xinference 1.5.0.post1__py3-none-any.whl → 1.5.0.post2__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of xinference might be problematic. Click here for more details.
- xinference/_version.py +3 -3
- xinference/core/supervisor.py +11 -2
- xinference/model/llm/transformers/gemma3.py +11 -2
- xinference/web/ui/build/asset-manifest.json +3 -3
- xinference/web/ui/build/index.html +1 -1
- xinference/web/ui/build/static/js/{main.58bd483c.js → main.4b67a723.js} +3 -3
- xinference/web/ui/build/static/js/main.4b67a723.js.map +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/e4ba658c6b3b0490910acdae0c535a892257efb61539a24adf8038fc653bd22f.json +1 -0
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/METADATA +1 -1
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/RECORD +15 -15
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/WHEEL +1 -1
- xinference/web/ui/build/static/js/main.58bd483c.js.map +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/69081049f0c7447544b7cfd73dd13d8846c02fe5febe4d81587e95c89a412d5b.json +0 -1
- /xinference/web/ui/build/static/js/{main.58bd483c.js.LICENSE.txt → main.4b67a723.js.LICENSE.txt} +0 -0
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/entry_points.txt +0 -0
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/licenses/LICENSE +0 -0
- {xinference-1.5.0.post1.dist-info → xinference-1.5.0.post2.dist-info}/top_level.txt +0 -0
xinference/_version.py
CHANGED
|
@@ -8,11 +8,11 @@ import json
|
|
|
8
8
|
|
|
9
9
|
version_json = '''
|
|
10
10
|
{
|
|
11
|
-
"date": "2025-04-
|
|
11
|
+
"date": "2025-04-21T17:53:44+0800",
|
|
12
12
|
"dirty": false,
|
|
13
13
|
"error": null,
|
|
14
|
-
"full-revisionid": "
|
|
15
|
-
"version": "1.5.0.
|
|
14
|
+
"full-revisionid": "a5d4be9f970137bde1d402420f71961826392224",
|
|
15
|
+
"version": "1.5.0.post2"
|
|
16
16
|
}
|
|
17
17
|
''' # END VERSION_JSON
|
|
18
18
|
|
xinference/core/supervisor.py
CHANGED
|
@@ -1356,7 +1356,12 @@ class SupervisorActor(xo.StatelessActor):
|
|
|
1356
1356
|
return model_uid
|
|
1357
1357
|
|
|
1358
1358
|
async def get_launch_builtin_model_progress(self, model_uid: str) -> float:
|
|
1359
|
-
|
|
1359
|
+
try:
|
|
1360
|
+
info = self._model_uid_to_replica_info[model_uid]
|
|
1361
|
+
except KeyError:
|
|
1362
|
+
# Not launched perhaps, just return 0.0 to prevent error
|
|
1363
|
+
return 0.0
|
|
1364
|
+
|
|
1360
1365
|
all_progress = 0.0
|
|
1361
1366
|
i = 0
|
|
1362
1367
|
for rep_model_uid in iter_replica_model_uid(model_uid, info.replica):
|
|
@@ -1370,7 +1375,11 @@ class SupervisorActor(xo.StatelessActor):
|
|
|
1370
1375
|
return all_progress / i if i > 0 else 0.0
|
|
1371
1376
|
|
|
1372
1377
|
async def cancel_launch_builtin_model(self, model_uid: str):
|
|
1373
|
-
|
|
1378
|
+
try:
|
|
1379
|
+
info = self._model_uid_to_replica_info[model_uid]
|
|
1380
|
+
except KeyError:
|
|
1381
|
+
raise RuntimeError(f"Model {model_uid} has not been launched yet")
|
|
1382
|
+
|
|
1374
1383
|
coros = []
|
|
1375
1384
|
for i, rep_model_uid in enumerate(
|
|
1376
1385
|
iter_replica_model_uid(model_uid, info.replica)
|
|
@@ -128,7 +128,12 @@ class Gemma3ChatModel(PytorchChatModel):
|
|
|
128
128
|
).to(self._device)
|
|
129
129
|
input_len = inputs["input_ids"].shape[-1]
|
|
130
130
|
|
|
131
|
-
generation = self._model.generate(
|
|
131
|
+
generation = self._model.generate(
|
|
132
|
+
**inputs,
|
|
133
|
+
do_sample=False,
|
|
134
|
+
max_new_tokens=config.get("max_tokens", 512),
|
|
135
|
+
temperature=config.get("temperature", 1),
|
|
136
|
+
)
|
|
132
137
|
generation = generation[0][input_len:]
|
|
133
138
|
|
|
134
139
|
decoded = self._processor.decode(generation, skip_special_tokens=True)
|
|
@@ -159,7 +164,11 @@ class Gemma3ChatModel(PytorchChatModel):
|
|
|
159
164
|
|
|
160
165
|
def model_generate():
|
|
161
166
|
try:
|
|
162
|
-
return self._model.generate(
|
|
167
|
+
return self._model.generate(
|
|
168
|
+
**gen_kwargs,
|
|
169
|
+
max_new_tokens=config.get("max_tokens", 512),
|
|
170
|
+
temperature=config.get("temperature", 1),
|
|
171
|
+
)
|
|
163
172
|
except Exception:
|
|
164
173
|
nonlocal error
|
|
165
174
|
error = sys.exc_info()
|
|
@@ -1,14 +1,14 @@
|
|
|
1
1
|
{
|
|
2
2
|
"files": {
|
|
3
3
|
"main.css": "./static/css/main.0f6523be.css",
|
|
4
|
-
"main.js": "./static/js/main.
|
|
4
|
+
"main.js": "./static/js/main.4b67a723.js",
|
|
5
5
|
"static/media/icon.webp": "./static/media/icon.4603d52c63041e5dfbfd.webp",
|
|
6
6
|
"index.html": "./index.html",
|
|
7
7
|
"main.0f6523be.css.map": "./static/css/main.0f6523be.css.map",
|
|
8
|
-
"main.
|
|
8
|
+
"main.4b67a723.js.map": "./static/js/main.4b67a723.js.map"
|
|
9
9
|
},
|
|
10
10
|
"entrypoints": [
|
|
11
11
|
"static/css/main.0f6523be.css",
|
|
12
|
-
"static/js/main.
|
|
12
|
+
"static/js/main.4b67a723.js"
|
|
13
13
|
]
|
|
14
14
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="./favicon.svg"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#000000"/><meta name="description" content="Web site created using create-react-app"/><link rel="apple-touch-icon" href="./logo192.png"/><link rel="manifest" href="./manifest.json"/><title>Xinference</title><script defer="defer" src="./static/js/main.
|
|
1
|
+
<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="./favicon.svg"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#000000"/><meta name="description" content="Web site created using create-react-app"/><link rel="apple-touch-icon" href="./logo192.png"/><link rel="manifest" href="./manifest.json"/><title>Xinference</title><script defer="defer" src="./static/js/main.4b67a723.js"></script><link href="./static/css/main.0f6523be.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>
|