Skip to content

Commit dce5d3a

Browse files
authored
chore(config): export accelerator literal type (#5384)
1 parent 409a9d8 commit dce5d3a

File tree

1 file changed

+33
-31
lines changed

1 file changed

+33
-31
lines changed

src/_bentoml_sdk/service/config.py

Lines changed: 33 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,37 @@
1717

1818
Posint = Annotated[int, Gt(0)]
1919
Posfloat = Annotated[float, Gt(0.0)]
20+
GpuLiteralType = Literal[
21+
"nvidia-backwell-gb200",
22+
"nvidia-backwell-b100",
23+
"nvidia-h200-141gb",
24+
"nvidia-tesla-h100",
25+
"nvidia-tesla-t4",
26+
"nvidia-tesla-a100",
27+
"nvidia-a100-80gb",
28+
"nvidia-h100-80gb",
29+
"nvidia-a10g",
30+
"nvidia-l4",
31+
"nvidia-tesla-v100",
32+
"nvidia-tesla-p100",
33+
"nvidia-tesla-k80",
34+
"nvidia-tesla-p4",
35+
]
36+
TpuLiteralType = Literal[
37+
"v4-2x2x1",
38+
"v4-2x2x2",
39+
"v4-2x2x4",
40+
"v4-2x4x4",
41+
"v5p-2x2x1",
42+
"v5p-2x2x2",
43+
"v5p-2x2x4",
44+
"v5p-2x4x4",
45+
"v5e-1x1",
46+
"v5e-2x2",
47+
"v5e-2x4",
48+
"v5e-4x4",
49+
"v5e-4x8",
50+
]
2051

2152

2253
class TrafficSchema(TypedDict, total=False):
@@ -61,37 +92,8 @@ class ResourceSchema(TypedDict, total=False):
6192
gpu type defined here is only a annotation, it will use as an recommendation choice of instance type when deploying this service to bentocloud
6293
gpu_type follows the naming convention of AWS EC2 GPU instances, GCP GPU instances etc.
6394
"""
64-
gpu_type: Literal[
65-
"nvidia-backwell-gb200",
66-
"nvidia-backwell-b100",
67-
"nvidia-h200-141gb",
68-
"nvidia-tesla-h100",
69-
"nvidia-tesla-t4",
70-
"nvidia-tesla-a100",
71-
"nvidia-a100-80gb",
72-
"nvidia-h100-80gb",
73-
"nvidia-a10g",
74-
"nvidia-l4",
75-
"nvidia-tesla-v100",
76-
"nvidia-tesla-p100",
77-
"nvidia-tesla-k80",
78-
"nvidia-tesla-p4",
79-
]
80-
tpu_type: Literal[
81-
"v4-2x2x1",
82-
"v4-2x2x2",
83-
"v4-2x2x4",
84-
"v4-2x4x4",
85-
"v5p-2x2x1",
86-
"v5p-2x2x2",
87-
"v5p-2x2x4",
88-
"v5p-2x4x4",
89-
"v5e-1x1",
90-
"v5e-2x2",
91-
"v5e-2x4",
92-
"v5e-4x4",
93-
"v5e-4x8",
94-
]
95+
gpu_type: GpuLiteralType
96+
tpu_type: TpuLiteralType
9597

9698

9799
WorkerSchema = Union[Posint, Literal["cpu_count"]]

0 commit comments

Comments
 (0)