, n: Optional[int] = 5, t: Optional[SearchType] = None, r: Optional[bool] = False):
|
|
if q is None or q == '':
|
|
print(f'No query param (q) passed in API call to initiate search')
|
|
return {}
|
|
|
|
device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
|
|
user_query = q
|
|
results_count = n
|
|
results = {}
|
|
|
|
if (t == SearchType.Org or t == None) and model.orgmode_search:
|
|
# query org-mode notes
|
|
query_start = time.time()
|
|
hits, entries = text_search.query(user_query, model.orgmode_search, rank_results=r, device=device, filters=[DateFilter(), ExplicitFilter()], verbose=verbose)
|
|
query_end = time.time()
|
|
|
|
# collate and return results
|
|
collate_start = time.time()
|
|
results = text_search.collate_results(hits, entries, results_count)
|
|
collate_end = time.time()
|
|
|
|
if (t == SearchType.Music or t == None) and model.music_search:
|
|
# query music library
|
|
query_start = time.time()
|
|
hits, entries = text_search.query(user_query, model.music_search, rank_results=r, device=device, filters=[DateFilter(), ExplicitFilter()], verbose=verbose)
|
|
query_end = time.time()
|
|
|
|
# collate and return results
|
|
collate_start = time.time()
|
|
results = text_search.collate_results(hits, entries, results_count)
|
|
collate_end = time.time()
|
|
|
|
if (t == SearchType.Markdown or t == None) and model.orgmode_search:
|
|
# query markdown files
|
|
query_start = time.time()
|
|
hits, entries = text_search.query(user_query, model.markdown_search, rank_results=r, device=device, filters=[ExplicitFilter(), DateFilter()], verbose=verbose)
|
|
query_end = time.time()
|
|
|
|
# collate and return results
|
|
collate_start = time.time()
|
|
results = text_search.collate_results(hits, entries, results_count)
|
|
collate_end = time.time()
|
|
|
|
if (t == SearchType.Ledger or t == None) and model.ledger_search:
|
|
# query transactions
|
|
query_start = time.time()
|
|
hits, entries = text_search.query(user_query, model.ledger_search, rank_results=r, device=device, filters=[ExplicitFilter(), DateFilter()], verbose=verbose)
|
|
query_end = time.time()
|
|
|
|
# collate and return results
|
|
collate_start = time.time()
|
|
results = text_search.collate_results(hits, entries, results_count)
|
|
collate_end = time.time()
|
|
|
|
if (t == SearchType.Image or t == None) and model.image_search:
|
|
# query images
|
|
query_start = time.time()
|
|
hits = image_search.query(user_query, results_count, model.image_search, verbose=verbose)
|
|
output_directory = f'{os.getcwd()}/{web_directory}'
|
|
query_end = time.time()
|
|
|
|
# collate and return results
|
|
collate_start = time.time()
|
|
results = image_search.collate_results(
|
|
hits,
|
|
image_names=model.image_search.image_names,
|
|
output_directory=output_directory,
|
|
static_files_url='/static',
|
|
count=results_count)
|
|
collate_end = time.time()
|
|
|
|
if verbose > 1:
|
|
print(f"Query took {query_end - query_start:.3f} seconds")
|
|
print(f"Collating results took {collate_end - collate_start:.3f} seconds")
|
|
|
|
return results
|
|
|
|
|
|
@app.get('/reload')
|
|
def regenerate(t: Optional[SearchType] = None):
|
|
global model
|
|
device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
|
|
model = initialize_search(config, regenerate=False, t=t, device=device)
|
|
return {'status': 'ok', 'message': 'reload completed'}
|
|
|
|
|
|
@app.get('/regenerate')
|
|
def regenerate(t: Optional[SearchType] = None):
|
|
global model
|
|
device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
|
|
model = initialize_search(config, regenerate=True, t=t, device=device)
|
|
return {'status': 'ok', 'message': 'regeneration completed'}
|
|
|
|
|
|
@app.get('/beta/search')
|
|
def search_beta(q: str, n: Optional[int] = 1):
|
|
# Extract Search Type using GPT
|
|
metadata = extract_search_type(q, api_key=processor_config.conversation.openai_api_key, verbose=verbose)
|
|
search_type = get_from_dict(metadata, "search-type")
|
|
|
|
# Search
|
|
search_results = search(q, n=n, t=SearchType(search_type))
|
|
|
|
# Return response
|
|
return {'status': 'ok', 'result': search_results, 'type': search_type}
|
|
|
|
|
|
@app.get('/chat')
|
|
def chat(q: str):
|
|
# Load Conversation History
|
|
chat_session = processor_config.conversation.chat_session
|
|
meta_log = processor_config.conversation.meta_log
|
|
|
|
# Converse with OpenAI GPT
|
|
metadata = understand(q, api_key=processor_config.conversation.openai_api_key, verbose=verbose)
|
|
if verbose > 1:
|
|
print(f'Understood: {get_from_dict(metadata, "intent")}')
|
|
|
|
if get_from_dict(metadata, "intent", "memory-type") == "notes":
|
|
query = get_from_dict(metadata, "intent", "query")
|
|
result_list = search(query, n=1, t=SearchType.Org)
|
|
collated_result = "\n".join([item["entry"] for item in result_list])
|
|
if verbose > 1:
|
|
print(f'Semantically Similar Notes:\n{collated_result}')
|
|
gpt_response = summarize(collated_result, summary_type="notes", user_query=q, api_key=processor_config.conversation.openai_api_key)
|
|
else:
|
|
gpt_response = converse(q, chat_session, api_key=processor_config.conversation.openai_api_key)
|
|
|
|
# Update Conversation History
|
|
processor_config.conversation.chat_session = message_to_prompt(q, chat_session, gpt_message=gpt_response)
|
|
processor_config.conversation.meta_log['chat'] = message_to_log(q, metadata, gpt_response, meta_log.get('chat', []))
|
|
|
|
return {'status': 'ok', 'response': gpt_response}
|
|
|
|
|
|
def initialize_search(config: FullConfig, regenerate: bool, t: SearchType = None, device=torch.device("cpu")):
|
|
# Initialize Org Notes Search
|
|
if (t == SearchType.Org or t == None) and config.content_type.org:
|
|
# Extract Entries, Generate Notes Embeddings
|
|
model.orgmode_search = text_search.setup(org_to_jsonl, config.content_type.org, search_config=config.search_type.asymmetric, regenerate=regenerate, device=device, verbose=verbose)
|
|
|
|
# Initialize Org Music Search
|
|
if (t == SearchType.Music or t == None) and config.content_type.music:
|
|
# Extract Entries, Generate Music Embeddings
|
|
model.music_search = text_search.setup(org_to_jsonl, config.content_type.music, search_config=config.search_type.asymmetric, regenerate=regenerate, device=device, verbose=verbose)
|
|
|
|
# Initialize Markdown Search
|
|
if (t == SearchType.Markdown or t == None) and config.content_type.markdown:
|
|
# Extract Entries, Generate Markdown Embeddings
|
|
model.markdown_search = text_search.setup(markdown_to_jsonl, config.content_type.markdown, search_config=config.search_type.asymmetric, regenerate=regenerate, device=device, verbose=verbose)
|
|
|
|
# Initialize Ledger Search
|
|
if (t == SearchType.Ledger or t == None) and config.content_type.ledger:
|
|
# Extract Entries, Generate Ledger Embeddings
|
|
model.ledger_search = text_search.setup(beancount_to_jsonl, config.content_type.ledger, search_config=config.search_type.symmetric, regenerate=regenerate, verbose=verbose)
|
|
|
|
# Initialize Image Search
|
|
if (t == SearchType.Image or t == None) and config.content_type.image:
|
|
# Extract Entries, Generate Image Embeddings
|
|
model.image_search = image_search.setup(config.content_type.image, search_config=config.search_type.image, regenerate=regenerate, verbose=verbose)
|
|
|
|
return model
|
|
|
|
|
|
def initialize_processor(config: FullConfig):
|
|
if not config.processor:
|
|
return
|
|
|
|
processor_config = ProcessorConfigModel()
|
|
|
|
# Initialize Conversation Processor
|
|
processor_config.conversation = ConversationProcessorConfigModel(config.processor.conversation, verbose)
|
|
|
|
conversation_logfile = processor_config.conversation.conversation_logfile
|
|
if processor_config.conversation.verbose:
|
|
print('INFO:\tLoading conversation logs from disk...')
|
|
|
|
if conversation_logfile.expanduser().absolute().is_file():
|
|
# Load Metadata Logs from Conversation Logfile
|
|
with open(get_absolute_path(conversation_logfile), 'r') as f:
|
|
processor_config.conversation.meta_log = json.load(f)
|
|
|
|
print('INFO:\tConversation logs loaded from disk.')
|
|
else:
|
|
# Initialize Conversation Logs
|
|
processor_config.conversation.meta_log = {}
|
|
processor_config.conversation.chat_session = ""
|
|
|
|
return processor_config
|
|
|
|
|
|
@app.on_event('shutdown')
|
|
def shutdown_event():
|
|
# No need to create empty log file
|
|
if not processor_config.conversation.meta_log:
|
|
return
|
|
elif processor_config.conversation.verbose:
|
|
print('INFO:\tSaving conversation logs to disk...')
|
|
|
|
# Summarize Conversation Logs for this Session
|
|
chat_session = processor_config.conversation.chat_session
|
|
openai_api_key = processor_config.conversation.openai_api_key
|
|
conversation_log = processor_config.conversation.meta_log
|
|
session = {
|
|
"summary": summarize(chat_session, summary_type="chat", api_key=openai_api_key),
|
|
"session-start": conversation_log.get("session", [{"session-end": 0}])[-1]["session-end"],
|
|
"session-end": len(conversation_log["chat"])
|
|
}
|
|
if 'session' in conversation_log:
|
|
conversation_log['session'].append(session)
|
|
else:
|
|
conversation_log['session'] = [session]
|
|
|
|
# Save Conversation Metadata Logs to Disk
|
|
conversation_logfile = get_absolute_path(processor_config.conversation.conversation_logfile)
|
|
with open(conversation_logfile, "w+", encoding='utf-8') as logfile:
|
|
json.dump(conversation_log, logfile)
|
|
|
|
print('INFO:\tConversation logs saved to disk.')
|
|
|
|
|
|
if __name__ == '__main__':
|
|
# Load config from CLI
|
|
args = cli(sys.argv[1:])
|
|
|
|
# Stores the file path to the config file.
|
|
config_file = args.config_file
|
|
|
|
# Store the verbose flag
|
|
verbose = args.verbose
|
|
|
|
# Store the raw config data.
|
|
config = args.config
|
|
|
|
# Set device to GPU if available
|
|
device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
|
|
|
|
# Initialize the search model from Config
|
|
model = initialize_search(args.config, args.regenerate, device=device)
|
|
|
|
# Initialize Processor from Config
|
|
processor_config = initialize_processor(args.config)
|
|
|
|
# Start Application Server
|
|
if args.socket:
|
|
uvicorn.run(app, proxy_headers=True, uds=args.socket)
|
|
else:
|
|
uvicorn.run(app, host=args.host, port=args.port)
|