unitorch.cli.fastapis
InfoFastAPI
Tip
core/fastapi/info is the section for configuration of InfoFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/info.py
14
15
16
17
18
19
20
21
22 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/info")
self._section = f"core/fastapi/info"
router = config.getoption("router", "/core/fastapi/info")
self._device = config.getoption("device", "cpu")
self._router = APIRouter(prefix=router)
self._router.add_api_route("/status", self.status, methods=["GET"])
self._lock = asyncio.Lock()
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/info'
_device
instance-attribute
_device = getoption('device', 'cpu')
_router
instance-attribute
_router = APIRouter(prefix=router)
start
Source code in src/unitorch/cli/fastapis/info.py
| def start(self):
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/info.py
| def stop(self):
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/info.py
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69 | def status(self):
mem_info = psutil.virtual_memory()
stats = {
"cpu": {
"total": mem_info[0] / 1024**3,
"free": mem_info[1] / 1024**3,
"used": mem_info[3] / 1024**3,
}
}
if self._device != "cpu":
if isinstance(self._device, list):
for device in self._device:
free, total = torch.cuda.mem_get_info(device)
total = total / 1024**3
free = free / 1024**3
used = total - free
stats = {
**stats,
**{
f"cuda:{device}": {
"total": total,
"free": free,
"used": used,
}
},
}
else:
free, total = torch.cuda.mem_get_info(self._device)
total = total / 1024**3
free = free / 1024**3
used = total - free
stats = {
**stats,
**{"cuda": {"total": total, "free": free, "used": used}},
}
return stats
|
BRIAFastAPI
Tip
core/fastapi/bria is the section for configuration of BRIAFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/bria.py
79
80
81
82
83
84
85
86
87
88
89 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/bria")
self._section = f"core/fastapi/bria"
router = config.getoption("router", "/core/fastapi/bria")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/bria'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
Source code in src/unitorch/cli/fastapis/bria.py
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120 | def start(self):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
BRIAForSegmentationPipeline.from_config(
self.config,
pretrained_weight_path="https://huggingface.co/datasets/fuliucansheng/hubfiles/resolve/main/bria_rmbg2.0_pytorch_model.bin",
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/bria.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/bria.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(image: UploadFile, threshold: float = 0.5)
Source code in src/unitorch/cli/fastapis/bria.py
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150 | async def generate(
self,
image: UploadFile,
threshold: float = 0.5,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
mask = pipe(image, threshold=threshold)
finally:
pipe.release()
buffer = io.BytesIO()
mask.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
ClipForClassificationFastAPI
Tip
core/fastapi/clip is the section for configuration of ClipForClassificationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/clip.py
860
861
862
863
864
865
866
867
868
869
870 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/clip")
self._section = "core/fastapi/clip"
router = config.getoption("router", "/core/fastapi/clip")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/clip'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'clip-vit-base-patch16')
Source code in src/unitorch/cli/fastapis/clip.py
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901 | def start(self, pretrained_name: str = "clip-vit-base-patch16"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
ClipForClassificationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/clip.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/clip.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: UploadFile,
max_seq_length: Optional[int] = 512,
)
Source code in src/unitorch/cli/fastapis/clip.py
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930 | async def generate(
self,
text: str,
image: UploadFile,
max_seq_length: Optional[int] = 512,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result = pipe(
text,
image,
max_seq_length=max_seq_length,
)
finally:
pipe.release()
return result
|
ClipForTextClassificationFastAPI
Tip
core/fastapi/clip/text is the section for configuration of ClipForTextClassificationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/clip.py
935
936
937
938
939
940
941
942
943
944
945 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/clip/text")
self._section = "core/fastapi/clip/text"
router = config.getoption("router", "/core/fastapi/clip/text")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/clip/text'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'clip-vit-base-patch16')
Source code in src/unitorch/cli/fastapis/clip.py
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976 | def start(self, pretrained_name: str = "clip-vit-base-patch16"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
ClipForTextClassificationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/clip.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/clip.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(text: str, max_seq_length: Optional[int] = 512)
Source code in src/unitorch/cli/fastapis/clip.py
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001 | async def generate(
self,
text: str,
max_seq_length: Optional[int] = 512,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
result = pipe(
text,
max_seq_length=max_seq_length,
)
finally:
pipe.release()
return result
|
ClipForImageClassificationFastAPI
Tip
core/fastapi/clip/image is the section for configuration of ClipForImageClassificationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/clip.py
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/clip/image")
self._section = "core/fastapi/clip/image"
router = config.getoption("router", "/core/fastapi/clip/image")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/clip/image'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'clip-vit-base-patch16')
Source code in src/unitorch/cli/fastapis/clip.py
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047 | def start(self, pretrained_name: str = "clip-vit-base-patch16"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
ClipForImageClassificationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/clip.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/clip.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(image: UploadFile)
Source code in src/unitorch/cli/fastapis/clip.py
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070 | async def generate(
self,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result = pipe(image)
finally:
pipe.release()
return result
|
Tip
core/fastapi/clip/matching is the section for configuration of ClipForMatchingFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/clip.py
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/clip/matching")
self._section = "core/fastapi/clip/matching"
router = config.getoption("router", "/core/fastapi/clip/matching")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
_section = 'core/fastapi/clip/matching'
_router = APIRouter(prefix=router)
start(pretrained_name: str = 'clip-vit-base-patch16')
Source code in src/unitorch/cli/fastapis/clip.py
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184 | def start(self, pretrained_name: str = "clip-vit-base-patch16"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
ClipForMatchingPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
Source code in src/unitorch/cli/fastapis/clip.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
Source code in src/unitorch/cli/fastapis/clip.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate(
text: str,
image: UploadFile,
max_seq_length: Optional[int] = 77,
)
Source code in src/unitorch/cli/fastapis/clip.py
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213 | async def generate(
self,
text: str,
image: UploadFile,
max_seq_length: Optional[int] = 77,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result = pipe(
text,
image,
max_seq_length=max_seq_length,
)
finally:
pipe.release()
return result
|
DetrForDetectionFastAPI
Tip
core/fastapi/detr is the section for configuration of DetrForDetectionFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/detr.py
152
153
154
155
156
157
158
159
160
161
162 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/detr")
self._section = "core/fastapi/detr"
router = config.getoption("router", "/core/fastapi/detr")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/detr'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: Optional[str] = 'detr-resnet-50')
Source code in src/unitorch/cli/fastapis/detr.py
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193 | def start(self, pretrained_name: Optional[str] = "detr-resnet-50"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
DetrForDetectionPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/detr.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/detr.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(image: UploadFile, threshold: float = 0.5)
Source code in src/unitorch/cli/fastapis/detr.py
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223 | async def generate(
self,
image: UploadFile,
threshold: float = 0.5,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result_image = pipe(image, threshold=threshold)
finally:
pipe.release()
buffer = io.BytesIO()
result_image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
DPTForDepthEstimationFastAPI
Tip
core/fastapi/dpt is the section for configuration of DPTForDepthEstimationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/dpt.py
131
132
133
134
135
136
137
138
139
140
141 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/dpt")
self._section = "core/fastapi/dpt"
router = config.getoption("router", "/core/fastapi/dpt")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/dpt'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: Optional[str] = 'dpt-large')
Source code in src/unitorch/cli/fastapis/dpt.py
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172 | def start(self, pretrained_name: Optional[str] = "dpt-large"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
DPTForDepthEstimationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/dpt.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/dpt.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(image: UploadFile)
Source code in src/unitorch/cli/fastapis/dpt.py
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201 | async def generate(
self,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result_image = pipe(image)
finally:
pipe.release()
buffer = io.BytesIO()
result_image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
GemmaFastAPI
Tip
core/fastapi/gemma is the section for configuration of GemmaFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/gemma.py
192
193
194
195
196
197
198
199
200
201
202 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/gemma")
self._section = "core/fastapi/gemma"
router = config.getoption("router", "/core/fastapi/gemma")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/gemma'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'gemma-4-12b')
Source code in src/unitorch/cli/fastapis/gemma.py
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233 | def start(self, pretrained_name: str = "gemma-4-12b"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
GemmaForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/gemma.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/gemma.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
use_chat_template: Optional[bool] = False,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 2,
decoder_end_token_id: Optional[
Union[int, List[int]]
] = 1,
decoder_pad_token_id: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/gemma.py
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295 | async def generate(
self,
text: str,
use_chat_template: Optional[bool] = False,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 2,
decoder_end_token_id: Optional[Union[int, List[int]]] = 1,
decoder_pad_token_id: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None, "Service not started. Call /start first."
pipe = self._pipes.acquire()
try:
outputs = pipe(
prompt=text,
use_chat_template=use_chat_template,
max_seq_length=max_seq_length,
num_beams=num_beams,
decoder_start_token_id=decoder_start_token_id,
decoder_end_token_id=decoder_end_token_id,
decoder_pad_token_id=decoder_pad_token_id,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
early_stopping=early_stopping,
length_penalty=length_penalty,
num_beam_groups=num_beam_groups,
diversity_penalty=diversity_penalty,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return outputs
|
GemmaVLFastAPI
Tip
core/fastapi/gemma_vl is the section for configuration of GemmaVLFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/gemma_vl.py
214
215
216
217
218
219
220
221
222
223
224 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/gemma_vl")
self._section = "core/fastapi/gemma_vl"
router = config.getoption("router", "/core/fastapi/gemma_vl")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/gemma_vl'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'gemma-4-12b')
Source code in src/unitorch/cli/fastapis/gemma_vl.py
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255 | def start(self, pretrained_name: str = "gemma-4-12b"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
GemmaVLForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/gemma_vl.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/gemma_vl.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: Optional[UploadFile] = File(default=None),
use_chat_template: Optional[bool] = False,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 2,
decoder_end_token_id: Optional[
Union[int, List[int]]
] = 1,
decoder_pad_token_id: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/gemma_vl.py
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325 | async def generate(
self,
text: str,
image: Optional[UploadFile] = File(default=None),
use_chat_template: Optional[bool] = False,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 2,
decoder_end_token_id: Optional[Union[int, List[int]]] = 1,
decoder_pad_token_id: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None, "Service not started. Call /start first."
pil_image = None
if image is not None:
content = await image.read()
pil_image = Image.open(io.BytesIO(content)).convert("RGB")
pipe = self._pipes.acquire()
try:
outputs = pipe(
prompt=text,
images=[pil_image] if pil_image is not None else [],
use_chat_template=use_chat_template,
max_seq_length=max_seq_length,
num_beams=num_beams,
decoder_start_token_id=decoder_start_token_id,
decoder_end_token_id=decoder_end_token_id,
decoder_pad_token_id=decoder_pad_token_id,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
early_stopping=early_stopping,
length_penalty=length_penalty,
num_beam_groups=num_beam_groups,
diversity_penalty=diversity_penalty,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return outputs
|
GroundingDinoForDetectionFastAPI
Tip
core/fastapi/grounding_dino is the section for configuration of GroundingDinoForDetectionFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/grounding_dino.py
180
181
182
183
184
185
186
187
188
189
190 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/grounding_dino")
self._section = "core/fastapi/grounding_dino"
router = config.getoption("router", "/core/fastapi/grounding_dino")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/grounding_dino'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(
pretrained_name: Optional[str] = "grounding-dino-tiny",
)
Source code in src/unitorch/cli/fastapis/grounding_dino.py
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221 | def start(self, pretrained_name: Optional[str] = "grounding-dino-tiny"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
GroundingDinoForDetectionPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/grounding_dino.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/grounding_dino.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: UploadFile,
text_threshold: float = 0.25,
box_threshold: float = 0.25,
)
Source code in src/unitorch/cli/fastapis/grounding_dino.py
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258 | async def generate(
self,
text: str,
image: UploadFile,
text_threshold: float = 0.25,
box_threshold: float = 0.25,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result_image = pipe(
text,
image,
text_threshold=text_threshold,
box_threshold=box_threshold,
)
finally:
pipe.release()
buffer = io.BytesIO()
result_image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
LlamaForGenerationFastAPI
Tip
core/fastapi/llama is the section for configuration of LlamaForGenerationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/llama.py
217
218
219
220
221
222
223
224
225
226
227 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/llama")
self._section = "core/fastapi/llama"
router = config.getoption("router", "/core/fastapi/llama")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/llama'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'llama-3.2-1b-instruct')
Source code in src/unitorch/cli/fastapis/llama.py
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258 | def start(self, pretrained_name: str = "llama-3.2-1b-instruct"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
LlamaForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/llama.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/llama.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
prompt: str,
max_seq_length: Optional[int] = 512,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 1,
decoder_end_token_id: Optional[
Union[int, List[int]]
] = [2],
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/llama.py
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316 | async def generate(
self,
prompt: str,
max_seq_length: Optional[int] = 512,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 1,
decoder_end_token_id: Optional[Union[int, List[int]]] = [2],
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
result = pipe(
prompt,
max_seq_length=max_seq_length,
num_beams=num_beams,
decoder_start_token_id=decoder_start_token_id,
decoder_end_token_id=decoder_end_token_id,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
early_stopping=early_stopping,
length_penalty=length_penalty,
num_beam_groups=num_beam_groups,
diversity_penalty=diversity_penalty,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return result
|
LlavaMistralClipFastAPI
Tip
core/fastapi/llava/mistral_clip is the section for configuration of LlavaMistralClipFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/llava.py
444
445
446
447
448
449
450
451
452
453
454 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/llava/mistral_clip")
self._section = f"core/fastapi/llava/mistral_clip"
router = config.getoption("router", "/core/fastapi/llava/mistral_clip")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/llava/mistral_clip'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
Source code in src/unitorch/cli/fastapis/llava.py
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485 | def start(self):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
LlavaMistralClipForGenerationPipeline.from_config(
self.config,
pretrained_name="llava-v1.6-mistral-7b-hf",
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/llava.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/llava.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(text: str, image: UploadFile)
Source code in src/unitorch/cli/fastapis/llava.py
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514 | async def generate(
self,
text: str,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
text = f"[INST] <image>\n {text} [/INST]"
pipe = self._pipes.acquire()
try:
caption = pipe(
text,
image,
)
finally:
pipe.release()
return caption
|
LlavaLlamaSiglipFastAPI
Tip
core/fastapi/llava/joycaption2 is the section for configuration of LlavaLlamaSiglipFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/llava.py
519
520
521
522
523
524
525
526
527
528
529 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/llava/joycaption2")
self._section = f"core/fastapi/llava/joycaption2"
router = config.getoption("router", "/core/fastapi/llava/joycaption2")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/llava/joycaption2'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
Source code in src/unitorch/cli/fastapis/llava.py
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560 | def start(self):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
LlavaLlamaSiglipForGenerationPipeline.from_config(
self.config,
pretrained_name="llava-v1.6-joycaption-2",
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/llava.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/llava.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(text: str, image: UploadFile)
Source code in src/unitorch/cli/fastapis/llava.py
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589 | async def generate(
self,
text: str,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
text = f"<|start_header_id|>system<|end_header_id|>\\n\\nCutting Knowledge Date: December 2023\\nToday Date: 26 July 2024\\n\\nYou are a helpful image captioner.<|eot_id|><|start_header_id|>user<|end_header_id|>\\n\\n<|reserved_special_token_70|><|reserved_special_token_69|><|reserved_special_token_71|>{text}|eot_id|><|start_header_id|>assistant<|end_header_id|>\\n\\n"
pipe = self._pipes.acquire()
try:
caption = pipe(
text,
image,
)
finally:
pipe.release()
return caption
|
Tip
core/fastapi/mask2former is the section for configuration of Mask2FormerForSegmentationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/mask2former.py
142
143
144
145
146
147
148
149
150
151
152 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/mask2former")
self._section = "core/fastapi/mask2former"
router = config.getoption("router", "/core/fastapi/mask2former")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
_section = 'core/fastapi/mask2former'
_router = APIRouter(prefix=router)
start(
pretrained_name: Optional[
str
] = "mask2former-swin-tiny-ade-semantic",
)
Source code in src/unitorch/cli/fastapis/mask2former.py
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185 | def start(
self, pretrained_name: Optional[str] = "mask2former-swin-tiny-ade-semantic"
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
Mask2FormerForSegmentationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
Source code in src/unitorch/cli/fastapis/mask2former.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
Source code in src/unitorch/cli/fastapis/mask2former.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate(image: UploadFile)
Source code in src/unitorch/cli/fastapis/mask2former.py
196
197
198
199
200
201
202
203
204
205
206
207
208 | async def generate(
self,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
results = pipe(image)
finally:
pipe.release()
return [(mask.tolist(), label) for mask, label in results]
|
MistralForGenerationFastAPI
Tip
core/fastapi/mistral is the section for configuration of MistralForGenerationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/mistral.py
219
220
221
222
223
224
225
226
227
228
229 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/mistral")
self._section = "core/fastapi/mistral"
router = config.getoption("router", "/core/fastapi/mistral")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/mistral'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'mistral-7b-instruct-v0.1')
Source code in src/unitorch/cli/fastapis/mistral.py
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260 | def start(self, pretrained_name: str = "mistral-7b-instruct-v0.1"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
MistralForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/mistral.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/mistral.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
prompt: str,
max_seq_length: Optional[int] = 512,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 1,
decoder_end_token_id: Optional[
Union[int, List[int]]
] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/mistral.py
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318 | async def generate(
self,
prompt: str,
max_seq_length: Optional[int] = 512,
num_beams: Optional[int] = 2,
decoder_start_token_id: Optional[int] = 1,
decoder_end_token_id: Optional[Union[int, List[int]]] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
repetition_penalty: Optional[float] = 1.0,
no_repeat_ngram_size: Optional[int] = 0,
early_stopping: Optional[bool] = True,
length_penalty: Optional[float] = 1.0,
num_beam_groups: Optional[int] = 1,
diversity_penalty: Optional[float] = 0.0,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
result = pipe(
prompt,
max_seq_length=max_seq_length,
num_beams=num_beams,
decoder_start_token_id=decoder_start_token_id,
decoder_end_token_id=decoder_end_token_id,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
early_stopping=early_stopping,
length_penalty=length_penalty,
num_beam_groups=num_beam_groups,
diversity_penalty=diversity_penalty,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return result
|
QWen3FastAPI
Tip
core/fastapi/qwen3 is the section for configuration of QWen3FastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/qwen.py
216
217
218
219
220
221
222
223
224
225
226 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/qwen3")
self._section = f"core/fastapi/qwen3"
router = config.getoption("router", "/core/fastapi/qwen3")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/qwen3'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'qwen3-4b-thinking')
Source code in src/unitorch/cli/fastapis/qwen.py
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257 | def start(self, pretrained_name: str = "qwen3-4b-thinking"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWen3ForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/qwen.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/qwen.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
use_chat_template: Optional[bool] = True,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/qwen.py
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301 | async def generate(
self,
text: str,
use_chat_template: Optional[bool] = True,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
result = pipe(
text,
use_chat_template=use_chat_template,
max_seq_length=max_seq_length,
num_beams=num_beams,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return result
|
QWen3VLFastAPI
Tip
core/fastapi/qwen3_vl is the section for configuration of QWen3VLFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/qwen_vl.py
233
234
235
236
237
238
239
240
241
242
243 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/qwen3_vl")
self._section = f"core/fastapi/qwen3_vl"
router = config.getoption("router", "/core/fastapi/qwen3_vl")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/qwen3_vl'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'qwen3-vl-8b-instruct')
Source code in src/unitorch/cli/fastapis/qwen_vl.py
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274 | def start(self, pretrained_name: str = "qwen3-vl-8b-instruct"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWen3VLForGenerationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/qwen_vl.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/qwen_vl.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: UploadFile = File(...),
use_chat_template: Optional[bool] = True,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
)
Source code in src/unitorch/cli/fastapis/qwen_vl.py
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322 | async def generate(
self,
text: str,
image: UploadFile = File(...),
use_chat_template: Optional[bool] = True,
max_seq_length: Optional[int] = 12800,
num_beams: Optional[int] = 2,
num_return_sequences: Optional[int] = 1,
min_gen_seq_length: Optional[int] = 0,
max_gen_seq_length: Optional[int] = 512,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
):
assert self._pipes is not None
image = await image.read()
image = Image.open(io.BytesIO(image)).convert("RGB")
pipe = self._pipes.acquire()
try:
result = pipe(
text,
images=image,
use_chat_template=use_chat_template,
max_seq_length=max_seq_length,
num_beams=num_beams,
num_return_sequences=num_return_sequences,
min_gen_seq_length=min_gen_seq_length,
max_gen_seq_length=max_gen_seq_length,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
)
finally:
pipe.release()
return result
|
SamForSegmentationFastAPI
Tip
core/fastapi/sam is the section for configuration of SamForSegmentationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/sam.py
192
193
194
195
196
197
198
199
200
201
202 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/sam")
self._section = "core/fastapi/sam"
router = config.getoption("router", "/core/fastapi/sam")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/sam'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: Optional[str] = 'sam-vit-base')
Source code in src/unitorch/cli/fastapis/sam.py
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233 | def start(self, pretrained_name: Optional[str] = "sam-vit-base"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
SamForSegmentationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/sam.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/sam.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
image: UploadFile,
points: Optional[List] = None,
boxes: Optional[List] = None,
mask_threshold: float = 0.1,
)
Source code in src/unitorch/cli/fastapis/sam.py
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277 | async def generate(
self,
image: UploadFile,
points: Optional[List] = None,
boxes: Optional[List] = None,
mask_threshold: float = 0.1,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
mask_image = pipe(
image,
points=points,
boxes=boxes,
mask_threshold=mask_threshold,
)
finally:
pipe.release()
if mask_image is None:
return StreamingResponse(
io.BytesIO(),
media_type="image/png",
)
buffer = io.BytesIO()
mask_image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
Tip
core/fastapi/segformer is the section for configuration of SegformerForSegmentationFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/segformer.py
150
151
152
153
154
155
156
157
158
159
160 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/segformer")
self._section = "core/fastapi/segformer"
router = config.getoption("router", "/core/fastapi/segformer")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
_section = 'core/fastapi/segformer'
_router = APIRouter(prefix=router)
start(
pretrained_name: Optional[
str
] = "segformer-swin-tiny-ade-semantic",
)
Source code in src/unitorch/cli/fastapis/segformer.py
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193 | def start(
self, pretrained_name: Optional[str] = "segformer-swin-tiny-ade-semantic"
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
SegformerForSegmentationPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
Source code in src/unitorch/cli/fastapis/segformer.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
Source code in src/unitorch/cli/fastapis/segformer.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate(image: UploadFile)
Source code in src/unitorch/cli/fastapis/segformer.py
204
205
206
207
208
209
210
211
212
213
214
215
216 | async def generate(
self,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
results = pipe(image)
finally:
pipe.release()
return [(mask.tolist(), label) for mask, label in results]
|
Tip
core/fastapi/siglip is the section for configuration of Siglip2ForMatchingFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/siglip.py
156
157
158
159
160
161
162
163
164
165
166 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/siglip")
self._section = "core/fastapi/siglip"
router = config.getoption("router", "/core/fastapi/siglip")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
_section = 'core/fastapi/siglip'
_router = APIRouter(prefix=router)
start(pretrained_name: str = 'siglip-base-patch16-224')
Source code in src/unitorch/cli/fastapis/siglip.py
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197 | def start(self, pretrained_name: str = "siglip-base-patch16-224"):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
Siglip2ForMatchingPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
Source code in src/unitorch/cli/fastapis/siglip.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
Source code in src/unitorch/cli/fastapis/siglip.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate(text: str, image: UploadFile)
Source code in src/unitorch/cli/fastapis/siglip.py
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225 | async def generate(
self,
text: str,
image: UploadFile,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
result = pipe(
text,
image,
)
finally:
pipe.release()
return result
|
WanForText2VideoFastAPI
Tip
core/fastapi/wan/text2video is the section for configuration of WanForText2VideoFastAPI.
Use examples/configs/fastapis/wan.ini for a Wan-only FastAPI setup backed by the unified wan-v2.2-ti2v-5b checkpoint.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/wan/text2video.py
319
320
321
322
323
324
325
326
327
328
329 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/wan/text2video")
self._section = f"core/fastapi/wan/text2video"
router = config.getoption("router", "/core/fastapi/wan/text2video")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["POST"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/wan/text2video'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(
pretrained_name: Optional[str] = None,
pretrained_lora_names: Optional[
Union[str, List[str]]
] = None,
pretrained_lora_weights: Optional[
Union[float, List[float]]
] = 1.0,
pretrained_lora_alphas: Optional[
Union[float, List[float]]
] = 32.0,
)
Source code in src/unitorch/cli/fastapis/wan/text2video.py
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369 | def start(
self,
pretrained_name: Optional[str] = None,
pretrained_lora_names: Optional[Union[str, List[str]]] = None,
pretrained_lora_weights: Optional[Union[float, List[float]]] = 1.0,
pretrained_lora_alphas: Optional[Union[float, List[float]]] = 32.0,
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
WanForText2VideoFastAPIPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
pretrained_lora_names=pretrained_lora_names,
pretrained_lora_weights=pretrained_lora_weights,
pretrained_lora_alphas=pretrained_lora_alphas,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/wan/text2video.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/wan/text2video.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
neg_text: Optional[str] = "",
height: Optional[int] = 480,
width: Optional[int] = 832,
num_frames: Optional[int] = 81,
num_fps: Optional[int] = 16,
guidance_scale: Optional[float] = 5.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
)
Source code in src/unitorch/cli/fastapis/wan/text2video.py
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416 | async def generate(
self,
text: str,
neg_text: Optional[str] = "",
height: Optional[int] = 480,
width: Optional[int] = 832,
num_frames: Optional[int] = 81,
num_fps: Optional[int] = 16,
guidance_scale: Optional[float] = 5.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
video = pipe(
text,
neg_text=neg_text,
height=height,
width=width,
num_frames=num_frames,
num_fps=num_fps,
guidance_scale=guidance_scale,
num_timesteps=num_timesteps,
seed=seed,
)
finally:
pipe.release()
buffer = io.BytesIO()
with open(video, "rb") as f:
buffer.write(f.read())
buffer.seek(0)
return StreamingResponse(
buffer,
media_type="video/mp4",
headers={"Content-Disposition": "attachment; filename=output.mp4"},
)
|
WanForImage2VideoFastAPI
Tip
core/fastapi/wan/image2video is the section for configuration of WanForImage2VideoFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/wan/image2video.py
330
331
332
333
334
335
336
337
338
339
340 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/wan/image2video")
self._section = f"core/fastapi/wan/image2video"
router = config.getoption("router", "/core/fastapi/wan/image2video")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["POST"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/wan/image2video'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(
pretrained_name: Optional[str] = None,
pretrained_lora_names: Optional[
Union[str, List[str]]
] = None,
pretrained_lora_weights: Optional[
Union[float, List[float]]
] = 1.0,
pretrained_lora_alphas: Optional[
Union[float, List[float]]
] = 32.0,
)
Source code in src/unitorch/cli/fastapis/wan/image2video.py
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380 | def start(
self,
pretrained_name: Optional[str] = None,
pretrained_lora_names: Optional[Union[str, List[str]]] = None,
pretrained_lora_weights: Optional[Union[float, List[float]]] = 1.0,
pretrained_lora_alphas: Optional[Union[float, List[float]]] = 32.0,
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
WanForImage2VideoFastAPIPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
pretrained_lora_names=pretrained_lora_names,
pretrained_lora_weights=pretrained_lora_weights,
pretrained_lora_alphas=pretrained_lora_alphas,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/wan/image2video.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/wan/image2video.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: UploadFile,
neg_text: Optional[str] = "",
num_frames: Optional[int] = 81,
num_fps: Optional[int] = 16,
guidance_scale: Optional[float] = 5.0,
strength: Optional[float] = 1.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
)
Source code in src/unitorch/cli/fastapis/wan/image2video.py
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429 | async def generate(
self,
text: str,
image: UploadFile,
neg_text: Optional[str] = "",
num_frames: Optional[int] = 81,
num_fps: Optional[int] = 16,
guidance_scale: Optional[float] = 5.0,
strength: Optional[float] = 1.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
video = pipe(
text,
image,
neg_text=neg_text,
num_frames=num_frames,
num_fps=num_fps,
guidance_scale=guidance_scale,
strength=strength,
num_timesteps=num_timesteps,
seed=seed,
)
finally:
pipe.release()
buffer = io.BytesIO()
with open(video, "rb") as f:
buffer.write(f.read())
buffer.seek(0)
return StreamingResponse(
buffer,
media_type="video/mp4",
headers={"Content-Disposition": "attachment; filename=output.mp4"},
)
|
QWenImageText2ImageFastAPI
Tip
core/fastapi/qwen_image/text2image is the section for configuration of QWenImageText2ImageFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/qwen_image/text2image.py
345
346
347
348
349
350
351
352
353
354
355 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/qwen_image/text2image")
self._section = f"core/fastapi/qwen_image/text2image"
router = config.getoption("router", "/core/fastapi/qwen_image/text2image")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["GET"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["POST"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/qwen_image/text2image'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(
pretrained_name: Optional[str] = "qwen-image",
pretrained_lora_names: Optional[
Union[str, List[str]]
] = None,
pretrained_lora_weights: Optional[
Union[float, List[float]]
] = 1.0,
pretrained_lora_alphas: Optional[
Union[float, List[float]]
] = 32.0,
)
Source code in src/unitorch/cli/fastapis/qwen_image/text2image.py
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395 | def start(
self,
pretrained_name: Optional[str] = "qwen-image",
pretrained_lora_names: Optional[Union[str, List[str]]] = None,
pretrained_lora_weights: Optional[Union[float, List[float]]] = 1.0,
pretrained_lora_alphas: Optional[Union[float, List[float]]] = 32.0,
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWenImageForText2ImageFastAPIPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
pretrained_lora_names=pretrained_lora_names,
pretrained_lora_weights=pretrained_lora_weights,
pretrained_lora_alphas=pretrained_lora_alphas,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/qwen_image/text2image.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/qwen_image/text2image.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
height: Optional[int] = 512,
width: Optional[int] = 512,
guidance_scale: Optional[float] = 4.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
)
Source code in src/unitorch/cli/fastapis/qwen_image/text2image.py
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434 | async def generate(
self,
text: str,
height: Optional[int] = 512,
width: Optional[int] = 512,
guidance_scale: Optional[float] = 4.0,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
):
assert self._pipes is not None
pipe = self._pipes.acquire()
try:
image = pipe(
text,
height=height,
width=width,
guidance_scale=guidance_scale,
num_timesteps=num_timesteps,
seed=seed,
)
finally:
pipe.release()
buffer = io.BytesIO()
image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
QWenImageEditingFastAPI
Tip
core/fastapi/qwen_image/editing is the section for configuration of QWenImageEditingFastAPI.
Bases: GenericFastAPI
Source code in src/unitorch/cli/fastapis/qwen_image/image_editing.py
366
367
368
369
370
371
372
373
374
375
376 | def __init__(self, config: Config):
self.config = config
config.set_default_section(f"core/fastapi/qwen_image/editing")
self._section = f"core/fastapi/qwen_image/editing"
router = config.getoption("router", "/core/fastapi/qwen_image/editing")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["POST"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = f'core/fastapi/qwen_image/editing'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(
pretrained_name: Optional[str] = "qwen-image-editing",
pretrained_lora_names: Optional[
Union[str, List[str]]
] = None,
pretrained_lora_weights: Optional[
Union[float, List[float]]
] = 1.0,
pretrained_lora_alphas: Optional[
Union[float, List[float]]
] = 32.0,
)
Source code in src/unitorch/cli/fastapis/qwen_image/image_editing.py
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416 | def start(
self,
pretrained_name: Optional[str] = "qwen-image-editing",
pretrained_lora_names: Optional[Union[str, List[str]]] = None,
pretrained_lora_weights: Optional[Union[float, List[float]]] = 1.0,
pretrained_lora_alphas: Optional[Union[float, List[float]]] = 32.0,
):
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWenImageForImageEditingFastAPIPipeline.from_config(
self.config,
pretrained_name=pretrained_name,
pretrained_lora_names=pretrained_lora_names,
pretrained_lora_weights=pretrained_lora_weights,
pretrained_lora_alphas=pretrained_lora_alphas,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Source code in src/unitorch/cli/fastapis/qwen_image/image_editing.py
| def stop(self):
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Source code in src/unitorch/cli/fastapis/qwen_image/image_editing.py
| def status(self):
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: UploadFile,
height: Optional[int] = 512,
width: Optional[int] = 512,
guidance_scale: Optional[float] = 2.5,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
)
Source code in src/unitorch/cli/fastapis/qwen_image/image_editing.py
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459 | async def generate(
self,
text: str,
image: UploadFile,
height: Optional[int] = 512,
width: Optional[int] = 512,
guidance_scale: Optional[float] = 2.5,
num_timesteps: Optional[int] = 50,
seed: Optional[int] = 1123,
):
assert self._pipes is not None
image_bytes = await image.read()
image = Image.open(io.BytesIO(image_bytes))
pipe = self._pipes.acquire()
try:
image = pipe(
text,
image=image,
height=height,
width=width,
guidance_scale=guidance_scale,
num_timesteps=num_timesteps,
seed=seed,
)
finally:
pipe.release()
buffer = io.BytesIO()
image.save(buffer, format="PNG")
return StreamingResponse(
io.BytesIO(buffer.getvalue()),
media_type="image/png",
)
|
QWen3VLLMFastAPI
Tip
core/fastapi/vllm/qwen3 is the section for configuration of QWen3VLLMFastAPI.
Bases: GenericFastAPI
FastAPI service for QWen3 text generation powered by vLLM.
Exposes /generate, /status, /start, and /stop endpoints
under a configurable router prefix (default /core/fastapi/vllm/qwen3).
Source code in src/unitorch/cli/fastapis/qwen_vllm.py
25
26
27
28
29
30
31
32
33
34
35 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/vllm/qwen3")
self._section = "core/fastapi/vllm/qwen3"
router = config.getoption("router", "/core/fastapi/vllm/qwen3")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/vllm/qwen3'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'qwen3-4b-thinking') -> str
Loads and starts the vLLM QWen3 engine.
Parameters:
| Name |
Type |
Description |
Default |
pretrained_name
|
str
|
Pretrained model name to load. Defaults to "qwen3-4b-thinking".
|
'qwen3-4b-thinking'
|
Source code in src/unitorch/cli/fastapis/qwen_vllm.py
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80 | def start(self, pretrained_name: str = "qwen3-4b-thinking") -> str:
"""
Loads and starts the vLLM QWen3 engine.
Args:
pretrained_name (str): Pretrained model name to load. Defaults to ``"qwen3-4b-thinking"``.
"""
pretrained_name_or_path = nested_dict_value(
pretrained_vllm_infos, pretrained_name, "pretrained_name_or_path"
)
self.config.set_default_section(self._section)
if pretrained_name_or_path is not None:
self.config.set(
self._section, "pretrained_name", pretrained_name
)
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWen3VLLMForGeneration.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Stops and unloads the vLLM engine, releasing GPU memory.
Source code in src/unitorch/cli/fastapis/qwen_vllm.py
| def stop(self) -> str:
"""
Stops and unloads the vLLM engine, releasing GPU memory.
"""
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Returns "running" if the engine is loaded, otherwise "stopped".
Source code in src/unitorch/cli/fastapis/qwen_vllm.py
| def status(self) -> str:
"""Returns ``"running"`` if the engine is loaded, otherwise ``"stopped"``."""
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
use_chat_template: Optional[bool] = True,
max_gen_seq_length: Optional[int] = 512,
min_gen_seq_length: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
num_beams: Optional[int] = 1,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
repetition_penalty: Optional[float] = 1.0,
stop: Optional[Union[str, List[str]]] = None,
) -> Union[str, List[str]]
Generates a text completion for the given prompt.
Parameters:
| Name |
Type |
Description |
Default |
text
|
str
|
Input prompt or JSON-encoded message list (when use_chat_template=True).
|
required
|
use_chat_template
|
bool
|
Apply chat template formatting. Defaults to True.
|
True
|
max_gen_seq_length
|
int
|
Maximum tokens to generate. Defaults to 512.
|
512
|
min_gen_seq_length
|
int
|
Minimum tokens to generate. Defaults to 0.
|
0
|
num_return_sequences
|
int
|
Number of completions to return. Defaults to 1.
|
1
|
num_beams
|
int
|
Beam search width. Defaults to 1.
|
1
|
do_sample
|
bool
|
Enable sampling-based decoding. Defaults to False.
|
False
|
temperature
|
float
|
Sampling temperature. Defaults to 1.0.
|
1.0
|
top_k
|
int
|
Top-k sampling. Defaults to 50.
|
50
|
top_p
|
float
|
Top-p (nucleus) sampling. Defaults to 1.0.
|
1.0
|
repetition_penalty
|
float
|
Repetition penalty. Defaults to 1.0.
|
1.0
|
stop
|
str or List[str]
|
Stop string(s) to end generation.
|
None
|
Returns:
| Type |
Description |
Union[str, List[str]]
|
str or List[str]: Generated text. Single string when num_return_sequences=1.
|
Source code in src/unitorch/cli/fastapis/qwen_vllm.py
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158 | async def generate(
self,
text: str,
use_chat_template: Optional[bool] = True,
max_gen_seq_length: Optional[int] = 512,
min_gen_seq_length: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
num_beams: Optional[int] = 1,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
repetition_penalty: Optional[float] = 1.0,
stop: Optional[Union[str, List[str]]] = None,
) -> Union[str, List[str]]:
"""
Generates a text completion for the given prompt.
Args:
text (str): Input prompt or JSON-encoded message list (when ``use_chat_template=True``).
use_chat_template (bool): Apply chat template formatting. Defaults to True.
max_gen_seq_length (int): Maximum tokens to generate. Defaults to 512.
min_gen_seq_length (int): Minimum tokens to generate. Defaults to 0.
num_return_sequences (int): Number of completions to return. Defaults to 1.
num_beams (int): Beam search width. Defaults to 1.
do_sample (bool): Enable sampling-based decoding. Defaults to False.
temperature (float): Sampling temperature. Defaults to 1.0.
top_k (int): Top-k sampling. Defaults to 50.
top_p (float): Top-p (nucleus) sampling. Defaults to 1.0.
repetition_penalty (float): Repetition penalty. Defaults to 1.0.
stop (str or List[str], optional): Stop string(s) to end generation.
Returns:
str or List[str]: Generated text. Single string when ``num_return_sequences=1``.
"""
assert self._pipes is not None, "Service not started. Call /start first."
pipe = self._pipes.acquire()
try:
processor = pipe.processor
prompt = (
processor.chat_template(messages=json.loads(text))
if use_chat_template
else text
)
inputs = processor.generation_inputs(text=prompt)
input_ids = inputs.input_ids.unsqueeze(0)
outputs = pipe.generate(
input_ids=input_ids,
max_gen_seq_length=max_gen_seq_length,
min_gen_seq_length=min_gen_seq_length,
num_return_sequences=num_return_sequences,
num_beams=num_beams,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
repetition_penalty=repetition_penalty,
stop=stop,
)
finally:
pipe.release()
decoded = processor.detokenize(sequences=outputs.sequences)
sequences = decoded[0]
return sequences[0] if num_return_sequences == 1 else sequences
|
QWen3VLVLLMFastAPI
Tip
core/fastapi/vllm/qwen3_vl is the section for configuration of QWen3VLVLLMFastAPI.
Bases: GenericFastAPI
FastAPI service for QWen3-VL vision-language generation powered by vLLM.
Exposes /generate, /status, /start, and /stop endpoints
under a configurable router prefix (default /core/fastapi/vllm/qwen3_vl).
Accepts both text-only and multimodal (text + image) generation requests.
Source code in src/unitorch/cli/fastapis/qwen_vl_vllm.py
28
29
30
31
32
33
34
35
36
37
38 | def __init__(self, config: Config):
self.config = config
config.set_default_section("core/fastapi/vllm/qwen3_vl")
self._section = "core/fastapi/vllm/qwen3_vl"
router = config.getoption("router", "/core/fastapi/vllm/qwen3_vl")
self._pipes = None
self._router = APIRouter(prefix=router)
self._router.add_api_route("/generate", self.generate, methods=["POST"])
self._router.add_api_route("/status", self.status, methods=["GET"])
self._router.add_api_route("/start", self.start, methods=["GET"])
self._router.add_api_route("/stop", self.stop, methods=["GET"])
|
config
instance-attribute
_section
instance-attribute
_section = 'core/fastapi/vllm/qwen3_vl'
_pipes
instance-attribute
_router
instance-attribute
_router = APIRouter(prefix=router)
start
start(pretrained_name: str = 'qwen3-vl-2b-instruct') -> str
Loads and starts the vLLM QWen3-VL multimodal engine.
Parameters:
| Name |
Type |
Description |
Default |
pretrained_name
|
str
|
Pretrained model name to load. Defaults to "qwen3-vl-2b-instruct".
|
'qwen3-vl-2b-instruct'
|
Source code in src/unitorch/cli/fastapis/qwen_vl_vllm.py
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79 | def start(self, pretrained_name: str = "qwen3-vl-2b-instruct") -> str:
"""
Loads and starts the vLLM QWen3-VL multimodal engine.
Args:
pretrained_name (str): Pretrained model name to load. Defaults to ``"qwen3-vl-2b-instruct"``.
"""
self.config.set_default_section(self._section)
self.config.set(
self._section, "pretrained_name", pretrained_name
)
num_replicas = int(
self.config.getdefault(self._section, "pipeline_num_replicas", 1)
)
devices = self.config.getdefault(
self._section, "pipeline_replica_devices", "cpu"
)
lock = self.config.getdefault(
self._section, "pipeline_replica_lock", True
)
if devices is None:
devices = []
if isinstance(devices, str):
devices = [devices] * num_replicas
pipelines = [
QWen3VLVLLMForGeneration.from_config(
self.config,
pretrained_name=pretrained_name,
)
for _ in range(num_replicas)
]
for pipe, device in zip(pipelines, devices):
if device is not None and hasattr(pipe, "to"):
pipe.to(device)
self._pipes = PipelineReplicaPool(pipelines, lock=lock)
return "start success"
|
stop
Stops and unloads the vLLM engine, releasing GPU memory.
Source code in src/unitorch/cli/fastapis/qwen_vl_vllm.py
| def stop(self) -> str:
"""
Stops and unloads the vLLM engine, releasing GPU memory.
"""
if self._pipes is not None:
self._pipes.close()
self._pipes = None
return "stop success"
|
status
Returns "running" if the engine is loaded, otherwise "stopped".
Source code in src/unitorch/cli/fastapis/qwen_vl_vllm.py
| def status(self) -> str:
"""Returns ``"running"`` if the engine is loaded, otherwise ``"stopped"``."""
return "running" if self._pipes is not None else "stopped"
|
generate
async
generate(
text: str,
image: Optional[UploadFile] = File(default=None),
use_chat_template: Optional[bool] = True,
max_gen_seq_length: Optional[int] = 512,
min_gen_seq_length: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
repetition_penalty: Optional[float] = 1.0,
stop: Optional[Union[str, List[str]]] = None,
) -> Union[str, List[str]]
Generates a text completion for the given prompt and optional image.
Parameters:
| Name |
Type |
Description |
Default |
text
|
str
|
Input prompt or JSON-encoded message list (when use_chat_template=True).
|
required
|
image
|
UploadFile
|
Uploaded image file for multimodal generation.
|
File(default=None)
|
use_chat_template
|
bool
|
Apply chat template formatting. Defaults to True.
|
True
|
max_gen_seq_length
|
int
|
Maximum tokens to generate. Defaults to 512.
|
512
|
min_gen_seq_length
|
int
|
Minimum tokens to generate. Defaults to 0.
|
0
|
num_return_sequences
|
int
|
Number of completions to return. Defaults to 1.
|
1
|
do_sample
|
bool
|
Enable sampling-based decoding. Defaults to False.
|
False
|
temperature
|
float
|
Sampling temperature. Defaults to 1.0.
|
1.0
|
top_k
|
int
|
Top-k sampling. Defaults to 50.
|
50
|
top_p
|
float
|
Top-p (nucleus) sampling. Defaults to 1.0.
|
1.0
|
repetition_penalty
|
float
|
Repetition penalty. Defaults to 1.0.
|
1.0
|
stop
|
str or List[str]
|
|
None
|
Returns:
| Type |
Description |
Union[str, List[str]]
|
str or List[str]: Generated text. Single string when num_return_sequences=1.
|
Source code in src/unitorch/cli/fastapis/qwen_vl_vllm.py
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171 | async def generate(
self,
text: str,
image: Optional[UploadFile] = File(default=None),
use_chat_template: Optional[bool] = True,
max_gen_seq_length: Optional[int] = 512,
min_gen_seq_length: Optional[int] = 0,
num_return_sequences: Optional[int] = 1,
do_sample: Optional[bool] = False,
temperature: Optional[float] = 1.0,
top_k: Optional[int] = 50,
top_p: Optional[float] = 1.0,
repetition_penalty: Optional[float] = 1.0,
stop: Optional[Union[str, List[str]]] = None,
) -> Union[str, List[str]]:
"""
Generates a text completion for the given prompt and optional image.
Args:
text (str): Input prompt or JSON-encoded message list (when ``use_chat_template=True``).
image (UploadFile, optional): Uploaded image file for multimodal generation.
use_chat_template (bool): Apply chat template formatting. Defaults to True.
max_gen_seq_length (int): Maximum tokens to generate. Defaults to 512.
min_gen_seq_length (int): Minimum tokens to generate. Defaults to 0.
num_return_sequences (int): Number of completions to return. Defaults to 1.
do_sample (bool): Enable sampling-based decoding. Defaults to False.
temperature (float): Sampling temperature. Defaults to 1.0.
top_k (int): Top-k sampling. Defaults to 50.
top_p (float): Top-p (nucleus) sampling. Defaults to 1.0.
repetition_penalty (float): Repetition penalty. Defaults to 1.0.
stop (str or List[str], optional): Stop string(s).
Returns:
str or List[str]: Generated text. Single string when ``num_return_sequences=1``.
"""
assert self._pipes is not None, "Service not started. Call /start first."
pil_image = None
if image is not None:
content = await image.read()
pil_image = Image.open(io.BytesIO(content)).convert("RGB")
pipe = self._pipes.acquire()
try:
processor = pipe.processor
prompt = (
processor.chat_template(messages=json.loads(text))
if use_chat_template
else text
)
inputs = processor.generation_inputs(
text=prompt,
images=[pil_image] if pil_image is not None else [],
)
input_ids = inputs.input_ids.unsqueeze(0)
pixel_values = (
inputs.pixel_values.unsqueeze(0) if pil_image is not None else None
)
image_grid_thw = inputs.image_grid_thw if pil_image is not None else None
outputs = pipe.generate(
input_ids=input_ids,
pixel_values=pixel_values,
image_grid_thw=image_grid_thw,
max_gen_seq_length=max_gen_seq_length,
min_gen_seq_length=min_gen_seq_length,
num_return_sequences=num_return_sequences,
do_sample=do_sample,
temperature=temperature,
top_k=top_k,
top_p=top_p,
repetition_penalty=repetition_penalty,
stop=stop,
)
finally:
pipe.release()
decoded = processor.detokenize(sequences=outputs.sequences)
sequences = decoded[0]
return sequences[0] if num_return_sequences == 1 else sequences
|