xinference 1.3.0.post2__py3-none-any.whl → 1.3.1.post1__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.

Files changed (53) hide show
  1. xinference/_version.py +3 -3
  2. xinference/api/restful_api.py +1 -0
  3. xinference/conftest.py +7 -0
  4. xinference/core/chat_interface.py +39 -24
  5. xinference/core/model.py +3 -1
  6. xinference/core/scheduler.py +3 -0
  7. xinference/core/worker.py +1 -1
  8. xinference/model/embedding/core.py +12 -5
  9. xinference/model/llm/__init__.py +2 -1
  10. xinference/model/llm/core.py +10 -0
  11. xinference/model/llm/llama_cpp/core.py +266 -3
  12. xinference/model/llm/llm_family.json +390 -17
  13. xinference/model/llm/llm_family_modelscope.json +348 -29
  14. xinference/model/llm/mlx/core.py +15 -4
  15. xinference/model/llm/{reasoning_parsers/deepseek_r1_reasoning_parser.py → reasoning_parser.py} +9 -13
  16. xinference/model/llm/sglang/core.py +7 -2
  17. xinference/model/llm/transformers/chatglm.py +4 -4
  18. xinference/model/llm/transformers/core.py +22 -5
  19. xinference/model/llm/transformers/intern_vl.py +2 -1
  20. xinference/model/llm/transformers/utils.py +1 -1
  21. xinference/model/llm/utils.py +134 -60
  22. xinference/model/llm/vllm/core.py +31 -42
  23. xinference/types.py +4 -0
  24. xinference/web/ui/build/asset-manifest.json +3 -3
  25. xinference/web/ui/build/index.html +1 -1
  26. xinference/web/ui/build/static/js/main.55b70cb7.js +3 -0
  27. xinference/web/ui/build/static/js/main.55b70cb7.js.map +1 -0
  28. xinference/web/ui/node_modules/.cache/babel-loader/0f0adb2283a8f469d097a7a0ebb754624fa52414c83b83696c41f2e6a737ceda.json +1 -0
  29. xinference/web/ui/node_modules/.cache/babel-loader/2deac8d5636974533e3714f34e94fc754f9153a07c6ee11e72846cb8eae47e4b.json +1 -0
  30. xinference/web/ui/node_modules/.cache/babel-loader/8157db83995c671eb57abc316c337f867d1dc63fb83520bb4ff351fee57dcce2.json +1 -0
  31. xinference/web/ui/node_modules/.cache/babel-loader/87a9b13f2466f375ae5c6e7c08b279cc38351d29710d7f7626bbb07a85262b79.json +1 -0
  32. xinference/web/ui/node_modules/.cache/babel-loader/e23d476fcbf6fd69c8986bf82133d257d28aa8fc9a5cab231d81c1c75c58cd99.json +1 -0
  33. xinference/web/ui/node_modules/.cache/babel-loader/e547bbb18abb4a474b675a8d5782d25617566bea0af8caa9b836ce5649e2250a.json +1 -0
  34. xinference/web/ui/node_modules/.cache/babel-loader/e7a8c37fda8725cab69c7ef8c627060bd7fc806adc67e00fe628ba148cb86d7f.json +1 -0
  35. xinference/web/ui/src/locales/en.json +9 -1
  36. xinference/web/ui/src/locales/zh.json +9 -1
  37. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/METADATA +9 -5
  38. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/RECORD +43 -44
  39. xinference/model/llm/reasoning_parsers/__init__.py +0 -13
  40. xinference/model/llm/reasoning_parsers/abs_reasoning_parsers.py +0 -98
  41. xinference/web/ui/build/static/js/main.ad42919c.js +0 -3
  42. xinference/web/ui/build/static/js/main.ad42919c.js.map +0 -1
  43. xinference/web/ui/node_modules/.cache/babel-loader/074a42304bbbaa79e1bfc3b28502457a390df55708de9006f4cc8e35c60aea87.json +0 -1
  44. xinference/web/ui/node_modules/.cache/babel-loader/279ace390216236a82b3d8995c78eca4d637ac9a523e9f521a2d9c76607a43d7.json +0 -1
  45. xinference/web/ui/node_modules/.cache/babel-loader/630a7bd592596cc6e291fc32238ce7c08238038a64ed8ccee0eb0c13c9902910.json +0 -1
  46. xinference/web/ui/node_modules/.cache/babel-loader/914c33e91c1012e3bcd3e96f3a25884cbef148290632d0266dab972b8cc1e95f.json +0 -1
  47. xinference/web/ui/node_modules/.cache/babel-loader/b7939cd3a48adf12fccfdd0803019b5cc235ff7de3a297dae70ce635e0eea13e.json +0 -1
  48. xinference/web/ui/node_modules/.cache/babel-loader/fecf076bcd198a458c2a6ab0e85e40dc1c99994c353164e79c469be162cb74c9.json +0 -1
  49. /xinference/web/ui/build/static/js/{main.ad42919c.js.LICENSE.txt → main.55b70cb7.js.LICENSE.txt} +0 -0
  50. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/LICENSE +0 -0
  51. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/WHEEL +0 -0
  52. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/entry_points.txt +0 -0
  53. {xinference-1.3.0.post2.dist-info → xinference-1.3.1.post1.dist-info}/top_level.txt +0 -0
xinference/types.py CHANGED
@@ -78,6 +78,7 @@ class EmbeddingData(TypedDict):
78
78
  class Embedding(TypedDict):
79
79
  object: Literal["list"]
80
80
  model: str
81
+ model_replica: str
81
82
  data: List[EmbeddingData]
82
83
  usage: EmbeddingUsage
83
84
 
@@ -276,6 +277,7 @@ class LlamaCppModelConfig(TypedDict, total=False):
276
277
  use_mmap: bool
277
278
  use_mlock: bool
278
279
  n_threads: Optional[int]
280
+ n_parallel: Optional[int]
279
281
  n_batch: int
280
282
  last_n_tokens_size: int
281
283
  lora_base: Optional[str]
@@ -284,6 +286,7 @@ class LlamaCppModelConfig(TypedDict, total=False):
284
286
  n_gqa: Optional[int] # (TEMPORARY) must be 8 for llama2 70b
285
287
  rms_norm_eps: Optional[float] # (TEMPORARY)
286
288
  verbose: bool
289
+ reasoning_content: bool
287
290
 
288
291
 
289
292
  class PytorchGenerateConfig(TypedDict, total=False):
@@ -330,6 +333,7 @@ class PytorchModelConfig(TypedDict, total=False):
330
333
  trust_remote_code: bool
331
334
  max_num_seqs: int
332
335
  enable_tensorizer: Optional[bool]
336
+ reasoning_content: bool
333
337
 
334
338
 
335
339
  def get_pydantic_model_from_method(
xinference/web/ui/build/asset-manifest.json CHANGED
@@ -1,14 +1,14 @@
1
1
  {
2
2
  "files": {
3
3
  "main.css": "./static/css/main.f8177338.css",
4
- "main.js": "./static/js/main.ad42919c.js",
4
+ "main.js": "./static/js/main.55b70cb7.js",
5
5
  "static/media/icon.webp": "./static/media/icon.4603d52c63041e5dfbfd.webp",
6
6
  "index.html": "./index.html",
7
7
  "main.f8177338.css.map": "./static/css/main.f8177338.css.map",
8
- "main.ad42919c.js.map": "./static/js/main.ad42919c.js.map"
8
+ "main.55b70cb7.js.map": "./static/js/main.55b70cb7.js.map"
9
9
  },
10
10
  "entrypoints": [
11
11
  "static/css/main.f8177338.css",
12
- "static/js/main.ad42919c.js"
12
+ "static/js/main.55b70cb7.js"
13
13
  ]
14
14
  }
xinference/web/ui/build/index.html CHANGED
@@ -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.ad42919c.js"></script><link href="./static/css/main.f8177338.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>
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.55b70cb7.js"></script><link href="./static/css/main.f8177338.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>