load config and model

hydra_node/config.py

+import logging
+import os
+import platform
+import socket
+import sys
+import time
+import sentry_sdk
+import torch
+from dotmap import DotMap
+from icecream import ic
+from sentry_sdk import capture_exception
+from sentry_sdk.integrations.threading import ThreadingIntegration
+
+def init_config_model():
+    config = DotMap()
+    config.model_type = "GPT"
+    is_dev = ""
+    environment = "production"
+    if os.environ['DEV'] == "True":
+        is_dev = "_dev"
+        environment = "staging"
+    config.is_dev = is_dev
+
+    # So we know all about the errors.
+    sentry_url = os.getenv("SENTRY_URL")
+    sentry_sdk.init(
+        sentry_url,
+        server_name=socket.gethostname(),
+        traces_sample_rate=0.002,
+        environment=environment,
+        integrations=[ThreadingIntegration(propagate_hub=True)],
+    )
+
+    # Setup logger
+    logger = logging.getLogger(__name__)
+    logger.setLevel(level=logging.INFO)
+    fh = logging.StreamHandler()
+    fh_formatter = logging.Formatter(
+        "%(asctime)s %(levelname)s %(filename)s(%(process)d) - %(message)s"
+    )
+    fh.setFormatter(fh_formatter)
+    logger.addHandler(fh)
+    config.logger = logger
+
+    # Gather node information
+    config.cuda_dev = torch.cuda.current_device()
+    cpu_id = platform.processor()
+    if os.path.exists('/proc/cpuinfo'):
+        cpu_id = [line for line in open("/proc/cpuinfo", 'r').readlines() if
+         'model name' in line][0].rstrip().split(': ')[-1]
+
+    config.cpu_id = cpu_id
+    config.gpu_id = torch.cuda.get_device_name(config.cuda_dev)
+    config.node_id = platform.node()
+
+    # Report on our CUDA memory and model.
+    gb_gpu = int(torch.cuda.get_device_properties(
+        config.cuda_dev).total_memory / (1000 * 1000 * 1000))
+    logger.info(f"CPU: {config.cpu_id}")
+    logger.info(f"GPU: {config.gpu_id}")
+    logger.info(f"GPU RAM: {gb_gpu}gb")
+
+    config.model_name = os.environ['MODEL']
+    logger.info(f"MODEL: {config.model_name}")
+
+    # Resolve where we get our model and data from.
+    config.model_path = os.getenv('MODEL_PATH', None)
+
+    # Misc settings
+    config.model_alias = os.getenv('MODEL_ALIAS')
+
+    # Instantiate our actual model.
+    load_time = time.time()
+
+    try:
+        model = GPTModel(config)
+    except Exception as e:
+        ic(e)
+        capture_exception(e)
+        logger.error("Failed to load model, restarting.")
+        sys.exit(1)
+
+    config.model = model
+
+    # Mark that our model is loaded.
+    f = open("/tmp/health_startup", "w")
+    f.close()
+    time_load = time.time() - load_time
+    logger.info(f"Models loaded in {time_load:.2f}s")
+
+    return model, config
\ No newline at end of file

main.py

 from fastapi import FastAPI
 from pydantic import BaseModel
 from fastapi.responses import HTMLResponse
+from sentry_sdk import capture_exception
+from sentry_sdk import capture_message
+from sentry_sdk import start_transaction
+from hydra_node.config import init_config_model
 app = FastAPI()
 
 @app.on_event("startup")
 async def startup_event():
-    print('Start')
+    print("Startup")
+    #model, config = init_config()
 
 @app.on_event("shutdown")
 async def shutdown_event():