concurrent gens setting, online status

config/config.yml

@@ -4,6 +4,7 @@ log_prompts: true
 
 mode: oobabooga
 auth_required: false
+concurrent_gens: 3
 
 backend_url: http://172.0.0.2:9104
 

llm_server/opts.py

@@ -1,7 +1,7 @@
 # Global settings that never change after startup
 
 running_model = 'none'
-concurrent_generates = 3
+concurrent_gens = 3
 mode = 'oobabooga'
 backend_url = None
 token_limit = 5555

llm_server/routes/stats.py

@@ -5,10 +5,10 @@ from threading import Semaphore, Thread
 
 from llm_server import opts
 from llm_server.integer import ThreadSafeInteger
-from llm_server.opts import concurrent_generates
+from llm_server.opts import concurrent_gens
 
 proompters_1_min = 0
-concurrent_semaphore = Semaphore(concurrent_generates)
+concurrent_semaphore = Semaphore(concurrent_gens)
 proompts = ThreadSafeInteger(0)
 start_time = datetime.now()
 
@@ -23,7 +23,7 @@ class SemaphoreCheckerThread(Thread):
     def run(self):
         global proompters_1_min
         while True:
-            self.values.append(opts.concurrent_generates - concurrent_semaphore._value)
+            self.values.append(opts.concurrent_gens - concurrent_semaphore._value)
             proompters_1_min = sum(self.values) / len(self.values)
             time.sleep(1)
 

llm_server/routes/v1/proxy.py

@@ -10,16 +10,25 @@ from .. import stats
 from ..cache import cache
 from ..helpers.http import cache_control
 from ..stats import proompters_1_min
+from ...llm.oobabooga.info import get_running_model
 
 
 @bp.route('/stats', methods=['GET'])
 @cache.cached(timeout=60, query_string=True)
 @cache_control(60)
 def get_stats():
+    model_list = get_running_model()
+    if isinstance(model_list, bool):
+        # get_running_model() will return False when the fetch fails
+        online = False
+    else:
+        online = True
+
     return jsonify({
-        'proompters_now': opts.concurrent_generates - concurrent_semaphore._value,
+        'proompters_now': opts.concurrent_gens - concurrent_semaphore._value,
         'proompters_1_min': proompters_1_min,
         'total_proompts': stats.proompts.value,
         'uptime': int((datetime.now() - stats.start_time).total_seconds()),
+        'online': online,
         'timestamp': int(time.time())
     }), 200

requirements.txt

@@ -4,4 +4,5 @@ pyyaml
 flask_caching
 requests
 tiktoken
-gunicorn
+gunicorn
+redis

server.py

@@ -22,7 +22,7 @@ if config_path_environ:
 else:
     config_path = Path(script_path, 'config', 'config.yml')
 
-default_vars = {'mode': 'oobabooga', 'log_prompts': False, 'database_path': './proxy-server.db', 'auth_required': False}
+default_vars = {'mode': 'oobabooga', 'log_prompts': False, 'database_path': './proxy-server.db', 'auth_required': False, 'concurrent_gens': 3}
 required_vars = []
 config_loader = ConfigLoader(config_path, default_vars, required_vars)
 success, config, msg = config_loader.load_config()
@@ -44,6 +44,7 @@ if config['mode'] not in ['oobabooga', 'hf-textgen']:
 opts.mode = config['mode']
 opts.auth_required = config['auth_required']
 opts.log_prompts = config['log_prompts']
+opts.concurrent_gens = config['concurrent_gens']
 
 opts.running_model = get_running_model()