xinference 0.11.3__py3-none-any.whl → 0.12.1__py3-none-any.whl

xinference/model/audio/model_spec.json

@@ -4,6 +4,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-tiny",
     "model_revision": "167c219b21f11ef214220b8fdb7536b8a88c2475",
+    "ability": "audio-to-text",
     "multilingual": true
   },
   {
@@ -11,6 +12,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-tiny.en",
     "model_revision": "87c7102498dcde7456f24cfd30239ca606ed9063",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -18,6 +20,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-base",
     "model_revision": "8c1db9b51951100007a96a525d83a8ec81b3c237",
+    "ability": "audio-to-text",
     "multilingual": true
   },
   {
@@ -25,6 +28,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-base.en",
     "model_revision": "911407f4214e0e1d82085af863093ec0b66f9cd6",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -32,6 +36,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-small",
     "model_revision": "998cb1a777c20db53d6033a61b977ed4c3792cac",
+    "ability": "audio-to-text",
     "multilingual": true
   },
   {
@@ -39,6 +44,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-small.en",
     "model_revision": "e8727524f962ee844a7319d92be39ac1bd25655a",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -46,6 +52,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-medium",
     "model_revision": "16688beb1294bedd0a6f5cd86fe7eec57bce41ed",
+    "ability": "audio-to-text",
     "multilingual": true
   },
   {
@@ -53,6 +60,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-medium.en",
     "model_revision": "2e98eb6279edf5095af0c8dedb36bdec0acd172b",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -60,6 +68,7 @@
     "model_family": "whisper",
     "model_id": "openai/whisper-large-v3",
     "model_revision": "6cdf07a7e3ec3806e5d55f787915b85d4cd020b1",
+    "ability": "audio-to-text",
     "multilingual": true
   },
   {
@@ -67,6 +76,7 @@
     "model_family": "whisper",
     "model_id": "BELLE-2/Belle-distilwhisper-large-v2-zh",
     "model_revision": "ed25d13498fa5bac758b2fc479435b698532dfe8",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -74,6 +84,7 @@
     "model_family": "whisper",
     "model_id": "BELLE-2/Belle-whisper-large-v2-zh",
     "model_revision": "ec5bd5d78598545b7585814edde86dac2002b5b9",
+    "ability": "audio-to-text",
     "multilingual": false
   },
   {
@@ -81,6 +92,15 @@
     "model_family": "whisper",
     "model_id": "BELLE-2/Belle-whisper-large-v3-zh",
     "model_revision": "3bebc7247696b39f5ab9ed22db426943ac33f600",
+    "ability": "audio-to-text",
     "multilingual": false
+  },
+  {
+    "model_name": "ChatTTS",
+    "model_family": "ChatTTS",
+    "model_id": "2Noise/ChatTTS",
+    "model_revision": "ce5913842aebd78e4a01a02d47244b8d62ac4ee3",
+    "ability": "text-to-audio",
+    "multilingual": true
   }
 ]

xinference/model/audio/model_spec_modelscope.json

@@ -0,0 +1,20 @@
+[
+  {
+    "model_name": "whisper-large-v3",
+    "model_family": "whisper",
+    "model_hub": "modelscope",
+    "model_id": "AI-ModelScope/whisper-large-v3",
+    "model_revision": "master",
+    "ability": "audio-to-text",
+    "multilingual": true
+  },
+  {
+    "model_name": "ChatTTS",
+    "model_family": "ChatTTS",
+    "model_hub": "modelscope",
+    "model_id": "pzc163/chatTTS",
+    "model_revision": "master",
+    "ability": "text-to-audio",
+    "multilingual": true
+  }
+]

xinference/model/llm/__init__.py

@@ -25,6 +25,7 @@ from .core import (
     get_llm_model_descriptions,
 )
 from .llm_family import (
+    BUILTIN_CSGHUB_LLM_FAMILIES,
     BUILTIN_LLM_FAMILIES,
     BUILTIN_LLM_MODEL_CHAT_FAMILIES,
     BUILTIN_LLM_MODEL_GENERATE_FAMILIES,
@@ -117,9 +118,11 @@ def _install():
     from .pytorch.core import PytorchChatModel, PytorchModel
     from .pytorch.deepseek_vl import DeepSeekVLChatModel
     from .pytorch.falcon import FalconPytorchChatModel, FalconPytorchModel
+    from .pytorch.glm4v import Glm4VModel
     from .pytorch.intern_vl import InternVLChatModel
     from .pytorch.internlm2 import Internlm2PytorchChatModel
     from .pytorch.llama_2 import LlamaPytorchChatModel, LlamaPytorchModel
+    from .pytorch.minicpmv25 import MiniCPMV25Model
     from .pytorch.qwen_vl import QwenVLChatModel
     from .pytorch.vicuna import VicunaPytorchChatModel
     from .pytorch.yi_vl import YiVLChatModel
@@ -161,6 +164,8 @@ def _install():
             InternVLChatModel,
             PytorchModel,
             CogVLM2Model,
+            MiniCPMV25Model,
+            Glm4VModel,
         ]
     )
     if OmniLMMModel:  # type: ignore
@@ -217,13 +222,44 @@ def _install():
         if "tools" in model_spec.model_ability:
             BUILTIN_LLM_MODEL_TOOL_CALL_FAMILIES.add(model_spec.model_name)
 
-    for llm_specs in [BUILTIN_LLM_FAMILIES, BUILTIN_MODELSCOPE_LLM_FAMILIES]:
+    csghub_json_path = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)), "llm_family_csghub.json"
+    )
+    for json_obj in json.load(codecs.open(csghub_json_path, "r", encoding="utf-8")):
+        model_spec = LLMFamilyV1.parse_obj(json_obj)
+        BUILTIN_CSGHUB_LLM_FAMILIES.append(model_spec)
+
+        # register prompt style, in case that we have something missed
+        # if duplicated with huggingface json, keep it as the huggingface style
+        if (
+            "chat" in model_spec.model_ability
+            and isinstance(model_spec.prompt_style, PromptStyleV1)
+            and model_spec.model_name not in BUILTIN_LLM_PROMPT_STYLE
+        ):
+            BUILTIN_LLM_PROMPT_STYLE[model_spec.model_name] = model_spec.prompt_style
+        # register model family
+        if "chat" in model_spec.model_ability:
+            BUILTIN_LLM_MODEL_CHAT_FAMILIES.add(model_spec.model_name)
+        else:
+            BUILTIN_LLM_MODEL_GENERATE_FAMILIES.add(model_spec.model_name)
+        if "tools" in model_spec.model_ability:
+            BUILTIN_LLM_MODEL_TOOL_CALL_FAMILIES.add(model_spec.model_name)
+
+    for llm_specs in [
+        BUILTIN_LLM_FAMILIES,
+        BUILTIN_MODELSCOPE_LLM_FAMILIES,
+        BUILTIN_CSGHUB_LLM_FAMILIES,
+    ]:
         for llm_spec in llm_specs:
             if llm_spec.model_name not in LLM_MODEL_DESCRIPTIONS:
                 LLM_MODEL_DESCRIPTIONS.update(generate_llm_description(llm_spec))
 
     # traverse all families and add engine parameters corresponding to the model name
-    for families in [BUILTIN_LLM_FAMILIES, BUILTIN_MODELSCOPE_LLM_FAMILIES]:
+    for families in [
+        BUILTIN_LLM_FAMILIES,
+        BUILTIN_MODELSCOPE_LLM_FAMILIES,
+        BUILTIN_CSGHUB_LLM_FAMILIES,
+    ]:
         for family in families:
             generate_engine_config_by_model_family(family)