nexaai 1.0.4rc15__cp310-cp310-win_amd64.whl → 1.0.5__cp310-cp310-win_amd64.whl

nexaai/__init__.py

@@ -19,7 +19,10 @@ except ImportError:
     __version__ = "0.0.1"
 
 # Import common configuration classes first (no external dependencies)
-from .common import ModelConfig, GenerationConfig, ChatMessage, SamplerConfig
+from .common import ModelConfig, GenerationConfig, ChatMessage, SamplerConfig, PluginID
+
+# Create alias for PluginID to be accessible as plugin_id
+plugin_id = PluginID
 
 # Import new feature classes (no external dependencies in base classes)
 from .llm import LLM
@@ -40,6 +43,8 @@ __all__ = [
     "ChatMessage",
     "SamplerConfig",
     "EmbeddingConfig",
+    "PluginID",
+    "plugin_id",
 
     "LLM",
     "Embedder",

nexaai/_version.py

@@ -1,4 +1,4 @@
 # This file is generated by CMake from _version.py.in
 # Do not modify this file manually - it will be overwritten
 
-__version__ = "1.0.4-rc15" 
+__version__ = "1.0.5" 

nexaai/asr.py

@@ -1,8 +1,9 @@
-from typing import List, Optional, Sequence, Tuple
+from typing import List, Optional, Sequence, Tuple, Union
 from abc import abstractmethod
 from dataclasses import dataclass
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -33,11 +34,14 @@ class ASR(BaseModel):
                    model_path: str,
                    tokenizer_path: Optional[str] = None,
                    language: Optional[str] = None,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'ASR':
         """Load ASR model from local path, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.asr_impl.mlx_asr_impl import MLXASRImpl
             return MLXASRImpl._load_from(model_path, tokenizer_path, language, plugin_id, device_id)
         else:

nexaai/asr_impl/mlx_asr_impl.py

@@ -1,7 +1,8 @@
 # Note: This code is generated by Cursor, not tested yet.
 
-from typing import List, Optional
+from typing import List, Optional, Union
 
+from nexaai.common import PluginID
 from nexaai.asr import ASR, ASRConfig, ASRResult
 from nexaai.mlx_backend.asr.interface import MlxAsr as MLXASRInterface
 from nexaai.mlx_backend.ml import ModelConfig as MLXModelConfig, SamplerConfig as MLXSamplerConfig, GenerationConfig as MLXGenerationConfig, EmbeddingConfig
@@ -18,7 +19,7 @@ class MLXASRImpl(ASR):
                    model_path: str,
                    tokenizer_path: Optional[str] = None,
                    language: Optional[str] = None,
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MLXASRImpl':
         """Load ASR model from local path using MLX backend."""

nexaai/asr_impl/pybind_asr_impl.py

@@ -1,5 +1,6 @@
-from typing import List, Optional
+from typing import List, Optional, Union
 
+from nexaai.common import PluginID
 from nexaai.asr import ASR, ASRConfig, ASRResult
 
 
@@ -14,7 +15,7 @@ class PyBindASRImpl(ASR):
                    model_path: str,
                    tokenizer_path: Optional[str] = None,
                    language: Optional[str] = None,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindASRImpl':
         """Load ASR model from local path using PyBind backend."""

nexaai/common.py

@@ -1,5 +1,12 @@
 from dataclasses import dataclass
 from typing import TypedDict, Literal, Optional, List
+from enum import Enum
+
+
+class PluginID(str, Enum):
+    """Enum for plugin identifiers."""
+    MLX = "mlx"
+    LLAMA_CPP = "llama_cpp"
 
 
 class ChatMessage(TypedDict):
@@ -52,10 +59,45 @@ class ModelConfig:
 
 @dataclass(frozen=True) # Read-only
 class ProfilingData:
-    start_time: int
-    end_time: int
-    prompt_start_time: int = None
-    prompt_end_time: int = None
-    decode_start_time: int = None
-    decode_ent_time: int = None
-    first_token_time: int = None
+    """Profiling data structure for LLM/VLM performance metrics."""
+    ttft: int = 0             # Time to first token (us)
+    prompt_time: int = 0      # Prompt processing time (us)
+    decode_time: int = 0      # Token generation time (us)
+    prompt_tokens: int = 0    # Number of prompt tokens
+    generated_tokens: int = 0  # Number of generated tokens
+    audio_duration: int = 0   # Audio duration (us)
+    prefill_speed: float = 0.0  # Prefill speed (tokens/sec)
+    decoding_speed: float = 0.0 # Decoding speed (tokens/sec)
+    real_time_factor: float = 0.0 # Real-Time Factor (RTF)
+    stop_reason: str = ""     # Stop reason: "eos", "length", "user", "stop_sequence"
+    
+    @classmethod
+    def from_dict(cls, data: dict) -> "ProfilingData":
+        """Create ProfilingData from dictionary."""
+        return cls(
+            ttft=data.get("ttft", 0),
+            prompt_time=data.get("prompt_time", 0),
+            decode_time=data.get("decode_time", 0),
+            prompt_tokens=data.get("prompt_tokens", 0),
+            generated_tokens=data.get("generated_tokens", 0),
+            audio_duration=data.get("audio_duration", 0),
+            prefill_speed=data.get("prefill_speed", 0.0),
+            decoding_speed=data.get("decoding_speed", 0.0),
+            real_time_factor=data.get("real_time_factor", 0.0),
+            stop_reason=data.get("stop_reason", "")
+        )
+    
+    def to_dict(self) -> dict:
+        """Convert to dictionary."""
+        return {
+            "ttft": self.ttft,
+            "prompt_time": self.prompt_time,
+            "decode_time": self.decode_time,
+            "prompt_tokens": self.prompt_tokens,
+            "generated_tokens": self.generated_tokens,
+            "audio_duration": self.audio_duration,
+            "prefill_speed": self.prefill_speed,
+            "decoding_speed": self.decoding_speed,
+            "real_time_factor": self.real_time_factor,
+            "stop_reason": self.stop_reason
+        }

nexaai/cv.py

@@ -1,8 +1,9 @@
-from typing import List, Optional
+from typing import List, Optional, Union
 from abc import abstractmethod
 from dataclasses import dataclass
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -71,11 +72,14 @@ class CVModel(BaseModel):
     def _load_from(cls,
                    _: str,  # TODO: remove this argument, this is a hack to make api design happy
                    config: CVModelConfig,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'CVModel':
         """Load CV model from configuration, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.cv_impl.mlx_cv_impl import MLXCVImpl
             return MLXCVImpl._load_from(config, plugin_id, device_id)
         else:

nexaai/cv_impl/mlx_cv_impl.py

@@ -1,8 +1,9 @@
 # Note: This code is generated by Cursor, not tested yet.
 
-from typing import Optional
+from typing import Optional, Union
 import os
 
+from nexaai.common import PluginID
 from nexaai.cv import CVModel, CVModelConfig, CVResults
 from nexaai.mlx_backend.cv.interface import CVModel as MLXCVInterface, create_cv_model
 
@@ -16,7 +17,7 @@ class MLXCVImpl(CVModel):
     @classmethod
     def _load_from(cls,
                    config: CVModelConfig,
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MLXCVImpl':
         """Load CV model from configuration using MLX backend."""

nexaai/cv_impl/pybind_cv_impl.py

@@ -1,5 +1,6 @@
-from typing import Optional
+from typing import Optional, Union
 
+from nexaai.common import PluginID
 from nexaai.cv import CVModel, CVModelConfig, CVResults
 
 
@@ -12,7 +13,7 @@ class PyBindCVImpl(CVModel):
     @classmethod
     def _load_from(cls,
                    config: CVModelConfig,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindCVImpl':
         """Load CV model from configuration using PyBind backend."""

nexaai/embedder.py

@@ -4,6 +4,7 @@ from abc import abstractmethod
 import numpy as np
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -21,19 +22,22 @@ class Embedder(BaseModel):
         pass
 
     @classmethod
-    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: str = "llama_cpp"):
+    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP):
         """
         Load an embedder from model files, routing to appropriate implementation.
 
         Args:
             model_path: Path to the model file
             tokenizer_file: Path to the tokenizer file (default: "tokenizer.json")
-            plugin_id: Plugin ID to use for the model (default: "llama_cpp")
+            plugin_id: Plugin ID to use for the model (default: PluginID.LLAMA_CPP)
 
         Returns:
             Embedder instance
         """
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.embedder_impl.mlx_embedder_impl import MLXEmbedderImpl
             return MLXEmbedderImpl._load_from(model_path, tokenizer_file, plugin_id)
         else:

nexaai/embedder_impl/mlx_embedder_impl.py

@@ -1,6 +1,7 @@
 from typing import List, Union
 import numpy as np
 
+from nexaai.common import PluginID
 from nexaai.embedder import Embedder, EmbeddingConfig
 from nexaai.mlx_backend.embedding.interface import Embedder as MLXEmbedderInterface
 from nexaai.mlx_backend.ml import ModelConfig as MLXModelConfig, SamplerConfig as MLXSamplerConfig, GenerationConfig as MLXGenerationConfig, EmbeddingConfig
@@ -13,14 +14,14 @@ class MLXEmbedderImpl(Embedder):
         self._mlx_embedder = None
 
     @classmethod
-    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: str = "mlx"):
+    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: Union[PluginID, str] = PluginID.MLX):
         """
         Load an embedder from model files using MLX backend.
 
         Args:
             model_path: Path to the model file
             tokenizer_file: Path to the tokenizer file (default: "tokenizer.json")
-            plugin_id: Plugin ID to use for the model (default: "mlx")
+            plugin_id: Plugin ID to use for the model (default: PluginID.MLX)
 
         Returns:
             MLXEmbedderImpl instance

nexaai/embedder_impl/pybind_embedder_impl.py

@@ -1,6 +1,7 @@
 from typing import List, Union
 import numpy as np
 
+from nexaai.common import PluginID
 from nexaai.embedder import Embedder, EmbeddingConfig
 from nexaai.binds import embedder_bind
 from nexaai.runtime import _ensure_runtime
@@ -15,20 +16,22 @@ class PyBindEmbedderImpl(Embedder):
         self._handle = _handle_ptr
 
     @classmethod
-    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: str = "llama_cpp"):
+    def _load_from(cls, model_path: str, tokenizer_file: str = "tokenizer.json", plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP):
         """
         Load an embedder from model files
 
         Args:
             model_path: Path to the model file
             tokenizer_file: Path to the tokenizer file (default: "tokenizer.json")
-            plugin_id: Plugin ID to use for the model (default: "llama_cpp")
+            plugin_id: Plugin ID to use for the model (default: PluginID.LLAMA_CPP)
 
         Returns:
             PyBindEmbedderImpl instance
         """
         _ensure_runtime()
-        handle = embedder_bind.ml_embedder_create(model_path, tokenizer_file, plugin_id)
+        # Convert enum to string for C++ binding
+        plugin_id_str = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        handle = embedder_bind.ml_embedder_create(model_path, tokenizer_file, plugin_id_str)
         return cls(handle)
 
     def eject(self):

nexaai/image_gen.py

@@ -3,6 +3,7 @@ from abc import abstractmethod
 from dataclasses import dataclass
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -67,13 +68,16 @@ class ImageGen(BaseModel):
     def _load_from(cls,
                    model_path: str,
                    scheduler_config_path: str = "",
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None,
                    float16: bool = True,
                    quantize: bool = False
         ) -> 'ImageGen':
         """Load image generation model from local path, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.image_gen_impl.mlx_image_gen_impl import MLXImageGenImpl
             return MLXImageGenImpl._load_from(model_path, scheduler_config_path, plugin_id, device_id, float16, quantize)
         else:

nexaai/image_gen_impl/mlx_image_gen_impl.py

@@ -1,8 +1,9 @@
 # Note: This code is generated by Cursor, not tested yet.
 
-from typing import List, Optional
+from typing import List, Optional, Union
 import os
 
+from nexaai.common import PluginID
 from nexaai.image_gen import ImageGen, ImageGenerationConfig, ImageSamplerConfig, SchedulerConfig, Image
 from nexaai.mlx_backend.sd.interface import ImageGen as MLXImageGenInterface
 
@@ -17,7 +18,7 @@ class MLXImageGenImpl(ImageGen):
     def _load_from(cls,
                    model_path: str,
                    scheduler_config_path: str = "",
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None,
                    float16: bool = True,
                    quantize: bool = False

nexaai/image_gen_impl/pybind_image_gen_impl.py

@@ -1,5 +1,6 @@
-from typing import List, Optional
+from typing import List, Optional, Union
 
+from nexaai.common import PluginID
 from nexaai.image_gen import ImageGen, ImageGenerationConfig, ImageSamplerConfig, SchedulerConfig, Image
 
 
@@ -13,7 +14,7 @@ class PyBindImageGenImpl(ImageGen):
     def _load_from(cls,
                    model_path: str,
                    scheduler_config_path: str = "",
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None,
                    float16: bool = True,
                    quantize: bool = False

nexaai/llm.py

@@ -1,10 +1,10 @@
-from typing import Generator, Optional
+from typing import Generator, Optional, Union
 from abc import abstractmethod
 import queue
 import threading
 
-from nexaai.common import ModelConfig, GenerationConfig, ChatMessage
-from nexaai.base import BaseModel
+from nexaai.common import ModelConfig, GenerationConfig, ChatMessage, PluginID
+from nexaai.base import BaseModel, ProfilingData
 
 class LLM(BaseModel):
     def __init__(self, m_cfg: ModelConfig = ModelConfig()):
@@ -17,11 +17,14 @@ class LLM(BaseModel):
                    local_path: str,
                    tokenizer_path: Optional[str] = None,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'LLM':
         """Load model from local path, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.llm_impl.mlx_llm_impl import MLXLLMImpl
             return MLXLLMImpl._load_from(local_path, tokenizer_path, m_cfg, plugin_id, device_id)
         else:
@@ -37,7 +40,7 @@ class LLM(BaseModel):
         self._cancel_event.clear()
 
     @abstractmethod
-    def apply_chat_template(self, messages: list[ChatMessage]) -> str:
+    def apply_chat_template(self, messages: list[ChatMessage], tools: Optional[str] = None, enable_thinking: bool = True, add_generation_prompt: bool = True) -> str:
         """Apply the chat template to messages."""
         pass
 
@@ -60,6 +63,10 @@ class LLM(BaseModel):
         """
         pass
 
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data from the last generation."""  
+        pass
+
     @abstractmethod
     def save_kv_cache(self, path: str):
         """

nexaai/llm_impl/mlx_llm_impl.py

@@ -1,6 +1,7 @@
-from typing import Generator, Optional, Any
+from typing import Generator, Optional, Any, Sequence, Union
 
-from nexaai.common import ModelConfig, GenerationConfig, ChatMessage
+from nexaai.base import ProfilingData
+from nexaai.common import ModelConfig, GenerationConfig, ChatMessage, PluginID
 from nexaai.llm import LLM
 from nexaai.mlx_backend.llm.interface import LLM as MLXLLMInterface
 from nexaai.mlx_backend.ml import ModelConfig as MLXModelConfig, SamplerConfig as MLXSamplerConfig, GenerationConfig as MLXGenerationConfig, EmbeddingConfig
@@ -17,7 +18,7 @@ class MLXLLMImpl(LLM):
                    local_path: str,
                    tokenizer_path: Optional[str] = None,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MLXLLMImpl':
         """Load model from local path using MLX backend."""
@@ -54,7 +55,13 @@ class MLXLLMImpl(LLM):
             self._mlx_llm.destroy()
             self._mlx_llm = None
 
-    def apply_chat_template(self, messages: list[ChatMessage]) -> str:
+    def apply_chat_template(
+        self,
+        messages: Sequence[ChatMessage], 
+        tools: Optional[str] = None,
+        enable_thinking: bool = True, 
+        add_generation_prompt: bool = True
+    ) -> str:
         """Apply the chat template to messages."""
         if not self._mlx_llm:
             raise RuntimeError("MLX LLM not loaded")
@@ -68,9 +75,16 @@ class MLXLLMImpl(LLM):
                     def __init__(self, role, content):
                         self.role = role
                         self.content = content
-                mlx_messages.append(MLXChatMessage(msg["role"], msg["content"]))
+                
+                # Handle both dict-style and attribute-style access
+                if hasattr(msg, 'role') and hasattr(msg, 'content'):
+                    # Message is already an object with attributes
+                    mlx_messages.append(MLXChatMessage(msg.role, msg.content))
+                else:
+                    # Message is a dict
+                    mlx_messages.append(MLXChatMessage(msg["role"], msg["content"]))
             
-            return self._mlx_llm.apply_chat_template(mlx_messages)
+            return self._mlx_llm.apply_chat_template(mlx_messages, tools=tools, enable_thinking=enable_thinking, add_generation_prompt=add_generation_prompt)
         except Exception as e:
             raise RuntimeError(f"Failed to apply chat template: {str(e)}")
 
@@ -202,6 +216,12 @@ class MLXLLMImpl(LLM):
         except Exception as e:
             raise RuntimeError(f"Failed to generate text: {str(e)}")
 
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data from the last generation."""
+        if not self._mlx_llm:
+            raise RuntimeError("MLX LLM not loaded")
+        return self._mlx_llm.get_profiling_data()
+
     def save_kv_cache(self, path: str):
         """
         Save the key-value cache to the file.

nexaai/llm_impl/pybind_llm_impl.py

@@ -1,8 +1,9 @@
-from typing import Generator, Optional
+from typing import Generator, Optional, Union
 import queue
 import threading
 
-from nexaai.common import ModelConfig, GenerationConfig, ChatMessage
+from nexaai.base import ProfilingData
+from nexaai.common import ModelConfig, GenerationConfig, ChatMessage, PluginID
 from nexaai.binds import llm_bind, common_bind
 from nexaai.runtime import _ensure_runtime
 from nexaai.llm import LLM
@@ -13,13 +14,14 @@ class PyBindLLMImpl(LLM):
         """Private constructor, should not be called directly."""
         super().__init__(m_cfg)
         self._handle = handle  # This is a py::capsule
+        self._profiling_data = None
 
     @classmethod
     def _load_from(cls,
                    local_path: str,
                    tokenizer_path: Optional[str] = None,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindLLMImpl':
         """Load model from local path."""
@@ -49,11 +51,13 @@ class PyBindLLMImpl(LLM):
             config.chat_template_content = m_cfg.chat_template_content
 
         # Create handle : returns py::capsule with automatic cleanup
+        # Convert enum to string for C++ binding
+        plugin_id_str = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
         handle = llm_bind.ml_llm_create(
             model_path=local_path,
             tokenizer_path=tokenizer_path,
             model_config=config,
-            plugin_id=plugin_id,
+            plugin_id=plugin_id_str,
             device_id=device_id
         )
         return cls(handle, m_cfg)
@@ -64,7 +68,7 @@ class PyBindLLMImpl(LLM):
         del self._handle
         self._handle = None
 
-    def apply_chat_template(self, messages: list[ChatMessage]) -> str:
+    def apply_chat_template(self, messages: list[ChatMessage], tools: Optional[str] = None, enable_thinking: bool = True, add_generation_prompt: bool = True) -> str:
         """Apply the chat template to messages."""
         # Convert TypedDict to list of dicts for binding
         message_dicts = [
@@ -95,13 +99,14 @@ class PyBindLLMImpl(LLM):
         # Run generation in thread
         def generate():
             try:
-                llm_bind.ml_llm_generate(
+                result = llm_bind.ml_llm_generate(
                     handle=self._handle,
                     prompt=prompt,
                     config=config,
                     on_token=on_token,
                     user_data=None
                 )
+                self._profiling_data = ProfilingData.from_dict(result.get("profile_data", {}))
             except Exception as e:
                 exception_container[0] = e
             finally:
@@ -143,8 +148,14 @@ class PyBindLLMImpl(LLM):
             on_token=None,  # No callback for non-streaming
             user_data=None
         )
+
+        self._profiling_data = ProfilingData.from_dict(result.get("profile_data", {}))
         return result.get("text", "")
 
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data."""
+        return self._profiling_data
+
     def save_kv_cache(self, path: str):
         """
         Save the key-value cache to the file.

nexaai/rerank.py

@@ -1,8 +1,9 @@
-from typing import List, Optional, Sequence
+from typing import List, Optional, Sequence, Union
 from abc import abstractmethod
 from dataclasses import dataclass
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -24,11 +25,14 @@ class Reranker(BaseModel):
     def _load_from(cls,
                    model_path: str,
                    tokenizer_file: str = "tokenizer.json",
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'Reranker':
         """Load reranker model from local path, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.rerank_impl.mlx_rerank_impl import MLXRerankImpl
             return MLXRerankImpl._load_from(model_path, tokenizer_file, plugin_id, device_id)
         else:

nexaai/rerank_impl/mlx_rerank_impl.py

@@ -1,8 +1,9 @@
 # Note: This code is generated by Cursor, not tested yet.
 
-from typing import List, Optional, Sequence
+from typing import List, Optional, Sequence, Union
 import os
 
+from nexaai.common import PluginID
 from nexaai.rerank import Reranker, RerankConfig
 from nexaai.mlx_backend.rerank.interface import Reranker as MLXRerankInterface, create_reranker
 
@@ -17,7 +18,7 @@ class MLXRerankImpl(Reranker):
     def _load_from(cls,
                    model_path: str,
                    tokenizer_file: str = "tokenizer.json",
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MLXRerankImpl':
         """Load reranker model from local path using MLX backend."""

nexaai/rerank_impl/pybind_rerank_impl.py

@@ -1,5 +1,6 @@
-from typing import List, Optional, Sequence
+from typing import List, Optional, Sequence, Union
 
+from nexaai.common import PluginID
 from nexaai.rerank import Reranker, RerankConfig
 
 
@@ -13,7 +14,7 @@ class PyBindRerankImpl(Reranker):
     def _load_from(cls,
                    model_path: str,
                    tokenizer_file: str = "tokenizer.json",
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindRerankImpl':
         """Load reranker model from local path using PyBind backend."""

nexaai/tts.py

@@ -1,8 +1,9 @@
-from typing import List, Optional
+from typing import List, Optional, Union
 from abc import abstractmethod
 from dataclasses import dataclass
 
 from nexaai.base import BaseModel
+from nexaai.common import PluginID
 
 
 @dataclass
@@ -43,11 +44,14 @@ class TTS(BaseModel):
     def _load_from(cls,
                    model_path: str,
                    vocoder_path: str,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'TTS':
         """Load TTS model from local path, routing to appropriate implementation."""
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.tts_impl.mlx_tts_impl import MLXTTSImpl
             return MLXTTSImpl._load_from(model_path, vocoder_path, plugin_id, device_id)
         else:

nexaai/tts_impl/mlx_tts_impl.py

@@ -1,8 +1,9 @@
 # Note: This code is generated by Cursor, not tested yet.
 
-from typing import List, Optional
+from typing import List, Optional, Union
 import os
 
+from nexaai.common import PluginID
 from nexaai.tts import TTS, TTSConfig, TTSResult
 from nexaai.mlx_backend.tts.interface import MlxTts as MLXTTSInterface
 
@@ -17,7 +18,7 @@ class MLXTTSImpl(TTS):
     def _load_from(cls,
                    model_path: str,
                    vocoder_path: str,
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MLXTTSImpl':
         """Load TTS model from local path using MLX backend."""

nexaai/tts_impl/pybind_tts_impl.py

@@ -1,5 +1,6 @@
-from typing import List, Optional
+from typing import List, Optional, Union
 
+from nexaai.common import PluginID
 from nexaai.tts import TTS, TTSConfig, TTSResult
 
 
@@ -13,7 +14,7 @@ class PyBindTTSImpl(TTS):
     def _load_from(cls,
                    model_path: str,
                    vocoder_path: str,
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindTTSImpl':
         """Load TTS model from local path using PyBind backend."""

nexaai/vlm.py

@@ -5,8 +5,8 @@ import threading
 import base64
 from pathlib import Path
 
-from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage
-from nexaai.base import BaseModel
+from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage, PluginID
+from nexaai.base import BaseModel, ProfilingData
 
 
 class VLM(BaseModel):
@@ -20,7 +20,7 @@ class VLM(BaseModel):
                    local_path: str,
                    mmproj_path: str,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'VLM':
         """Load VLM model from local path, routing to appropriate implementation.
@@ -35,7 +35,10 @@ class VLM(BaseModel):
         Returns:
             VLM instance
         """
-        if plugin_id == "mlx":
+        # Check plugin_id value for routing - handle both enum and string
+        plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
+        
+        if plugin_value == "mlx":
             from nexaai.vlm_impl.mlx_vlm_impl import MlxVlmImpl
             return MlxVlmImpl._load_from(local_path, mmproj_path, m_cfg, plugin_id, device_id)
         else:
@@ -117,4 +120,8 @@ class VLM(BaseModel):
         Returns:
             str: The generated text.
         """
+        pass
+
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data from the last generation."""
         pass

nexaai/vlm_impl/mlx_vlm_impl.py

@@ -1,6 +1,7 @@
-from typing import Generator, Optional, List, Dict, Any
+from typing import Generator, Optional, List, Dict, Any, Union
 
-from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage
+from nexaai.base import ProfilingData
+from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage, PluginID
 from nexaai.vlm import VLM
 from nexaai.mlx_backend.vlm.interface import VLM as MLXVLMInterface
 from nexaai.mlx_backend.ml import ModelConfig as MLXModelConfig, SamplerConfig as MLXSamplerConfig, GenerationConfig as MLXGenerationConfig, EmbeddingConfig
@@ -17,7 +18,7 @@ class MlxVlmImpl(VLM):
                    local_path: str,
                    mmproj_path: str,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "mlx",
+                   plugin_id: Union[PluginID, str] = PluginID.MLX,
                    device_id: Optional[str] = None
         ) -> 'MlxVlmImpl':
         """Load VLM model from local path using MLX backend.
@@ -247,3 +248,9 @@ class MlxVlmImpl(VLM):
             
         except Exception as e:
             raise RuntimeError(f"Failed to generate text: {str(e)}")
+
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data from the last generation."""
+        if not self._mlx_vlm:
+            raise RuntimeError("MLX VLM not loaded")
+        return self._mlx_vlm.get_profiling_data()

nexaai/vlm_impl/pybind_vlm_impl.py

@@ -4,10 +4,11 @@ import threading
 import base64
 from pathlib import Path
 
-from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage
+from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage, PluginID
 from nexaai.binds import vlm_bind, common_bind
 from nexaai.runtime import _ensure_runtime
 from nexaai.vlm import VLM
+from nexaai.base import ProfilingData
 
 
 class PyBindVLMImpl(VLM):
@@ -15,13 +16,14 @@ class PyBindVLMImpl(VLM):
         """Private constructor, should not be called directly."""
         super().__init__(m_cfg)
         self._handle = handle  # This is a py::capsule
+        self._profiling_data = None
 
     @classmethod
     def _load_from(cls,
                    local_path: str,
                    mmproj_path: str,
                    m_cfg: ModelConfig = ModelConfig(),
-                   plugin_id: str = "llama_cpp",
+                   plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
                    device_id: Optional[str] = None
         ) -> 'PyBindVLMImpl':
         """Load VLM model from local path.
@@ -61,11 +63,13 @@ class PyBindVLMImpl(VLM):
             config.chat_template_content = m_cfg.chat_template_content
 
         # Create handle : returns py::capsule with automatic cleanup
+        # Convert enum to string for C++ binding
+        plugin_id_str = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
         handle = vlm_bind.create_vlm(
             model_path=local_path,
             mmproj_path=mmproj_path,
             model_config=config,
-            plugin_id=plugin_id,
+            plugin_id=plugin_id_str,
             device_id=device_id
         )
         return cls(handle, m_cfg)
@@ -141,13 +145,14 @@ class PyBindVLMImpl(VLM):
         # Run generation in thread
         def generate():
             try:
-                vlm_bind.ml_vlm_generate(
+                result = vlm_bind.ml_vlm_generate(
                     handle=self._handle,
                     prompt=prompt,
                     config=config,
                     on_token=on_token,
                     user_data=None
                 )
+                self._profiling_data = ProfilingData.from_dict(result.get("profile_data", {}))
             except Exception as e:
                 exception_container[0] = e
             finally:
@@ -189,8 +194,14 @@ class PyBindVLMImpl(VLM):
             on_token=None,  # No callback for non-streaming
             user_data=None
         )
+
+        self._profiling_data = ProfilingData.from_dict(result.get("profile_data", {}))
         return result.get("text", "")
 
+    def get_profiling_data(self) -> Optional[ProfilingData]:
+        """Get profiling data."""
+        return self._profiling_data
+
     def _convert_generation_config(self, g_cfg: GenerationConfig):
         """Convert GenerationConfig to binding format."""
         config = common_bind.GenerationConfig()

{nexaai-1.0.4rc15.dist-info → nexaai-1.0.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexaai
-Version: 1.0.4rc15
+Version: 1.0.5
 Summary: Python bindings for NexaSDK C-lib backend
 Author-email: "Nexa AI, Inc." <dev@nexa.ai>
 Project-URL: Homepage, https://github.com/NexaAI/nexasdk-bridge
@@ -17,11 +17,15 @@ Requires-Dist: tqdm
 Requires-Dist: hf_xet
 Requires-Dist: numpy
 Requires-Dist: httpx
-Requires-Dist: mlx
-Requires-Dist: mlx-lm
-Requires-Dist: scipy
-Requires-Dist: soundfile
-Requires-Dist: Pillow
-Requires-Dist: opencv-python
-Requires-Dist: shapely
-Requires-Dist: pyclipper
+Provides-Extra: mlx
+Requires-Dist: mlx; extra == "mlx"
+Requires-Dist: mlx-lm; extra == "mlx"
+Requires-Dist: mlx-vlm; extra == "mlx"
+Requires-Dist: tokenizers; extra == "mlx"
+Requires-Dist: safetensors; extra == "mlx"
+Requires-Dist: Pillow; extra == "mlx"
+Requires-Dist: scipy; extra == "mlx"
+Requires-Dist: soundfile; extra == "mlx"
+Requires-Dist: opencv-python; extra == "mlx"
+Requires-Dist: shapely; extra == "mlx"
+Requires-Dist: pyclipper; extra == "mlx"

nexaai-1.0.5.dist-info/RECORD

@@ -0,0 +1,61 @@
+nexaai/__init__.py,sha256=Lt8NU57eTMtWrDYzpFeYR9XtGAPXqizynP83TPU0UW0,2105
+nexaai/_stub.cp310-win_amd64.pyd,sha256=p44ZGmcsoolJ5aI2Oqu053_P9hSscKzoTQfDG6Lk4S0,10752
+nexaai/_version.py,sha256=gMJqvRn_xrvChdUIgn-DExgBJixnRda2-paThBgt3s0,142
+nexaai/asr.py,sha256=_fsGaxpiU137bUtO5ujtFSYCI1RLsyeEm3Gf4GhHVRk,2118
+nexaai/base.py,sha256=qQBCiQVNzgpkQjZX9aiFDEdbAAe56TROKC3WnWra2Zg,1021
+nexaai/common.py,sha256=6keIpdX5XS5us4z79EMoa6RSkVze9SbbXax13IJ9yvs,3525
+nexaai/cv.py,sha256=a6-csgYNDzPziJ0EojE9-BeM_xCny4UvWWbpnJ7GL-A,3365
+nexaai/embedder.py,sha256=3a81s7JapvYxCRbWPFKp_9EWBKW7WYqF03gk87YuGKU,2524
+nexaai/image_gen.py,sha256=4iASOKxJosMznLarTvOxJQDNaas251O81bfUWJTUBfE,4496
+nexaai/llm.py,sha256=Qwm1q_NStLfD-JYZQIvxniWnAmwNl1V6LUON3Me7w_I,3663
+nexaai/rerank.py,sha256=_zGWmX6eDigY2kViMKCtNssp4JMEeVycZZfJH9eAZOY,1927
+nexaai/runtime.py,sha256=LxAUejH9uaci8IGz9_h0l-MMeYcwTlBjVKN_0u4Q4Qo,2021
+nexaai/tts.py,sha256=afs6sx0w0Tvs_aJlyZRPm62qQpTrs-fW_jDHrMkc4AA,2241
+nexaai/vlm.py,sha256=STjXCw67ABrHrEll8A2NGiwmfo7MotfYgBh1k1aNxkk,4775
+nexaai/asr_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/asr_impl/mlx_asr_impl.py,sha256=XwMX3LYMeulp8cDS0TCCYcjvttFHAyDWQ_oMvABwQmI,3349
+nexaai/asr_impl/pybind_asr_impl.py,sha256=20o5SOPzhF9x41ra8L_qIM7YxCkYeLb5csSrNde-dds,1560
+nexaai/binds/__init__.py,sha256=tYvy0pFhoY29GstDT5r-oRiPRarPLECvJAkcamJItOg,83
+nexaai/binds/common_bind.cp310-win_amd64.pyd,sha256=wXqQGKiddXq8rQA2SluwuCmfvKlLK9bNt3IyhP-Q2t8,201216
+nexaai/binds/embedder_bind.cp310-win_amd64.pyd,sha256=jhCkw3xHYir2zJBnSwbLH6dsWuD4loaU2F44cvlShD4,182784
+nexaai/binds/libcrypto-3-x64.dll,sha256=-Lau6pL5DpDXzpg9MED63gCeL8oRrSLI_e2LeaxIHqk,7314432
+nexaai/binds/libssl-3-x64.dll,sha256=Tzzyu5jRpUugFxr_65hbFlAtFpjxIDpOYMU1E0ijkJw,1313792
+nexaai/binds/llm_bind.cp310-win_amd64.pyd,sha256=mbtjTaQ_Cc90HYs76QCs_T5DY_fZ7VpfzoTYGPSX2Vg,162816
+nexaai/binds/nexa_bridge.dll,sha256=taCRtuNWJQVvEv_2d2K1o0FAmHPDPD5Uez2npNkT700,178688
+nexaai/binds/nexa_llama_cpp/ggml-base.dll,sha256=b1Ei-roaQLUBPNd4PxCUAMAR4jyv-GIwbA144xttdXY,514560
+nexaai/binds/nexa_llama_cpp/ggml-cpu.dll,sha256=PNt0gYfIfuywKVYr4DyTTeVv8QN2-9nDKA6pI08TtkQ,663552
+nexaai/binds/nexa_llama_cpp/ggml-cuda.dll,sha256=4dJy4K4ZCCYMkL-9zcMsn10By_sMK_pfiqL7X3EOJSI,315100160
+nexaai/binds/nexa_llama_cpp/ggml-vulkan.dll,sha256=yJrka26Udhgzu3ZJKEldeHP1G0yw6Jo6830CD0mFifE,26204160
+nexaai/binds/nexa_llama_cpp/ggml.dll,sha256=WaVwGkJbgg2Cq6lJTKEZop0Av74YUh8M_-oovSdhtJ0,66560
+nexaai/binds/nexa_llama_cpp/llama.dll,sha256=oWDzqoNqkx4Klli2ZNEtBVIZuQ9ri1IyuK8mQNB7wlw,1587712
+nexaai/binds/nexa_llama_cpp/mtmd.dll,sha256=0UfR_8qiJMnIOCdHT9i9lZjBs2mKazhCm-1exSFCbVY,560128
+nexaai/binds/nexa_llama_cpp/nexa_plugin.dll,sha256=HcqcbJbEDPVTphVdqTt3CxQjFeQtQZAMSW1omX0JcKM,1384448
+nexaai/cv_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/cv_impl/mlx_cv_impl.py,sha256=QLd_8w90gtxH8kmssaDYatCTRvQNIJuUGKZNnYrmx6E,3317
+nexaai/cv_impl/pybind_cv_impl.py,sha256=aSOCAxmHrwJbEkSN6VX3Cykqlj_9RIpVrZXILul04GA,1096
+nexaai/embedder_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/embedder_impl/mlx_embedder_impl.py,sha256=MN5vAGohgyEreLn3H7dg2JeWp2v8emLhrfIDGndk-us,4498
+nexaai/embedder_impl/pybind_embedder_impl.py,sha256=FoLsUrzF5cNtEsSFchPlapkdqLGFOUGNPx0Kc8hdCvA,3589
+nexaai/image_gen_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/image_gen_impl/mlx_image_gen_impl.py,sha256=peUE9ue9ApaPlZVOICBWiHtd13sY40OWQbE8EjfIUMU,11511
+nexaai/image_gen_impl/pybind_image_gen_impl.py,sha256=514RFQMSO0Rhacq0IYzlEhEr6QfaprnGew0Rjz8HZI4,3777
+nexaai/llm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/llm_impl/mlx_llm_impl.py,sha256=r9Qa1ZuduRAcRjmqo2J_zlqhprVPe3ioZtUbi7yIvYY,11426
+nexaai/llm_impl/pybind_llm_impl.py,sha256=s2Cb035xDbh1ZhGNys0LRtAR6b6n4-YVKSC5voD4_Tk,8269
+nexaai/rerank_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/rerank_impl/mlx_rerank_impl.py,sha256=x8L6zCccV2I4bq4V5zAOcrWFe5Ckle7z5J_00tao8bI,3335
+nexaai/rerank_impl/pybind_rerank_impl.py,sha256=1IW925bYv4FRwZDNzf9epEzdDqR6T3OONgTLBd-cOn8,1556
+nexaai/tts_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/tts_impl/mlx_tts_impl.py,sha256=LcH9bVdIl3Q6lOzSUB_X2s-_nWFmlCl1yL7XSUK0fYI,3195
+nexaai/tts_impl/pybind_tts_impl.py,sha256=n3z4zmPQayQJgAwcvETw0IBUCp8IYROuYFSg0tAy_8Y,1487
+nexaai/utils/avatar_fetcher.py,sha256=D01f8je-37Nd68zGw8MYK2m7y3fvGlC6h0KR-aN9kdU,3925
+nexaai/utils/decode.py,sha256=0Z9jDH4ICzw4YXj8nD4L-sMouDaev-TISGRQ4KzidWE,421
+nexaai/utils/model_manager.py,sha256=Ksl-tKq-a3miTUxEn6-SSOC_KVdn6RPjcUdkWmDDwCk,49767
+nexaai/utils/progress_tracker.py,sha256=FmJBoOlzfQdc-TmccEav0cBR_iSNrrcskG3Fm1OrEJA,15482
+nexaai/vlm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nexaai/vlm_impl/mlx_vlm_impl.py,sha256=oY_qb9z_iF0zArBuY5CCYIvZcA3R0i_NKXrr_r-QSgg,10989
+nexaai/vlm_impl/pybind_vlm_impl.py,sha256=Hu8g8OXyPn8OzLQOpRSE5lfGmhjChiKj7fMRB8mC_cI,9147
+nexaai-1.0.5.dist-info/METADATA,sha256=JhdQmzKpw_EKjD-jcgZHKzn-KRMIoQEKbRGPlTwxrm8,1182
+nexaai-1.0.5.dist-info/WHEEL,sha256=KUuBC6lxAbHCKilKua8R9W_TM71_-9Sg5uEP3uDWcoU,101
+nexaai-1.0.5.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
+nexaai-1.0.5.dist-info/RECORD,,

nexaai-1.0.4rc15.dist-info/RECORD

@@ -1,59 +0,0 @@
-nexaai/__init__.py,sha256=d1bC_PUNduXYYPLrzKLyS0RapvcrKzLQGJREsoBZvXM,1977
-nexaai/_stub.cp310-win_amd64.pyd,sha256=C8_5OyVPadKYiSnRiPWYNr6Y7TJCu76TAZV3lBhcqjI,10752
-nexaai/_version.py,sha256=65huyjVsErx2svghMx8d4OaeLV72OhoKTE2Wt_RlEgk,147
-nexaai/asr.py,sha256=1XnwbrSoweBfIVAH6LbILv0DMStTQe_Uq5U_f-EyArY,1873
-nexaai/base.py,sha256=qQBCiQVNzgpkQjZX9aiFDEdbAAe56TROKC3WnWra2Zg,1021
-nexaai/common.py,sha256=00cP8uT9NdldBI3dRNHrQFx-uhdgtOGGxRAx4p96nw4,1586
-nexaai/cv.py,sha256=90lrW6o77E6uNMk5MYWsLp2f-fhLacjWTT1ENDhVYEg,3120
-nexaai/embedder.py,sha256=FtJtMKrniejTCi8_-ePLOymfkH8j1VzUqteOqGy5cO4,2279
-nexaai/image_gen.py,sha256=oliLxFN7Bd_3wzP4F6frMJ7GPvRn-1kn_8kAtdcy_pY,4258
-nexaai/llm.py,sha256=7V60E1cI1tt6CZ1ti2-tPqkYS56TcJE_kIhvyRyIBeQ,3194
-nexaai/rerank.py,sha256=l6KtNX9TMGjrtighTQrpXpG_XUlSoZ21AMDDFwR1acI,1682
-nexaai/runtime.py,sha256=LxAUejH9uaci8IGz9_h0l-MMeYcwTlBjVKN_0u4Q4Qo,2021
-nexaai/tts.py,sha256=1tJ3dSwphqPIAQmhMvigcyVFMki6_CC3yofCmaV_H9I,1996
-nexaai/vlm.py,sha256=Cnnw9gy9PaC46WvrCoF5MwFA_iUPaBQGGQ98THSn1jA,4410
-nexaai/asr_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/asr_impl/mlx_asr_impl.py,sha256=UpGOtl4JZmcmDRa56z5OyIozFVjfsYpKw_vYU_7HoWk,3282
-nexaai/asr_impl/pybind_asr_impl.py,sha256=C0Qnx-WDNmyC72dxuZVfUwuAoUSMvpo8IfOOkEbqsFA,1493
-nexaai/binds/__init__.py,sha256=tYvy0pFhoY29GstDT5r-oRiPRarPLECvJAkcamJItOg,83
-nexaai/binds/common_bind.cp310-win_amd64.pyd,sha256=tsCwzARi5fu-66q24KKLnWnBhNOt4UEbXbssgv4MPTw,201216
-nexaai/binds/embedder_bind.cp310-win_amd64.pyd,sha256=-JroOPxqlTEIU1BirfvjE0vqF__WCvsJaiuE3toWoyI,182784
-nexaai/binds/llm_bind.cp310-win_amd64.pyd,sha256=e7BWEt6TzJocCe_Qb3dB1c-73n6ETL9uPOMuTEsT0Ew,160256
-nexaai/binds/nexa_bridge.dll,sha256=HPGaSjp3uuNGWXKhxvNbyiwxuyLFWr6TzIBxTlaweqQ,176640
-nexaai/binds/nexa_llama_cpp/ggml-base.dll,sha256=WiLohLZ6F5t9Vc0o65CjOMijPjuSH6-G586FOwOJLFE,514560
-nexaai/binds/nexa_llama_cpp/ggml-cpu.dll,sha256=vUHtE3B6dAuJu-WCmXCD7GKayM5HBcoYHw7DVXMvG-4,663552
-nexaai/binds/nexa_llama_cpp/ggml-cuda.dll,sha256=2O7SF6N6VUYPq2yf3l-KFvEpee2X6Yh_7vSbfgWVi2o,315100160
-nexaai/binds/nexa_llama_cpp/ggml-vulkan.dll,sha256=vhNQXlF9_0f5WQkzckQ4lGzmI0ccTPWDODhWlej1ebs,26204160
-nexaai/binds/nexa_llama_cpp/ggml.dll,sha256=VYw6ZxxvlZf4Gi2jkTaOeWSB_pZCIuex4hBd6cWwSWI,66560
-nexaai/binds/nexa_llama_cpp/llama.dll,sha256=Zp6IS6RSJ72vDulABng_9Q59tKvi3znbm13pJwazzNY,1587712
-nexaai/binds/nexa_llama_cpp/mtmd.dll,sha256=9350YqtbNICY1eV1gLB7flBu6NpRUoMf3ESIyH3uoiI,560128
-nexaai/binds/nexa_llama_cpp/nexa_plugin.dll,sha256=DeKJPc4fgbgC14iBPghfjGID9vuT3FEkvx9FT6BDk1s,1086464
-nexaai/cv_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/cv_impl/mlx_cv_impl.py,sha256=vYN8ASbDr-VlQcia0ydpE3tUfnRcRIoRVQDAOhlZB_4,3250
-nexaai/cv_impl/pybind_cv_impl.py,sha256=oXT7Hcurg2YH_qgvwpGtgeQcIFxt6uzT9xN-cLvRHcU,1029
-nexaai/embedder_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/embedder_impl/mlx_embedder_impl.py,sha256=9tQ2-t31r7oo0WcvN3yWIYsk6VQgqSgcILEg0iz3x2I,4431
-nexaai/embedder_impl/pybind_embedder_impl.py,sha256=-sQkWxtwU_SpcSW6OcddmTn4YVTTRBTNlZRDdktKagk,3377
-nexaai/image_gen_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/image_gen_impl/mlx_image_gen_impl.py,sha256=q-kXDj7Pg0xTuI1XKCbCigU5HtXV2HNJOXWx85YznLI,11444
-nexaai/image_gen_impl/pybind_image_gen_impl.py,sha256=M3l9cgQStbStmJ2mtS2NRBZcjp0FJjDlHOhh8xhoaGI,3710
-nexaai/llm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/llm_impl/mlx_llm_impl.py,sha256=oJePBkpvQijIxTNgF6sj_lmWKbV9Egh0wJIkGrTHzFc,10459
-nexaai/llm_impl/pybind_llm_impl.py,sha256=ecDlzQ_4WE_xBE2pA98nkHdDdJgctpzK9IcpFINCdPc,7581
-nexaai/rerank_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/rerank_impl/mlx_rerank_impl.py,sha256=litO0iDf7KUrl8hE3ADETIvLkZcosroSgMpa3HEVFvQ,3268
-nexaai/rerank_impl/pybind_rerank_impl.py,sha256=F-bsWJAS5ZBr4ZOwecZJwiZeB4uEGg0u3bSJV035NDA,1489
-nexaai/tts_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/tts_impl/mlx_tts_impl.py,sha256=I7_ladKIhzn3C7zcx5_dN8EcGlgmRjWSFfwQeqTza9s,3128
-nexaai/tts_impl/pybind_tts_impl.py,sha256=JmWWoluN6I8325IwPL_Np8RLUq_yMENPoyszQ9G3B3M,1420
-nexaai/utils/avatar_fetcher.py,sha256=D01f8je-37Nd68zGw8MYK2m7y3fvGlC6h0KR-aN9kdU,3925
-nexaai/utils/decode.py,sha256=0Z9jDH4ICzw4YXj8nD4L-sMouDaev-TISGRQ4KzidWE,421
-nexaai/utils/model_manager.py,sha256=Ksl-tKq-a3miTUxEn6-SSOC_KVdn6RPjcUdkWmDDwCk,49767
-nexaai/utils/progress_tracker.py,sha256=FmJBoOlzfQdc-TmccEav0cBR_iSNrrcskG3Fm1OrEJA,15482
-nexaai/vlm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/vlm_impl/mlx_vlm_impl.py,sha256=O0NA0zkJ4mhv5lT4EBSVineGe_F0Rsz4ITQvaX_VgIk,10651
-nexaai/vlm_impl/pybind_vlm_impl.py,sha256=WGC21LaQnr4LEiytXX8wVBMLJDG75GIweW3kEMxJGXE,8561
-nexaai-1.0.4rc15.dist-info/METADATA,sha256=eHwHTGzl4PN_GfTApdOqE5dQTesECMWRo6r6CYZe7rY,910
-nexaai-1.0.4rc15.dist-info/WHEEL,sha256=KUuBC6lxAbHCKilKua8R9W_TM71_-9Sg5uEP3uDWcoU,101
-nexaai-1.0.4rc15.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
-nexaai-1.0.4rc15.dist-info/RECORD,,