nexaai 1.0.4rc16__cp310-cp310-macosx_13_0_x86_64.whl → 1.0.6rc1__cp310-cp310-macosx_13_0_x86_64.whl

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-rc16" 
+__version__ = "1.0.6-rc1" 

nexaai/common.py

@@ -59,10 +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/llm.py

@@ -4,7 +4,7 @@ import queue
 import threading
 
 from nexaai.common import ModelConfig, GenerationConfig, ChatMessage, PluginID
-from nexaai.base import BaseModel
+from nexaai.base import BaseModel, ProfilingData
 
 class LLM(BaseModel):
     def __init__(self, m_cfg: ModelConfig = ModelConfig()):
@@ -63,6 +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

@@ -1,5 +1,6 @@
 from typing import Generator, Optional, Any, Sequence, Union
 
+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
@@ -215,6 +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

@@ -2,6 +2,7 @@ from typing import Generator, Optional, Union
 import queue
 import threading
 
+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
@@ -13,6 +14,7 @@ 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,
@@ -97,13 +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:
@@ -145,8 +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/mlx_backend/llm/interface.py

@@ -371,19 +371,19 @@ class LLM(BaseLLM, ProfilingMixin):
         cached_tokens = 0
 
         # Only offset prefix kv-cache at first round
-        if is_first_round:
+        # if is_first_round:
         
-            # Handle KV cache prefix offset if available
-            if self.kv_cache is not None and len(self.kv_cache) > 0:
-                # Get the offset from the first cache layer
-                if hasattr(self.kv_cache[0], 'offset'):
-                    cached_tokens = self.kv_cache[0].offset - 1
+        #     # Handle KV cache prefix offset if available
+        #     if self.kv_cache is not None and len(self.kv_cache) > 0:
+        #         # Get the offset from the first cache layer
+        #         if hasattr(self.kv_cache[0], 'offset'):
+        #             cached_tokens = self.kv_cache[0].offset - 1
             
-            # Process only the non-cached tokens
-            incremental_tokens = incremental_tokens[cached_tokens:] if cached_tokens > 0 else incremental_tokens
+        #     # Process only the non-cached tokens
+        #     incremental_tokens = incremental_tokens[cached_tokens:] if cached_tokens > 0 else incremental_tokens
             
-            if len(incremental_tokens) == 0:
-                raise ValueError("No tokens to process, KV cache is too long.")
+        #     if len(incremental_tokens) == 0:
+        #         raise ValueError("No tokens to process, KV cache is too long.")
             
         # Since apply_chat_template now returns incremental prompts, we can use the prompt directly
         # The prompt is already the incremental part based on global_n_past

nexaai/vlm.py

@@ -6,7 +6,7 @@ import base64
 from pathlib import Path
 
 from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage, PluginID
-from nexaai.base import BaseModel
+from nexaai.base import BaseModel, ProfilingData
 
 
 class VLM(BaseModel):
@@ -120,4 +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

@@ -1,5 +1,6 @@
 from typing import Generator, Optional, List, Dict, Any, Union
 
+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
@@ -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

@@ -8,6 +8,7 @@ from nexaai.common import ModelConfig, GenerationConfig, MultiModalMessage, Plug
 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):
@@ -15,6 +16,7 @@ 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,
@@ -143,13 +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:
@@ -191,8 +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-1.0.4rc16.dist-info → nexaai-1.0.6rc1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexaai
-Version: 1.0.4rc16
+Version: 1.0.6rc1
 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-1.0.4rc16.dist-info → nexaai-1.0.6rc1.dist-info}/RECORD

@@ -1,17 +1,17 @@
 nexaai/__init__.py,sha256=jXdC4vv6DBK1fVewYTYSUhOOYfvf_Mk81UIeMGGIKUg,2029
-nexaai/_stub.cpython-310-darwin.so,sha256=7KDZERgfp9KR_dpQvZ7SMrGjlJpPyezLP7v-rYCdqFA,49832
-nexaai/_version.py,sha256=NGCgH5JHTkWsbmkVT9FhcM7m4cxgmEZiw51TUG210EA,143
+nexaai/_stub.cpython-310-darwin.so,sha256=i1tMWQy3ZjzlbqlF_pVQbifzoGO0ebEzk8dy5GEgn7Y,49832
+nexaai/_version.py,sha256=o8WPRe-h5be83JEwTPwBVdsZ20QQ2VFyIwzifvgbiPQ,142
 nexaai/asr.py,sha256=NljMXDErwPNMOPaRkJZMEDka9Nk8xyur7L8i924TStY,2054
 nexaai/base.py,sha256=N8PRgDFA-XPku2vWnQIofQ7ipz3pPlO6f8YZGnuhquE,982
-nexaai/common.py,sha256=5ElYo4uDP2CT3Kqxoo7XzqcJtDBuwwbIi_Wr14aT9Z4,1659
+nexaai/common.py,sha256=yBnIbqYaQYnfrl7IczOBh6MDibYZVxwaRJEglYcKgGs,3422
 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=QQDRg8zlu-xHmWjtSOsK1vhQBHaqRIdL3T9I4cVX7W4,3416
+nexaai/llm.py,sha256=S1o_k2VQoF5w2wO25f142OO1R75TP89Ii69VZv8pIGo,3567
 nexaai/rerank.py,sha256=vWaBucoQ1wz-2iYnZqyFIcEjm-4Xcs1KDbFN5X8zzDQ,1872
 nexaai/runtime.py,sha256=mxxHYsb5iBUAm2K_u-XJWr_U-spJ9S4eApc8kf9myjw,1957
 nexaai/tts.py,sha256=ZnBpWUxIfHhh7KfEjddtH7hHOTa91zg7ogGLakMIALo,2167
-nexaai/vlm.py,sha256=pZcMWkF2Ml9liVNbHxLqBJxwm2bxVNM1dkoelwWMyIE,4500
+nexaai/vlm.py,sha256=3voXmAVnGlXnOiwA3wcX4p0Lvmp0X1VKkQVPObJdwBY,4649
 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
@@ -19,16 +19,16 @@ nexaai/binds/__init__.py,sha256=T9Ua7SzHNglSeEqXlfH5ymYXRyXhNKkC9z_y_bWCNMo,80
 nexaai/binds/common_bind.cpython-310-darwin.so,sha256=FF5WuJj0fNCim_HjseBQu38vL-1M5zI_7EVTD7Bs-Bc,233960
 nexaai/binds/embedder_bind.cpython-310-darwin.so,sha256=mU6hP0SyH8vcmPpC2GIr7ioK7539dsg_YbmrBdmj7l0,202032
 nexaai/binds/libcrypto.dylib,sha256=ysW8ydmDPnnNRy3AHESjJwMTFfmGDKU9eLIaiR37ca0,5091432
-nexaai/binds/libnexa_bridge.dylib,sha256=8wjwefnWZLAzEqLlnFdjEWXNmTlRD9y9ogO0_ArRUB4,250712
+nexaai/binds/libnexa_bridge.dylib,sha256=Acbiop5I7SBo5MOOOfYCowkAHsnVcFAwR9xDpTnkwUs,250712
 nexaai/binds/libssl.dylib,sha256=JHPTSbRFnImmoWDO9rFdiKb0lJMT3q78VEsx-5-S0sk,889520
-nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=g4erKCUm2qdMZk1WUrr3IAXixRNp78ViUEkbE5jDOfE,182872
-nexaai/binds/nexa_llama_cpp/libggml-base.dylib,sha256=ChIX99NoLhsYVXJvv8iGMIpx-5Rst2gYwux-bEektB4,626992
-nexaai/binds/nexa_llama_cpp/libggml-cpu.so,sha256=tZcQGr6aWSQmTN12ieC2nIJ0lID5-mTkqoGjxJh07b4,1039744
-nexaai/binds/nexa_llama_cpp/libggml-metal.so,sha256=eDWuZ4ui8LsahlU05sNEMZ7lTtZfswKtcGcGvWTB0ro,713680
+nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=aYqMs5VhC07RNZZgyS9JeYJJgWCl-toZOmt6vXu5yp0,183008
+nexaai/binds/nexa_llama_cpp/libggml-base.dylib,sha256=oikz7Qxzx6A0mPROq7uHTUwWn66LvvOjcdVstG-M8Fw,629528
+nexaai/binds/nexa_llama_cpp/libggml-cpu.so,sha256=WepzOOeElmdOlsoMv7loLHsj8-Qx2O9ZJPlNnX11KJI,1039800
+nexaai/binds/nexa_llama_cpp/libggml-metal.so,sha256=ssn3Bqmnu7YA_FKL513Y18gbxG8WP9Udw71DNKV34eo,713680
 nexaai/binds/nexa_llama_cpp/libggml.dylib,sha256=Z2ZvkyEEpPtHhMYap-44p9Q0M6TXJbLcMy-smR2X5sk,58336
-nexaai/binds/nexa_llama_cpp/libllama.dylib,sha256=9pJFMHFlKHiQgLzi8YXextf5dPCYylQkpDv0EvCEssM,1958384
-nexaai/binds/nexa_llama_cpp/libmtmd.dylib,sha256=Etc0ZuYVNo9l1OTQRjZY4cTkgH2S2EL84DpxpWJeoJ4,682480
-nexaai/binds/nexa_llama_cpp/libnexa_plugin.dylib,sha256=GzhOq5vFIQAii3zh4oRN_TjcpqNLYEstsBiUAa96fUA,2589576
+nexaai/binds/nexa_llama_cpp/libllama.dylib,sha256=QZBn_w32g8NAJLE1unC_qx1BCVM531LeqTUqWipt9ks,1982280
+nexaai/binds/nexa_llama_cpp/libmtmd.dylib,sha256=F1QLNlfjiECRssUtEZeuqNqej-8COYcQjMZKPAB0CGk,701504
+nexaai/binds/nexa_llama_cpp/libnexa_plugin.dylib,sha256=csMdM6l21qpj-3_4z0xGsYM1snOBg4cJPfLXOQ8oTcI,2644752
 nexaai/cv_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/cv_impl/mlx_cv_impl.py,sha256=gKECQOv8iaWwG3bl7xeqVy2NN_9K7tYerIFzfn4eLo4,3228
 nexaai/cv_impl/pybind_cv_impl.py,sha256=uSmwBste4cT7c8DQmXzRLmzwDf773PAbXNYWW1UzVls,1064
@@ -39,8 +39,8 @@ nexaai/image_gen_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3h
 nexaai/image_gen_impl/mlx_image_gen_impl.py,sha256=BuDkksvXyb4J02GsdnbGAmYckfUU0Eah6BimoMD3QqY,11219
 nexaai/image_gen_impl/pybind_image_gen_impl.py,sha256=ms34VYoD5AxZFG6cIG0QAJDjCtfphaZ1bHzKzey1xF8,3692
 nexaai/llm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/llm_impl/mlx_llm_impl.py,sha256=2Ifc_mfTHDX64BWVHLjOhFCIMqM_Z-Cn4RfExlMtq0s,10865
-nexaai/llm_impl/pybind_llm_impl.py,sha256=DpO38rlGcvf0Zpe4bPKsbPD3EguBf0dDS9Ve64bgdvo,7653
+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=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
@@ -60,7 +60,7 @@ nexaai/mlx_backend/embedding/modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCe
 nexaai/mlx_backend/embedding/modeling/nexa_jina_v2.py,sha256=F9Z_9r-Dh0wNThiMp5W5hqE2dt5bf4ps5_c6h4BuWGw,15218
 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=YBLAdz_5gQ1VF9o98Tuj6xB_M2nUB9kX9VkM-Mp6ryc,29310
+nexaai/mlx_backend/llm/interface.py,sha256=SZFkuAUi2vxj_dSqj8RXf9vPTGMtpks_pZxxrF7iIe8,29330
 nexaai/mlx_backend/llm/main.py,sha256=gFDE4VZv_CLKMCTn0N521OfCKH_Ys26bHDh6g9VEFNc,1982
 nexaai/mlx_backend/mlx_audio/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/mlx_backend/mlx_audio/server.py,sha256=Pqy13Fafq4WX_cTuvRFz1jq89beQm2QQGpXmhK4b9jc,17547
@@ -361,9 +361,9 @@ nexaai/utils/decode.py,sha256=61n4Zf6c5QLyqGoctEitlI9BX3tPlP2a5aaKNHbw3T4,404
 nexaai/utils/model_manager.py,sha256=c07ocxxw1IHCQw6esbmYK0dX2R2OajfEIGsC_2teHXo,48572
 nexaai/utils/progress_tracker.py,sha256=76HlPkyN41IMHSsH56-qdlN_aY_oBfJz50J16Cx67R0,15102
 nexaai/vlm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nexaai/vlm_impl/mlx_vlm_impl.py,sha256=Dm-N38wqK3Cjdk3n7wfVGKC7hwxHvaM8pz37VzvJC-Y,10443
-nexaai/vlm_impl/pybind_vlm_impl.py,sha256=mvydHMHNWtkmyqouLIj1XSYZgsro3tcp3s_aqkjljE0,8510
-nexaai-1.0.4rc16.dist-info/METADATA,sha256=NuLsDWtJssKVjTNP4oo-tFItIBxIbiq-0hTq1rv706s,883
-nexaai-1.0.4rc16.dist-info/WHEEL,sha256=0KYp5feZ1CMUhsfFXKpSQTbSmQbXy4mv6yPPVBXg2EM,110
-nexaai-1.0.4rc16.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
-nexaai-1.0.4rc16.dist-info/RECORD,,
+nexaai/vlm_impl/mlx_vlm_impl.py,sha256=od1R1mRoIgPG3NHC7JiDlcB_YJY8aklX8Em3ZkeHNpE,10734
+nexaai/vlm_impl/pybind_vlm_impl.py,sha256=5ZMFgDATthmMzjrd-vE5KX5ZAMoWPYbF_FTLz8DBKIk,8908
+nexaai-1.0.6rc1.dist-info/METADATA,sha256=U2gJx8JlzG3wUYtVYk7VdDN7ildkHxWTQUE5Oya_Z_s,1154
+nexaai-1.0.6rc1.dist-info/WHEEL,sha256=0KYp5feZ1CMUhsfFXKpSQTbSmQbXy4mv6yPPVBXg2EM,110
+nexaai-1.0.6rc1.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
+nexaai-1.0.6rc1.dist-info/RECORD,,