Merge pull request #46 from gongchensu/feature/support_hygon

Compatible with Hygon DCU.

README.md

@@ -15,13 +15,13 @@ xmake && xmake install
 - 运行模型推理测试
 
 ```bash
-python scripts/jiuge.py [--cpu | --nvidia | --cambricon | --ascend | --metax | --moore] path/to/model_dir [n_device]
+python scripts/jiuge.py [--cpu | --nvidia | --cambricon | --ascend | --metax | --moore | --iluvatar | --kunlun | --hygon] path/to/model_dir [n_device]
 ```
 
 - 部署模型推理服务
 
 ```bash
-python scripts/launch_server.py --model-path MODEL_PATH [-h] [--dev {cpu,nvidia,cambricon,ascend,metax,moore}] [--ndev NDEV] [--max-batch MAX_BATCH] [--max-tokens MAX_TOKENS]
+python scripts/launch_server.py --model-path MODEL_PATH [-h] [--dev {cpu,nvidia,cambricon,ascend,metax,moore,iluvatar,kunlun,hygon}] [--ndev NDEV] [--max-batch MAX_BATCH] [--max-tokens MAX_TOKENS]
 ```
 
 - 测试模型推理服务性能

scripts/jiuge.py

@@ -656,7 +656,7 @@ class JiugeForCauslLM:
 def test():
     if len(sys.argv) < 3:
         print(
-            "Usage: python jiuge.py [--cpu | --nvidia| --cambricon | --ascend | --metax | --moore] <path/to/model_dir> [n_device]"
+            "Usage: python jiuge.py [--cpu | --nvidia| --cambricon | --ascend | --metax | --moore | --iluvatar | --kunlun | --hygon] <path/to/model_dir> [n_device]"
         )
         sys.exit(1)
     model_path = sys.argv[2]
@@ -677,9 +677,11 @@ def test():
         device_type = DeviceType.DEVICE_TYPE_ILUVATAR
     elif sys.argv[1] == "--kunlun":
         device_type = DeviceType.DEVICE_TYPE_KUNLUN
+    elif sys.argv[1] == "--hygon":
+        device_type = DeviceType.DEVICE_TYPE_HYGON
     else:
         print(
-            "Usage: python jiuge.py [--cpu | --nvidia| --cambricon | --ascend | --metax | --moore] <path/to/model_dir> [n_device]"
+            "Usage: python jiuge.py [--cpu | --nvidia| --cambricon | --ascend | --metax | --moore | --iluvatar | --kunlun | --hygon] <path/to/model_dir> [n_device]"
         )
         sys.exit(1)
 

scripts/jiuge_ppl.py

@@ -11,6 +11,9 @@ DEVICE_TYPE_MAP = {
     "ascend": DeviceType.DEVICE_TYPE_ASCEND,
     "metax": DeviceType.DEVICE_TYPE_METAX,
     "moore": DeviceType.DEVICE_TYPE_MOORE,
+    "iluvatar": DeviceType.DEVICE_TYPE_ILUVATAR,
+    "kunlun": DeviceType.DEVICE_TYPE_KUNLUN,
+    "hygon": DeviceType.DEVICE_TYPE_HYGON,
 }
 
 TORCH_DEVICE_TYPE_MAP = {
@@ -20,6 +23,9 @@ TORCH_DEVICE_TYPE_MAP = {
     "ascend": "npu",
     "metax": "cuda",
     "moore": "cuda",
+    "iluvatar": "cuda",
+    "kunlun": "cuda",
+    "hygon": "cuda",
 }
 
 

scripts/launch_server.py

@@ -24,6 +24,9 @@ DEVICE_TYPE_MAP = {
     "ascend": DeviceType.DEVICE_TYPE_ASCEND,
     "metax": DeviceType.DEVICE_TYPE_METAX,
     "moore": DeviceType.DEVICE_TYPE_MOORE,
+    "iluvatar": DeviceType.DEVICE_TYPE_ILUVATAR,
+    "kunlun": DeviceType.DEVICE_TYPE_KUNLUN,
+    "hygon": DeviceType.DEVICE_TYPE_HYGON,
 }
 
 

scripts/libinfinicore_infer/base.py

@@ -35,6 +35,7 @@ class DeviceType(ctypes.c_int):
     DEVICE_TYPE_MOORE = 5
     DEVICE_TYPE_ILUVATAR = 6
     DEVICE_TYPE_KUNLUN = 7
+    DEVICE_TYPE_HYGON = 8
 
 
 class KVCacheCStruct(ctypes.Structure):