nexaai 1.0.16rc9__cp310-cp310-macosx_14_0_universal2.whl → 1.0.16rc10__cp310-cp310-macosx_14_0_universal2.whl

nexaai/__init__.py

@@ -21,9 +21,6 @@ except ImportError:
 # Import common configuration classes first (no external dependencies)
 from .common import ModelConfig, GenerationConfig, ChatMessage, SamplerConfig, PluginID
 
-# Import logging functionality
-from .log import set_logger, get_error_message
-
 # Create alias for PluginID to be accessible as plugin_id
 plugin_id = PluginID
 
@@ -48,10 +45,6 @@ __all__ = [
     "EmbeddingConfig",
     "PluginID",
     "plugin_id",
-    
-    # Logging functionality
-    "set_logger",
-    "get_error_message",
 
     "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__ = "1.0.16-rc9" 
+__version__ = "1.0.16-rc10" 

nexaai/binds/nexa_mlx/py-lib/ml.py

@@ -1,9 +1,6 @@
 # This file defines the python interface that c-lib expects from a python backend
 
 from __future__ import annotations
-from typing import Optional
-from pathlib import Path
-from dataclasses import dataclass
 
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
@@ -104,12 +101,9 @@ class ModelConfig:
     n_threads_batch: int = 0  # number of threads to use for batch processing
     n_batch: int = 0  # logical maximum batch size that can be submitted to llama_decode
     n_ubatch: int = 0  # physical maximum batch size
-    # max number of sequences (i.e. distinct states for recurrent models)
-    n_seq_max: int = 0
-    # path to chat template file, optional
-    chat_template_path: Optional[Path] = None
-    # content of chat template file, optional
-    chat_template_content: Optional[str] = None
+    n_seq_max: int = 0  # max number of sequences (i.e. distinct states for recurrent models)
+    chat_template_path: Optional[Path] = None  # path to chat template file, optional
+    chat_template_content: Optional[str] = None  # content of chat template file, optional
 
 
 @dataclass
@@ -124,8 +118,7 @@ class SamplerConfig:
     frequency_penalty: float = 0.0
     seed: int = -1  # –1 for random
     grammar_path: Optional[Path] = None
-    # Optional grammar string (BNF-like format)
-    grammar_string: Optional[str] = None
+    grammar_string: Optional[str] = None  # Optional grammar string (BNF-like format)
 
 
 @dataclass
@@ -135,10 +128,8 @@ class GenerationConfig:
     stop: Sequence[str] = field(default_factory=tuple)
     n_past: int = 0
     sampler_config: Optional[SamplerConfig] = None
-    # Array of image paths for VLM (None if none)
-    image_paths: Optional[Sequence[Path]] = None
-    # Array of audio paths for VLM (None if none)
-    audio_paths: Optional[Sequence[Path]] = None
+    image_paths: Optional[Sequence[Path]] = None  # Array of image paths for VLM (None if none)
+    audio_paths: Optional[Sequence[Path]] = None  # Array of audio paths for VLM (None if none)
 
 
 @dataclass
@@ -179,32 +170,6 @@ class RerankConfig:
     normalize_method: str = "softmax"  # "softmax" | "min-max" | "none"
 
 
-# image-gen
-
-
-@dataclass
-class ImageGenTxt2ImgInput:
-    """Input structure for text-to-image generation."""
-    prompt: str
-    config: ImageGenerationConfig
-    output_path: Optional[Path] = None
-
-
-@dataclass
-class ImageGenImg2ImgInput:
-    """Input structure for image-to-image generation."""
-    init_image_path: Path
-    prompt: str
-    config: ImageGenerationConfig
-    output_path: Optional[Path] = None
-
-
-@dataclass
-class ImageGenOutput:
-    """Output structure for image generation."""
-    output_image_path: Path
-
-
 @dataclass
 class ImageSamplerConfig:
     """Configuration for image sampling."""
@@ -215,27 +180,17 @@ class ImageSamplerConfig:
     seed: int = -1  # –1 for random
 
 
-@dataclass
-class ImageGenCreateInput:
-    """Configuration for image generation."""
-    model_name: str
-    model_path: Path
-    config: ModelConfig
-    scheduler_config_path: Path
-    plugin_id: str
-    device_id: Optional[str] = None
-
-
 @dataclass
 class ImageGenerationConfig:
     """Configuration for image generation."""
-    prompts: List[str]
-    sampler_config: ImageSamplerConfig
-    scheduler_config: SchedulerConfig
-    strength: float
-    negative_prompts: Optional[List[str]] = None
+    prompts: str | List[str]
+    negative_prompts: str | List[str] | None = None
     height: int = 512
     width: int = 512
+    sampler_config: Optional[ImageSamplerConfig] = None
+    lora_id: int = -1  # –1 for none
+    init_image: Optional[Image] = None
+    strength: float = 1.0
 
 
 @dataclass
@@ -306,7 +261,7 @@ class TTSResult:
 class BoundingBox:
     """Generic bounding box structure."""
     x: float  # X coordinate (normalized or pixel, depends on model)
-    y: float  # Y coordinate (normalized or pixel, depends on model)
+    y: float  # Y coordinate (normalized or pixel, depends on model) 
     width: float  # Width
     height: float  # Height
 
@@ -320,8 +275,7 @@ class CVResult:
     confidence: float = 0.0  # Confidence score [0.0-1.0]
     bbox: Optional[BoundingBox] = None  # Bounding box (example: YOLO)
     text: Optional[str] = None  # Text result (example: OCR)
-    # Feature embedding (example: CLIP embedding)
-    embedding: Optional[List[float]] = None
+    embedding: Optional[List[float]] = None  # Feature embedding (example: CLIP embedding)
     embedding_dim: int = 0  # Embedding dimension
 
 

nexaai/mlx_backend/ml.py

@@ -1,9 +1,6 @@
 # This file defines the python interface that c-lib expects from a python backend
 
 from __future__ import annotations
-from typing import Optional
-from pathlib import Path
-from dataclasses import dataclass
 
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
@@ -104,12 +101,9 @@ class ModelConfig:
     n_threads_batch: int = 0  # number of threads to use for batch processing
     n_batch: int = 0  # logical maximum batch size that can be submitted to llama_decode
     n_ubatch: int = 0  # physical maximum batch size
-    # max number of sequences (i.e. distinct states for recurrent models)
-    n_seq_max: int = 0
-    # path to chat template file, optional
-    chat_template_path: Optional[Path] = None
-    # content of chat template file, optional
-    chat_template_content: Optional[str] = None
+    n_seq_max: int = 0  # max number of sequences (i.e. distinct states for recurrent models)
+    chat_template_path: Optional[Path] = None  # path to chat template file, optional
+    chat_template_content: Optional[str] = None  # content of chat template file, optional
 
 
 @dataclass
@@ -124,8 +118,7 @@ class SamplerConfig:
     frequency_penalty: float = 0.0
     seed: int = -1  # –1 for random
     grammar_path: Optional[Path] = None
-    # Optional grammar string (BNF-like format)
-    grammar_string: Optional[str] = None
+    grammar_string: Optional[str] = None  # Optional grammar string (BNF-like format)
 
 
 @dataclass
@@ -135,10 +128,8 @@ class GenerationConfig:
     stop: Sequence[str] = field(default_factory=tuple)
     n_past: int = 0
     sampler_config: Optional[SamplerConfig] = None
-    # Array of image paths for VLM (None if none)
-    image_paths: Optional[Sequence[Path]] = None
-    # Array of audio paths for VLM (None if none)
-    audio_paths: Optional[Sequence[Path]] = None
+    image_paths: Optional[Sequence[Path]] = None  # Array of image paths for VLM (None if none)
+    audio_paths: Optional[Sequence[Path]] = None  # Array of audio paths for VLM (None if none)
 
 
 @dataclass
@@ -179,32 +170,6 @@ class RerankConfig:
     normalize_method: str = "softmax"  # "softmax" | "min-max" | "none"
 
 
-# image-gen
-
-
-@dataclass
-class ImageGenTxt2ImgInput:
-    """Input structure for text-to-image generation."""
-    prompt: str
-    config: ImageGenerationConfig
-    output_path: Optional[Path] = None
-
-
-@dataclass
-class ImageGenImg2ImgInput:
-    """Input structure for image-to-image generation."""
-    init_image_path: Path
-    prompt: str
-    config: ImageGenerationConfig
-    output_path: Optional[Path] = None
-
-
-@dataclass
-class ImageGenOutput:
-    """Output structure for image generation."""
-    output_image_path: Path
-
-
 @dataclass
 class ImageSamplerConfig:
     """Configuration for image sampling."""
@@ -215,27 +180,17 @@ class ImageSamplerConfig:
     seed: int = -1  # –1 for random
 
 
-@dataclass
-class ImageGenCreateInput:
-    """Configuration for image generation."""
-    model_name: str
-    model_path: Path
-    config: ModelConfig
-    scheduler_config_path: Path
-    plugin_id: str
-    device_id: Optional[str] = None
-
-
 @dataclass
 class ImageGenerationConfig:
     """Configuration for image generation."""
-    prompts: List[str]
-    sampler_config: ImageSamplerConfig
-    scheduler_config: SchedulerConfig
-    strength: float
-    negative_prompts: Optional[List[str]] = None
+    prompts: str | List[str]
+    negative_prompts: str | List[str] | None = None
     height: int = 512
     width: int = 512
+    sampler_config: Optional[ImageSamplerConfig] = None
+    lora_id: int = -1  # –1 for none
+    init_image: Optional[Image] = None
+    strength: float = 1.0
 
 
 @dataclass
@@ -306,7 +261,7 @@ class TTSResult:
 class BoundingBox:
     """Generic bounding box structure."""
     x: float  # X coordinate (normalized or pixel, depends on model)
-    y: float  # Y coordinate (normalized or pixel, depends on model)
+    y: float  # Y coordinate (normalized or pixel, depends on model) 
     width: float  # Width
     height: float  # Height
 
@@ -320,8 +275,7 @@ class CVResult:
     confidence: float = 0.0  # Confidence score [0.0-1.0]
     bbox: Optional[BoundingBox] = None  # Bounding box (example: YOLO)
     text: Optional[str] = None  # Text result (example: OCR)
-    # Feature embedding (example: CLIP embedding)
-    embedding: Optional[List[float]] = None
+    embedding: Optional[List[float]] = None  # Feature embedding (example: CLIP embedding)
     embedding_dim: int = 0  # Embedding dimension
 
 

nexaai/mlx_backend/sd/modeling/model_io.py

@@ -1,7 +1,6 @@
 # Copyright © 2023-2024 Apple Inc.
 
 import json
-import os
 from typing import Optional
 
 import mlx.core as mx
@@ -177,37 +176,19 @@ def _load_safetensor_weights(mapper, model, weight_file, float16: bool = False):
 
 
 def _check_key(key: str, part: str):
-    # Check if it's a local path
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # For local paths, we'll use a default model structure
-        return
     if key not in _MODELS:
         raise ValueError(
             f"[{part}] '{key}' model not found, choose one of {{{','.join(_MODELS.keys())}}}"
         )
 
-def _get_model_path(key: str, file_path: str):
-    """Get the full path for a model file, supporting both local and HuggingFace paths"""
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path
-        return os.path.join(key, file_path)
-    else:
-        # HuggingFace path
-        return hf_hub_download(key, file_path)
-
 
 def load_unet(key: str = _DEFAULT_MODEL, float16: bool = False):
     """Load the stable diffusion UNet from Hugging Face Hub."""
     _check_key(key, "load_unet")
 
-    # Get the config path
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        unet_config = "unet/config.json"
-    else:
-        unet_config = _MODELS[key]["unet_config"]
-    
-    with open(_get_model_path(key, unet_config)) as f:
+    # Download the config and create the model
+    unet_config = _MODELS[key]["unet_config"]
+    with open(hf_hub_download(key, unet_config)) as f:
         config = json.load(f)
 
     n_blocks = len(config["block_out_channels"])
@@ -238,13 +219,8 @@ def load_unet(key: str = _DEFAULT_MODEL, float16: bool = False):
     )
 
     # Download the weights and map them into the model
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        unet_weights = "unet/diffusion_pytorch_model.safetensors"
-    else:
-        unet_weights = _MODELS[key]["unet"]
-    
-    weight_file = _get_model_path(key, unet_weights)
+    unet_weights = _MODELS[key]["unet"]
+    weight_file = hf_hub_download(key, unet_weights)
     _load_safetensor_weights(map_unet_weights, model, weight_file, float16)
 
     return model
@@ -262,13 +238,8 @@ def load_text_encoder(
     config_key = config_key or (model_key + "_config")
 
     # Download the config and create the model
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        text_encoder_config = f"{model_key}/config.json"
-    else:
-        text_encoder_config = _MODELS[key][config_key]
-    
-    with open(_get_model_path(key, text_encoder_config)) as f:
+    text_encoder_config = _MODELS[key][config_key]
+    with open(hf_hub_download(key, text_encoder_config)) as f:
         config = json.load(f)
 
     with_projection = "WithProjection" in config["architectures"][0]
@@ -286,13 +257,8 @@ def load_text_encoder(
     )
 
     # Download the weights and map them into the model
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        text_encoder_weights = f"{model_key}/model.safetensors"
-    else:
-        text_encoder_weights = _MODELS[key][model_key]
-    
-    weight_file = _get_model_path(key, text_encoder_weights)
+    text_encoder_weights = _MODELS[key][model_key]
+    weight_file = hf_hub_download(key, text_encoder_weights)
     _load_safetensor_weights(map_clip_text_encoder_weights, model, weight_file, float16)
 
     return model
@@ -303,13 +269,8 @@ def load_autoencoder(key: str = _DEFAULT_MODEL, float16: bool = False):
     _check_key(key, "load_autoencoder")
 
     # Download the config and create the model
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        vae_config = "vae/config.json"
-    else:
-        vae_config = _MODELS[key]["vae_config"]
-    
-    with open(_get_model_path(key, vae_config)) as f:
+    vae_config = _MODELS[key]["vae_config"]
+    with open(hf_hub_download(key, vae_config)) as f:
         config = json.load(f)
 
     model = Autoencoder(
@@ -326,13 +287,8 @@ def load_autoencoder(key: str = _DEFAULT_MODEL, float16: bool = False):
     )
 
     # Download the weights and map them into the model
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        vae_weights = "vae/diffusion_pytorch_model.safetensors"
-    else:
-        vae_weights = _MODELS[key]["vae"]
-    
-    weight_file = _get_model_path(key, vae_weights)
+    vae_weights = _MODELS[key]["vae"]
+    weight_file = hf_hub_download(key, vae_weights)
     _load_safetensor_weights(map_vae_weights, model, weight_file, float16)
 
     return model
@@ -342,13 +298,8 @@ def load_diffusion_config(key: str = _DEFAULT_MODEL):
     """Load the stable diffusion config from Hugging Face Hub."""
     _check_key(key, "load_diffusion_config")
 
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        diffusion_config = "scheduler/scheduler_config.json"
-    else:
-        diffusion_config = _MODELS[key]["diffusion_config"]
-    
-    with open(_get_model_path(key, diffusion_config)) as f:
+    diffusion_config = _MODELS[key]["diffusion_config"]
+    with open(hf_hub_download(key, diffusion_config)) as f:
         config = json.load(f)
 
     return DiffusionConfig(
@@ -366,17 +317,11 @@ def load_tokenizer(
 ):
     _check_key(key, "load_tokenizer")
 
-    if os.path.exists(key) or '/' in key or '\\' in key:
-        # Local path - use SDXL Turbo structure
-        vocab_file = _get_model_path(key, f"tokenizer/{vocab_key.split('_')[1]}.json")
-        merges_file = _get_model_path(key, f"tokenizer/{merges_key.split('_')[1]}.txt")
-    else:
-        vocab_file = _get_model_path(key, _MODELS[key][vocab_key])
-        merges_file = _get_model_path(key, _MODELS[key][merges_key])
-    
+    vocab_file = hf_hub_download(key, _MODELS[key][vocab_key])
     with open(vocab_file, encoding="utf-8") as f:
         vocab = json.load(f)
 
+    merges_file = hf_hub_download(key, _MODELS[key][merges_key])
     with open(merges_file, encoding="utf-8") as f:
         bpe_merges = f.read().strip().split("\n")[1 : 49152 - 256 - 2 + 1]
     bpe_merges = [tuple(m.split()) for m in bpe_merges]

nexaai/runtime.py

@@ -28,10 +28,6 @@ def _shutdown_runtime() -> None:
 # Public helper so advanced users can reclaim memory on demand
 shutdown = _shutdown_runtime
 
-def is_initialized() -> bool:
-    """Check if the runtime has been initialized."""
-    return _runtime_alive
-
 # ----------------------------------------------------------------------
 # Single public class
 # ----------------------------------------------------------------------

{nexaai-1.0.16rc9.dist-info → nexaai-1.0.16rc10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexaai
-Version: 1.0.16rc9
+Version: 1.0.16rc10
 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

{nexaai-1.0.16rc9.dist-info → nexaai-1.0.16rc10.dist-info}/RECORD

@@ -1,6 +1,6 @@
-nexaai/__init__.py,sha256=L8oB7GFZZMGnUpCg0PecDbI_ycKuQak-ZEJ4Y12_QIw,2184
-nexaai/_stub.cpython-310-darwin.so,sha256=9ySacVKPwR75hGn30_RMxti9XUG59wKNCGLIcLt--lc,66768
-nexaai/_version.py,sha256=ofT_bYXEi4fs898uta8KVZWG30txiae2CuE5yZCLBU0,143
+nexaai/__init__.py,sha256=jXdC4vv6DBK1fVewYTYSUhOOYfvf_Mk81UIeMGGIKUg,2029
+nexaai/_stub.cpython-310-darwin.so,sha256=Izt7MU4HQisUgbHwKkOH1Uo_GRUa9INqjjoGcBiZv5s,66768
+nexaai/_version.py,sha256=rM7mysA3I6mJ-jXt3DwjydINMplIqs78L_Oid1wMpI8,144
 nexaai/asr.py,sha256=NljMXDErwPNMOPaRkJZMEDka9Nk8xyur7L8i924TStY,2054
 nexaai/base.py,sha256=N8PRgDFA-XPku2vWnQIofQ7ipz3pPlO6f8YZGnuhquE,982
 nexaai/common.py,sha256=yBnIbqYaQYnfrl7IczOBh6MDibYZVxwaRJEglYcKgGs,3422
@@ -8,19 +8,18 @@ nexaai/cv.py,sha256=RHCDo8gvBH8BkGZx7qVyp-OKxqi7E1GG9XzyaXehCNA,3273
 nexaai/embedder.py,sha256=Cw0tSHkPgd-RI62afCqQAcTHMnQhaI2CvfTMO-1JKOg,2452
 nexaai/image_gen.py,sha256=0C_5Tjj4BYmxLbmMmvwajp-yy2mmEEOKwBFnDQNPzx4,4356
 nexaai/llm.py,sha256=S1o_k2VQoF5w2wO25f142OO1R75TP89Ii69VZv8pIGo,3567
-nexaai/log.py,sha256=Kwo2CIfWN6iP4M4F5EUIV8KIO5hAsvz6HZAaOwJ27Og,2628
 nexaai/rerank.py,sha256=vWaBucoQ1wz-2iYnZqyFIcEjm-4Xcs1KDbFN5X8zzDQ,1872
-nexaai/runtime.py,sha256=JvllhlNPgYGLbgGyX2yNvmGzT0lZ5XbvTvEo8sZG_Ho,2067
+nexaai/runtime.py,sha256=mxxHYsb5iBUAm2K_u-XJWr_U-spJ9S4eApc8kf9myjw,1957
 nexaai/tts.py,sha256=ZnBpWUxIfHhh7KfEjddtH7hHOTa91zg7ogGLakMIALo,2167
 nexaai/vlm.py,sha256=OCxwML-Z5uVGp3fjzJVtbCxfTLpgxkhQ8Wo6MVysoiw,4733
 nexaai/asr_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/asr_impl/mlx_asr_impl.py,sha256=eosd8-TIWAOwV0HltmoFrLwzXHcU4jyxtncvuZE9pgA,3257
 nexaai/asr_impl/pybind_asr_impl.py,sha256=pE9Hb_hMi5yAc4MF83bLVOb8zDtreCkB3_u7XED9YpA,1516
 nexaai/binds/__init__.py,sha256=eYuay_8DDXeOUWz2_R9HFSabohxs6hvZn391t2L0Po0,104
-nexaai/binds/common_bind.cpython-310-darwin.so,sha256=QUgK0eTGRU6ULHEJW0slE5WCpfXI9TAzILrbfS9902Y,235168
+nexaai/binds/common_bind.cpython-310-darwin.so,sha256=hVxY76tn7hN6uHDIgM7LWNvgoudHgNZVoaygM9X1RWE,217232
 nexaai/binds/embedder_bind.cpython-310-darwin.so,sha256=tPa0c0Dv_GiW66fgmAGWGCHXRGNApznqoQS0eQx9GFM,202064
-nexaai/binds/libnexa_bridge.dylib,sha256=lCp46qRAzUtBO2yK9qise3qMYPQfSYozrgmLWVMVAQk,251192
-nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=TAWfa1Hzq00TjtC1xVsiAeLp6hv2LrL5afDz4omUghc,182784
+nexaai/binds/libnexa_bridge.dylib,sha256=EsxYPjSQUQ6sbqhVcfFF10beAY8EAudisHH56uB0xg0,251144
+nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=6vRO5eOsL6s-hCKdz6onIjBKR_wcpCIsXHtUJY9vJBI,182704
 nexaai/binds/vlm_bind.cpython-310-darwin.so,sha256=nd6eG_m2EiPthzkSZ97hlXWUOZQir4cQfFJZ4p6eR2U,182704
 nexaai/binds/nexa_llama_cpp/libggml-base.dylib,sha256=JM4oOkie1su0ES5hMdtILeQHlRukRzH1vTleTupUXhg,650736
 nexaai/binds/nexa_llama_cpp/libggml-cpu.so,sha256=qiYxbTe4Nt7n36zJVvq3zovgSZEmrN2is6gzTern7UI,677728
@@ -28,9 +27,9 @@ nexaai/binds/nexa_llama_cpp/libggml-metal.so,sha256=zfaX7rIBYQazH2lf-vza007BMhPT
 nexaai/binds/nexa_llama_cpp/libggml.dylib,sha256=aOTj_6RrAMkfDO0ZI28_3nfcC-l4Y3dRCiS3C0d0_eI,58592
 nexaai/binds/nexa_llama_cpp/libllama.dylib,sha256=RkBd5usb8RvEIOamvxCW3UvMauI5bC66G_n6hw83NpY,1786128
 nexaai/binds/nexa_llama_cpp/libmtmd.dylib,sha256=o6mQqefzQNF0CS4j6odwJKj0gkXm15hIxwlNt88FOn4,605248
-nexaai/binds/nexa_llama_cpp/libnexa_plugin.dylib,sha256=f-ctmVHSL5x1LKLr0StYM5WscvUW_5olJ7al3ciVmmQ,1863000
-nexaai/binds/nexa_mlx/libnexa_plugin.dylib,sha256=s4oelR1kkFrk0MKL23I0m9Uei4yPf48rzdNvknpjodg,623192
-nexaai/binds/nexa_mlx/py-lib/ml.py,sha256=DKXVOAfh8cg7KTKljh7jpcPwfQFNigc6uv_ZXF6lse8,23977
+nexaai/binds/nexa_llama_cpp/libnexa_plugin.dylib,sha256=4rTxwHGSVUfGnvA6laPpLztMMFooy-inG2798HXqKPY,1862104
+nexaai/binds/nexa_mlx/libnexa_plugin.dylib,sha256=pSLYF-LN-s6ZaJomznSY8G6cNeKNwnsE2gZAaLfUdTw,598328
+nexaai/binds/nexa_mlx/py-lib/ml.py,sha256=LafDM_TeXmuQkld2tdQxUBGgooT0JPMXngLam2TADqU,23179
 nexaai/binds/nexa_mlx/py-lib/profiling.py,sha256=Dc-mybFwBdCIKFWL7CbSHjkOJGAoYHG7r_e_XPhzwBU,9361
 nexaai/binds/nexa_mlx/py-lib/mlx_audio/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/binds/nexa_mlx/py-lib/mlx_audio/server.py,sha256=Pqy13Fafq4WX_cTuvRFz1jq89beQm2QQGpXmhK4b9jc,17547
@@ -194,7 +193,7 @@ nexaai/image_gen_impl/pybind_image_gen_impl.py,sha256=ms34VYoD5AxZFG6cIG0QAJDjCt
 nexaai/llm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/llm_impl/mlx_llm_impl.py,sha256=4v7jUFzHfE7zw2uViekGQDaTROz8A6oaW31Z3iVe6tg,11157
 nexaai/llm_impl/pybind_llm_impl.py,sha256=aooqkcXZWhCo07wbSafGgBrA3WnijtnUADShjjgFsBQ,8051
-nexaai/mlx_backend/ml.py,sha256=DKXVOAfh8cg7KTKljh7jpcPwfQFNigc6uv_ZXF6lse8,23977
+nexaai/mlx_backend/ml.py,sha256=LafDM_TeXmuQkld2tdQxUBGgooT0JPMXngLam2TADqU,23179
 nexaai/mlx_backend/profiling.py,sha256=Dc-mybFwBdCIKFWL7CbSHjkOJGAoYHG7r_e_XPhzwBU,9361
 nexaai/mlx_backend/asr/__init__.py,sha256=fuT_9_xpYJ28m4yjly5L2jChUrzlSQz-b_S7nujxkSM,451
 nexaai/mlx_backend/asr/interface.py,sha256=pE5expr8sP7O9bEgWaUaJ4ITX0MsCxFovG9iVWqVDVU,4246
@@ -211,18 +210,6 @@ nexaai/mlx_backend/embedding/interface.py,sha256=M7AGiq_UVLNIi2Ie6H08ySnMxIjIhUl
 nexaai/mlx_backend/embedding/main.py,sha256=xKRebBcooKuf8DzWKwCicftes3MAcYAd1QvcT9_AAPQ,6003
 nexaai/mlx_backend/embedding/modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/mlx_backend/embedding/modeling/nexa_jina_v2.py,sha256=F9Z_9r-Dh0wNThiMp5W5hqE2dt5bf4ps5_c6h4BuWGw,15218
-nexaai/mlx_backend/image_gen/__init__.py,sha256=8eFAF00-yWdEs0LJiszjHEsgPDAF8tSruBYvujCNgE0,42
-nexaai/mlx_backend/image_gen/generate_sd.py,sha256=8DuPsJlxVf1LlFARVThiW807G-0cWPReYONj-x8qKRo,8616
-nexaai/mlx_backend/image_gen/interface.py,sha256=7ElIiLm5gZXAfKPs497wLkn8qqLFLL5T-oV-hXVmEdw,3285
-nexaai/mlx_backend/image_gen/main.py,sha256=jKkGDfqmGIsQwhHJaL_j_CTtr04xplD6dnnGe3AcilU,8123
-nexaai/mlx_backend/image_gen/stable_diffusion/__init__.py,sha256=wriLb0wA5vCBlCoQMtfKrVVWMJw8fhXCCk6R9_Nrb9c,9524
-nexaai/mlx_backend/image_gen/stable_diffusion/clip.py,sha256=feHQXi1NiGa01AMo7nK8M-sgBoZBHI95xAGMfAv64kE,3733
-nexaai/mlx_backend/image_gen/stable_diffusion/config.py,sha256=lq2sWXevpnCk8KXOywEVOzgFT6WBNnG4xr7NLWgidII,1773
-nexaai/mlx_backend/image_gen/stable_diffusion/model_io.py,sha256=ei-pdrqFxjeHLL6Ov5NZaHT5lceqZbHDyNFfUFUh_04,13755
-nexaai/mlx_backend/image_gen/stable_diffusion/sampler.py,sha256=nJxiSmF79E2aDGzLqygOat4zpgaTzfR7Kp1PD0bmhWA,3397
-nexaai/mlx_backend/image_gen/stable_diffusion/tokenizer.py,sha256=WMcmyQmb5MmhCp-NNU72Bqjq9vQvDzgR8KAfGKA2Qso,2991
-nexaai/mlx_backend/image_gen/stable_diffusion/unet.py,sha256=rYj1-baAugL7CNyYYvLch891WUgpunhAJCVBNKxfZSA,14762
-nexaai/mlx_backend/image_gen/stable_diffusion/vae.py,sha256=sQ1943x5SKyTmeL2xe3t84FL9DGO72Eab2m88r-BSe8,8044
 nexaai/mlx_backend/llm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/mlx_backend/llm/generate.py,sha256=Phes0tzxbbEWA2hDylQvD0LjorMaPwvcfZq9RKCAOt0,4399
 nexaai/mlx_backend/llm/interface.py,sha256=SZFkuAUi2vxj_dSqj8RXf9vPTGMtpks_pZxxrF7iIe8,29330
@@ -390,7 +377,7 @@ nexaai/mlx_backend/sd/main.py,sha256=cHlZhG8KHDFXbYRH-diKA7B1Qacq1euuGw0pKeXJGAI
 nexaai/mlx_backend/sd/modeling/__init__.py,sha256=wriLb0wA5vCBlCoQMtfKrVVWMJw8fhXCCk6R9_Nrb9c,9524
 nexaai/mlx_backend/sd/modeling/clip.py,sha256=feHQXi1NiGa01AMo7nK8M-sgBoZBHI95xAGMfAv64kE,3733
 nexaai/mlx_backend/sd/modeling/config.py,sha256=lq2sWXevpnCk8KXOywEVOzgFT6WBNnG4xr7NLWgidII,1773
-nexaai/mlx_backend/sd/modeling/model_io.py,sha256=xUjF5XNUfRLHQz-LtGT_D3XGQ1MI7ZQWknmdUDMpi_s,13732
+nexaai/mlx_backend/sd/modeling/model_io.py,sha256=0jpMHJdMoTk0Jo4Uz1ZGiXekAWzPuTTD5g7ms4m4idY,11668
 nexaai/mlx_backend/sd/modeling/sampler.py,sha256=nJxiSmF79E2aDGzLqygOat4zpgaTzfR7Kp1PD0bmhWA,3397
 nexaai/mlx_backend/sd/modeling/tokenizer.py,sha256=WMcmyQmb5MmhCp-NNU72Bqjq9vQvDzgR8KAfGKA2Qso,2991
 nexaai/mlx_backend/sd/modeling/unet.py,sha256=rYj1-baAugL7CNyYYvLch891WUgpunhAJCVBNKxfZSA,14762
@@ -541,7 +528,7 @@ nexaai/utils/quantization_utils.py,sha256=4gvp6UQfSO9G1FYBwnFtQspTzH9sDbi1PBXw2t
 nexaai/vlm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/vlm_impl/mlx_vlm_impl.py,sha256=pLtWm_ckz8a0U-AtAOMVseFDO4OVPvHyYO2KlfBaGYk,10833
 nexaai/vlm_impl/pybind_vlm_impl.py,sha256=FAbhpRJzHgI78r0mUvKybO97R1szvNhH0aTn_I52oT4,8597
-nexaai-1.0.16rc9.dist-info/METADATA,sha256=UmBO16cEnDyU2RUNoLt9UpeZcQ6QpZ9SZmkBxvhQyrI,1201
-nexaai-1.0.16rc9.dist-info/WHEEL,sha256=T2p57lol9__xkoU6aJTyN1Pm43ZpRU3q6km7mIbrAMs,114
-nexaai-1.0.16rc9.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
-nexaai-1.0.16rc9.dist-info/RECORD,,
+nexaai-1.0.16rc10.dist-info/METADATA,sha256=_p5OXiFfM3Nb6BN_grqE9--aJ4A9bz_B84UPqWYxp9I,1202
+nexaai-1.0.16rc10.dist-info/WHEEL,sha256=T2p57lol9__xkoU6aJTyN1Pm43ZpRU3q6km7mIbrAMs,114
+nexaai-1.0.16rc10.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
+nexaai-1.0.16rc10.dist-info/RECORD,,