Alpaca/src/connection_handler.py
2024-10-14 16:43:46 -06:00

157 lines
7.1 KiB
Python

# connection_handler.py
"""
Handles requests to remote and integrated instances of Ollama
"""
import json, os, requests, subprocess, threading, shutil
from .internal import data_dir, cache_dir
from logging import getLogger
from time import sleep
logger = getLogger(__name__)
window = None
AMD_support_label = "\n<a href='https://github.com/Jeffser/Alpaca/wiki/AMD-Support'>{}</a>".format(_('Alpaca Support'))
def log_output(pipe):
with open(os.path.join(data_dir, 'tmp.log'), 'a') as f:
with pipe:
try:
for line in iter(pipe.readline, ''):
print(line, end='')
f.write(line)
f.flush()
if 'msg="model request too large for system"' in line:
window.show_toast(_("Model request too large for system"), window.main_overlay)
elif 'msg="amdgpu detected, but no compatible rocm library found.' in line:
if bool(os.getenv("FLATPAK_ID")):
window.ollama_information_label.set_label(_("AMD GPU detected but the extension is missing, Ollama will use CPU.") + AMD_support_label)
else:
window.ollama_information_label.set_label(_("AMD GPU detected but ROCm is missing, Ollama will use CPU.") + AMD_support_label)
window.ollama_information_label.set_css_classes(['dim-label', 'error'])
elif 'msg="amdgpu is supported"' in line:
window.ollama_information_label.set_label(_("Using AMD GPU type '{}'").format(line.split('=')[-1]))
window.ollama_information_label.set_css_classes(['dim-label', 'success'])
except Exception as e:
pass
class instance():
def __init__(self, local_port:int, remote_url:str, remote:bool, tweaks:dict, overrides:dict, bearer_token:str, idle_timer_delay:int):
self.local_port=local_port
self.remote_url=remote_url
self.remote=remote
self.tweaks=tweaks
self.overrides=overrides
self.bearer_token=bearer_token
self.idle_timer_delay=idle_timer_delay
self.idle_timer_stop_event=threading.Event()
self.idle_timer=None
self.instance=None
self.busy=0
if not self.remote:
self.start()
def get_headers(self, include_json:bool) -> dict:
headers = {}
if include_json:
headers["Content-Type"] = "application/json"
if self.bearer_token and self.remote:
headers["Authorization"] = "Bearer " + self.bearer_token
return headers if len(headers.keys()) > 0 else None
def request(self, connection_type:str, connection_url:str, data:dict=None, callback:callable=None) -> requests.models.Response:
self.busy += 1
if self.idle_timer and not self.remote:
self.idle_timer_stop_event.set()
self.idle_timer=None
if not self.instance and not self.remote:
self.start()
connection_url = '{}/{}'.format(self.remote_url if self.remote else 'http://127.0.0.1:{}'.format(self.local_port), connection_url)
logger.info('{} : {}'.format(connection_type, connection_url))
response = None
match connection_type:
case "GET":
response = requests.get(connection_url, headers=self.get_headers(False))
case "POST":
if callback:
response = requests.post(connection_url, headers=self.get_headers(True), data=data, stream=True)
if response.status_code == 200:
for line in response.iter_lines():
if line:
callback(json.loads(line.decode("utf-8")))
else:
response = requests.post(connection_url, headers=self.get_headers(True), data=data, stream=False)
case "DELETE":
response = requests.delete(connection_url, headers=self.get_headers(False), data=data)
self.busy -= 1
if not self.idle_timer and not self.remote:
self.start_timer()
return response
def run_timer(self):
if not self.idle_timer_stop_event.wait(self.idle_timer_delay*60):
window.show_toast(_("Ollama instance was shut down due to inactivity"), window.main_overlay)
self.stop()
def start_timer(self):
if self.busy == 0:
if self.idle_timer:
self.idle_timer_stop_event.set()
self.idle_timer=None
if self.idle_timer_delay > 0 and self.busy == 0:
self.idle_timer_stop_event.clear()
self.idle_timer = threading.Thread(target=self.run_timer)
self.idle_timer.start()
def start(self):
self.stop()
if shutil.which('ollama'):
if not os.path.isdir(os.path.join(cache_dir, 'tmp/ollama')):
os.mkdir(os.path.join(cache_dir, 'tmp/ollama'))
self.instance = None
params = self.overrides.copy()
params["OLLAMA_DEBUG"] = "1"
params["OLLAMA_HOST"] = f"127.0.0.1:{self.local_port}" # You can't change this directly sorry :3
params["HOME"] = data_dir
params["TMPDIR"] = os.path.join(cache_dir, 'tmp/ollama')
instance = subprocess.Popen(["ollama", "serve"], env={**os.environ, **params}, stderr=subprocess.PIPE, stdout=subprocess.PIPE, text=True)
threading.Thread(target=log_output, args=(instance.stdout,)).start()
threading.Thread(target=log_output, args=(instance.stderr,)).start()
logger.info("Starting Alpaca's Ollama instance...")
logger.debug(params)
logger.info("Started Alpaca's Ollama instance")
try:
v_str = subprocess.check_output("ollama -v", shell=True).decode('utf-8')
logger.info(v_str.split('\n')[1].strip('Warning: ').strip())
except Exception as e:
logger.error(e)
self.instance = instance
if not self.idle_timer:
self.start_timer()
window.ollama_information_label.set_label(_("Integrated Ollama instance is running"))
window.ollama_information_label.set_css_classes(['dim-label', 'success'])
else:
self.remote = True
window.remote_connection_switch.set_sensitive(True)
window.remote_connection_switch.set_active(True)
def stop(self):
if self.idle_timer:
self.idle_timer_stop_event.set()
self.idle_timer=None
if self.instance:
logger.info("Stopping Alpaca's Ollama instance")
self.instance.terminate()
self.instance.wait()
self.instance = None
window.ollama_information_label.set_label(_("Integrated Ollama instance is not running"))
window.ollama_information_label.set_css_classes(['dim-label'])
logger.info("Stopped Alpaca's Ollama instance")
def reset(self):
logger.info("Resetting Alpaca's Ollama instance")
self.stop()
sleep(1)
self.start()