xinference 1.2.1__py3-none-any.whl → 1.3.0__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 (80) hide show
  1. xinference/_version.py +3 -3
  2. xinference/api/restful_api.py +4 -7
  3. xinference/client/handlers.py +3 -0
  4. xinference/client/restful/restful_client.py +9 -1
  5. xinference/core/model.py +19 -0
  6. xinference/core/resource.py +7 -1
  7. xinference/core/scheduler.py +4 -7
  8. xinference/core/status_guard.py +1 -0
  9. xinference/core/supervisor.py +228 -19
  10. xinference/core/utils.py +1 -29
  11. xinference/core/worker.py +28 -2
  12. xinference/deploy/cmdline.py +33 -3
  13. xinference/deploy/local.py +2 -1
  14. xinference/deploy/test/test_cmdline.py +32 -0
  15. xinference/device_utils.py +43 -1
  16. xinference/model/audio/core.py +5 -0
  17. xinference/model/audio/kokoro.py +122 -0
  18. xinference/model/audio/model_spec.json +8 -0
  19. xinference/model/audio/model_spec_modelscope.json +9 -0
  20. xinference/model/image/stable_diffusion/core.py +15 -6
  21. xinference/model/llm/llama_cpp/core.py +21 -14
  22. xinference/model/llm/llm_family.json +866 -46
  23. xinference/model/llm/llm_family.py +7 -2
  24. xinference/model/llm/llm_family_modelscope.json +873 -16
  25. xinference/model/llm/mlx/core.py +11 -3
  26. xinference/model/llm/reasoning_parsers/__init__.py +13 -0
  27. xinference/model/llm/reasoning_parsers/abs_reasoning_parsers.py +98 -0
  28. xinference/model/llm/reasoning_parsers/deepseek_r1_reasoning_parser.py +140 -0
  29. xinference/model/llm/sglang/core.py +99 -11
  30. xinference/model/llm/transformers/core.py +9 -1
  31. xinference/model/llm/transformers/intern_vl.py +23 -14
  32. xinference/model/llm/transformers/qwen2_audio.py +3 -1
  33. xinference/model/llm/transformers/qwen2_vl.py +20 -3
  34. xinference/model/llm/transformers/utils.py +22 -11
  35. xinference/model/llm/utils.py +164 -20
  36. xinference/model/llm/vllm/core.py +36 -4
  37. xinference/model/llm/vllm/xavier/executor.py +2 -2
  38. xinference/model/llm/vllm/xavier/scheduler.py +3 -3
  39. xinference/thirdparty/internvl/conversation.py +26 -17
  40. xinference/types.py +2 -0
  41. xinference/web/ui/build/asset-manifest.json +6 -6
  42. xinference/web/ui/build/index.html +1 -1
  43. xinference/web/ui/build/static/css/main.f8177338.css +2 -0
  44. xinference/web/ui/build/static/css/main.f8177338.css.map +1 -0
  45. xinference/web/ui/build/static/js/main.ad42919c.js +3 -0
  46. xinference/web/ui/build/static/js/main.ad42919c.js.map +1 -0
  47. xinference/web/ui/node_modules/.cache/babel-loader/074a42304bbbaa79e1bfc3b28502457a390df55708de9006f4cc8e35c60aea87.json +1 -0
  48. xinference/web/ui/node_modules/.cache/babel-loader/0acb065326560592b10888234242f94f67efe28458b90f273d4d4fba9daa0cd2.json +1 -0
  49. xinference/web/ui/node_modules/.cache/babel-loader/279ace390216236a82b3d8995c78eca4d637ac9a523e9f521a2d9c76607a43d7.json +1 -0
  50. xinference/web/ui/node_modules/.cache/babel-loader/630a7bd592596cc6e291fc32238ce7c08238038a64ed8ccee0eb0c13c9902910.json +1 -0
  51. xinference/web/ui/node_modules/.cache/babel-loader/6cb9f6c62ab4042f0b11c5d75e51187188e9d6f5f08b1d63e796e051bafdb457.json +1 -0
  52. xinference/web/ui/node_modules/.cache/babel-loader/8f9af2979e45d4648f0cfae108363e58ee421c29a9d4e7329b6f06d9adfd4133.json +1 -0
  53. xinference/web/ui/node_modules/.cache/babel-loader/914c33e91c1012e3bcd3e96f3a25884cbef148290632d0266dab972b8cc1e95f.json +1 -0
  54. xinference/web/ui/node_modules/.cache/babel-loader/9c8b1a86e7c65b2b2599a205e30920652d6c2105f926508ef5bcf29a3ef4ce76.json +1 -0
  55. xinference/web/ui/node_modules/.cache/babel-loader/b7939cd3a48adf12fccfdd0803019b5cc235ff7de3a297dae70ce635e0eea13e.json +1 -0
  56. xinference/web/ui/node_modules/.cache/babel-loader/efe7cd132c27a8f9fd5352a394c491fd5fb0da0348cf9fcbd923164a32365eab.json +1 -0
  57. xinference/web/ui/node_modules/.cache/babel-loader/f04f666b77b44d7be3e16034d6b0074de2ba9c254f1fae15222b3148608fa8b3.json +1 -0
  58. xinference/web/ui/node_modules/.cache/babel-loader/fecf076bcd198a458c2a6ab0e85e40dc1c99994c353164e79c469be162cb74c9.json +1 -0
  59. xinference/web/ui/src/locales/en.json +14 -1
  60. xinference/web/ui/src/locales/zh.json +14 -1
  61. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/METADATA +18 -17
  62. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/RECORD +67 -60
  63. xinference/web/ui/build/static/css/main.51a587ff.css +0 -2
  64. xinference/web/ui/build/static/css/main.51a587ff.css.map +0 -1
  65. xinference/web/ui/build/static/js/main.b0936c54.js +0 -3
  66. xinference/web/ui/build/static/js/main.b0936c54.js.map +0 -1
  67. xinference/web/ui/node_modules/.cache/babel-loader/0c2fb5375667931c4a331c99e0d87dc145e8f327cea3f44d6e56f54c7c1d4020.json +0 -1
  68. xinference/web/ui/node_modules/.cache/babel-loader/185ceb8872d562e032b47e79df6a45670e06345b8ed70aad1a131e0476783c5c.json +0 -1
  69. xinference/web/ui/node_modules/.cache/babel-loader/3eefb411b24c2b3ce053570ef50daccf154022f0e168be5ed0fec21394baf9f4.json +0 -1
  70. xinference/web/ui/node_modules/.cache/babel-loader/63c8e07687ea53a4f8a910ee5e42e0eb26cd1acbfbe820f3e3248a786ee51401.json +0 -1
  71. xinference/web/ui/node_modules/.cache/babel-loader/a3ff866acddf34917a7ee399e0e571a4dfd8ba66d5057db885f243e16a6eb17d.json +0 -1
  72. xinference/web/ui/node_modules/.cache/babel-loader/a7f1a71f6580dfe810c685a9c1d68e318f71e1fa258fbe50b87a6ac37cc0a598.json +0 -1
  73. xinference/web/ui/node_modules/.cache/babel-loader/bdee44abeadc4abc17d41c52eb49c6e19a4b1a267b6e16876ce91bdeeebfc52d.json +0 -1
  74. xinference/web/ui/node_modules/.cache/babel-loader/d7664d18c4ddbad9c3a6a31b91f7c00fb0dde804608674a9860ee50f33e54708.json +0 -1
  75. xinference/web/ui/node_modules/.cache/babel-loader/ed57202cb79649bb716400436590245547df241988fc7c8e1d85d132299542d2.json +0 -1
  76. /xinference/web/ui/build/static/js/{main.b0936c54.js.LICENSE.txt → main.ad42919c.js.LICENSE.txt} +0 -0
  77. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/LICENSE +0 -0
  78. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/WHEEL +0 -0
  79. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/entry_points.txt +0 -0
  80. {xinference-1.2.1.dist-info → xinference-1.3.0.dist-info}/top_level.txt +0 -0
xinference/model/llm/llm_family.py CHANGED
@@ -134,7 +134,7 @@ class LLMFamilyV1(BaseModel):
134
134
  model_name: str
135
135
  model_lang: List[str]
136
136
  model_ability: List[
137
- Literal["embed", "generate", "chat", "tools", "vision", "audio"]
137
+ Literal["embed", "generate", "chat", "tools", "vision", "audio", "reasoning"]
138
138
  ]
139
139
  model_description: Optional[str]
140
140
  # reason for not required str here: legacy registration
@@ -143,6 +143,8 @@ class LLMFamilyV1(BaseModel):
143
143
  chat_template: Optional[str]
144
144
  stop_token_ids: Optional[List[int]]
145
145
  stop: Optional[List[str]]
146
+ reasoning_start_tag: Optional[str]
147
+ reasoning_end_tag: Optional[str]
146
148
 
147
149
 
148
150
  class CustomLLMFamilyV1(LLMFamilyV1):
@@ -538,7 +540,10 @@ def _generate_model_file_names(
538
540
  )
539
541
  need_merge = False
540
542
 
541
- if llm_spec.quantization_parts is None:
543
+ if (
544
+ llm_spec.quantization_parts is None
545
+ or quantization not in llm_spec.quantization_parts
546
+ ):
542
547
  file_names.append(final_file_name)
543
548
  elif quantization is not None and quantization in llm_spec.quantization_parts:
544
549
  parts = llm_spec.quantization_parts[quantization]