cyberes
/
local-llm-server
Archived
1
1
Fork 0
Code Issues Pull Requests Packages Projects Releases 1 Wiki Activity

fix wrong datatype

This commit is contained in:
Cyberes 2023-10-15 17:24:18 -06:00
parent b3f0c4b28f
commit 3ec9b2347f
4 changed files with 21 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llm_server/routes/openai/chat_completions.py
View File

@ -102,8 +102,12 @@ def openai_chat_completions(model_name=None):
pubsub = redis.pubsub() pubsub = redis.pubsub()
pubsub.subscribe(event_id) pubsub.subscribe(event_id)
for item in pubsub.listen(): for item in pubsub.listen():
if item['type'] == 'message' and item['data'].decode('utf-8') == 'begin': if item['type'] == 'message':
break msg = item['data'].decode('utf-8')
if msg == 'begin':
break
elif msg == 'offline':
return return_invalid_model_err(handler.request_json_body['model'])
time.sleep(0.1) time.sleep(0.1)
# Double check the model is still online # Double check the model is still online

8
llm_server/routes/openai/completions.py
View File

@ -127,8 +127,12 @@ def openai_completions(model_name=None):
pubsub = redis.pubsub() pubsub = redis.pubsub()
pubsub.subscribe(event_id) pubsub.subscribe(event_id)
for item in pubsub.listen(): for item in pubsub.listen():
if item['type'] == 'message' and item['data'].decode('utf-8') == 'begin': if item['type'] == 'message':
break msg = item['data'].decode('utf-8')
if msg == 'begin':
break
elif msg == 'offline':
return return_invalid_model_err(handler.request_json_body['model'])
time.sleep(0.1) time.sleep(0.1)
# Double check the model is still online # Double check the model is still online

8
llm_server/routes/v1/generate_stream.py
View File

@ -143,8 +143,12 @@ def do_stream(ws, model_name):
pubsub = redis.pubsub() pubsub = redis.pubsub()
pubsub.subscribe(event_id) pubsub.subscribe(event_id)
for item in pubsub.listen(): for item in pubsub.listen():
if item['type'] == 'message' and item['data'].decode('utf-8') == 'begin': if item['type'] == 'message':
break msg = item['data'].decode('utf-8')
if msg == 'begin':
break
elif msg == 'offline':
return messages.BACKEND_OFFLINE, 404 # TODO: format this error
time.sleep(0.1) time.sleep(0.1)
# Double check the model is still online # Double check the model is still online

8
llm_server/workers/inferencer.py
View File

@ -2,7 +2,6 @@ import threading
import time import time
import traceback import traceback
from llm_server import messages
from llm_server.cluster.cluster_config import cluster_config from llm_server.cluster.cluster_config import cluster_config
from llm_server.custom_redis import redis from llm_server.custom_redis import redis
from llm_server.llm.generator import generator from llm_server.llm.generator import generator
@ -14,10 +13,11 @@ def worker(backend_url):
while True: while True:
(request_json_body, client_ip, token, parameters), event_id, selected_model = queue.get() (request_json_body, client_ip, token, parameters), event_id, selected_model = queue.get()
backend_info = cluster_config.get_backend(backend_url) backend_info = cluster_config.get_backend(backend_url)
pubsub = redis.pubsub()
pubsub.subscribe(event_id)
if not backend_info['online']: if not backend_info['online']:
event = DataEvent(event_id) redis.publish(event_id, 'offline')
event.set((False, None, messages.BACKEND_OFFLINE))
return return
if not selected_model: if not selected_model:
@ -34,8 +34,6 @@ def worker(backend_url):
# is finished. Since a lot of ratelimiting and stats are # is finished. Since a lot of ratelimiting and stats are
# based off the number of active workers, we must keep # based off the number of active workers, we must keep
# the generation based off the workers. # the generation based off the workers.
pubsub = redis.pubsub()
pubsub.subscribe(event_id)
redis.publish(event_id, 'begin') redis.publish(event_id, 'begin')
for item in pubsub.listen(): for item in pubsub.listen():
if item['type'] == 'message' and item['data'].decode('utf-8') == 'finished': if item['type'] == 'message' and item['data'].decode('utf-8') == 'finished':