INNER CODE UNIT · Python
build_falcon_app
tensorchord/modelz-llm · src/modelz_llm/falcon_service.py:109
def build_falcon_app(args: argparse.Namespace):
if args.dry_run:
snapshot_download(repo_id=args.model)
snapshot_download(repo_id=args.emb_model)
return
llm_uds_path = UNIX_DOMAIN_SOCKET_PATH.format("llm")
emb_uds_path = UNIX_DOMAIN_SOCKET_PATH.format("emb")
barrier = mp.get_context("spawn").Barrier(3)
model_proc = run_server(
llm_uds_path, barrier, LLM, model_name=args.model, device=args.device
)
emb_proc = run_server(
emb_uds_path, barrier, Emb, model_name=args.emb_model, device=args.device
)
barrier.wait()
# wait to detect the proc exitcode
time.sleep(1)