[Hardware][Neuron] Refactor neuron support (#3471)

This commit is contained in:
Zhuohan Li
2024-03-21 18:22:17 -07:00
committed by GitHub
parent ea5f14e6ff
commit e90fc21f2e
33 changed files with 615 additions and 549 deletions

View File

@@ -325,7 +325,12 @@ class AsyncLLMEngine:
# Create the engine configs.
engine_configs = engine_args.create_engine_configs()
parallel_config = engine_configs[2]
if parallel_config.worker_use_ray or engine_args.engine_use_ray:
device_config = engine_configs[4]
if device_config.device_type == "neuron":
raise NotImplementedError("Neuron is not supported for "
"async engine yet.")
elif parallel_config.worker_use_ray or engine_args.engine_use_ray:
initialize_ray_cluster(parallel_config)
from vllm.executor.ray_gpu_executor import RayGPUExecutorAsync
executor_class = RayGPUExecutorAsync