nexaai 1.0.16rc11__cp310-cp310-macosx_14_0_universal2.whl → 1.0.17rc1__cp310-cp310-macosx_14_0_universal2.whl

nexaai/__init__.py

@@ -21,6 +21,9 @@ 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
 
@@ -45,6 +48,10 @@ __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-rc11" 
+__version__ = "1.0.17-rc1" 

nexaai/common.py

@@ -7,6 +7,7 @@ class PluginID(str, Enum):
     """Enum for plugin identifiers."""
     MLX = "mlx"
     LLAMA_CPP = "llama_cpp"
+    NEXAML = "nexaml"
 
 
 class ChatMessage(TypedDict):

nexaai/log.py

@@ -0,0 +1,92 @@
+"""
+Logging configuration for NexaAI bridge.
+
+This module provides a minimal API to configure bridge-wide logging
+to route into Python's logging system.
+"""
+
+import logging
+import threading
+from enum import IntEnum
+from typing import Optional
+
+from nexaai.binds import common_bind
+from nexaai.runtime import is_initialized
+
+
+class LogLevel(IntEnum):
+    """Log levels matching ml_LogLevel from ml.h"""
+    TRACE = 0
+    DEBUG = 1
+    INFO = 2
+    WARN = 3
+    ERROR = 4
+
+
+# Module-level state
+_config_lock = threading.Lock()
+_current_logger: Optional[logging.Logger] = None
+
+
+def set_logger(logger: Optional[logging.Logger] = None, *, strict: bool = True) -> None:
+    """
+    Set the process-wide bridge logger.
+    
+    Args:
+        logger: Python logger to receive bridge logs. If None, uses "nexaai.ml" logger.
+        strict: If True, raises if called after runtime initialization.
+                If False, attempts to set anyway (best-effort).
+    
+    Raises:
+        RuntimeError: If strict=True and runtime is already initialized.
+    """
+    global _current_logger
+    
+    with _config_lock:
+        # Check initialization state if strict mode
+        if strict and is_initialized():
+            raise RuntimeError(
+                "Cannot configure logging after runtime initialization. "
+                "Call set_logger() before creating any models, or use strict=False for best-effort."
+            )
+        
+        # Use default logger if none provided
+        if logger is None:
+            logger = logging.getLogger("nexaai.ml")
+        
+        _current_logger = logger
+        
+        # Set the C callback
+        common_bind.ml_set_log(_log_callback)
+
+
+def _log_callback(level: int, message: str) -> None:
+    """Internal callback that forwards bridge logs to Python logger."""
+    if _current_logger is None:
+        return
+    
+    # Map bridge log levels to Python logging levels
+    if level == LogLevel.TRACE or level == LogLevel.DEBUG:
+        _current_logger.debug(message)
+    elif level == LogLevel.INFO:
+        _current_logger.info(message)
+    elif level == LogLevel.WARN:
+        _current_logger.warning(message)
+    elif level == LogLevel.ERROR:
+        _current_logger.error(message)
+    else:
+        # Fallback for unknown levels
+        _current_logger.info(f"[Level {level}] {message}")
+
+
+def get_error_message(error_code: int) -> str:
+    """
+    Get error message string for error code.
+    
+    Args:
+        error_code: ML error code (typically negative)
+        
+    Returns:
+        Human-readable error message
+    """
+    return common_bind.ml_get_error_message(error_code)

nexaai/mlx_backend/vlm/interface.py

@@ -80,6 +80,9 @@ class VLM(ProfilingMixin):
 
         # Init deafutl sampler config with defualt.
         self.sampler_config = SamplerConfig()
+        
+        # Track global character position for incremental processing
+        self.global_n_past_chars = 0
 
     def destroy(self) -> None:
         """Destroy the model and free resources."""
@@ -89,6 +92,7 @@ class VLM(ProfilingMixin):
     def reset(self) -> None:
         """Reset the model state."""
         self._reset_cache()
+        self.global_n_past_chars = 0
 
     def _reset_cache(self) -> None:
         """Reset the KV cache."""
@@ -141,6 +145,16 @@ class VLM(ProfilingMixin):
         image_list = [str(path) for path in image_paths] if image_paths else None
         audio_list = [str(path) for path in audio_paths] if audio_paths else None
 
+        # Extract incremental portion of the prompt (similar to llama.cpp VLM)
+        full_prompt_len = len(prompt)
+        incremental_prompt = prompt
+        
+        if self.global_n_past_chars < full_prompt_len:
+            incremental_prompt = prompt[self.global_n_past_chars:]
+        else:
+            # No new text to process
+            incremental_prompt = ""
+
         # End prompt processing, start decode
         self._prompt_end()
         self._decode_start()
@@ -152,7 +166,7 @@ class VLM(ProfilingMixin):
             text, stats = generate(
                 self.model,
                 self.processor,
-                prompt,
+                incremental_prompt,  # Use incremental prompt instead of full prompt
                 image=image_list,
                 audio=audio_list,
                 **gen_kwargs,
@@ -181,6 +195,10 @@ class VLM(ProfilingMixin):
             self._update_prompt_tokens(prompt_tokens)
             self._update_generated_tokens(generated_tokens)
             self._set_stop_reason(StopReason.ML_STOP_REASON_COMPLETED)
+            
+            # Update global character position
+            self.global_n_past_chars = full_prompt_len + len(text)
+            
             self._decode_end()
             self._end_profiling()
 
@@ -226,6 +244,16 @@ class VLM(ProfilingMixin):
         image_list = [str(path) for path in image_paths] if image_paths else None
         audio_list = [str(path) for path in audio_paths] if audio_paths else None
 
+        # Extract incremental portion of the prompt (similar to llama.cpp VLM)
+        full_prompt_len = len(prompt)
+        incremental_prompt = prompt
+        
+        if self.global_n_past_chars < full_prompt_len:
+            incremental_prompt = prompt[self.global_n_past_chars:]
+        else:
+            # No new text to process
+            incremental_prompt = ""
+
         # End prompt processing, start decode
         self._prompt_end()
         self._decode_start()
@@ -239,7 +267,7 @@ class VLM(ProfilingMixin):
             for result in stream_generate_impl(
                 self.model,
                 self.processor,
-                prompt,
+                incremental_prompt,  # Use incremental prompt instead of full prompt
                 image=image_list,
                 audio=audio_list,
                 **gen_kwargs,
@@ -266,6 +294,9 @@ class VLM(ProfilingMixin):
                 self._update_prompt_tokens(last_result.prompt_tokens)
                 self._update_generated_tokens(last_result.generation_tokens)
 
+            # Update global character position
+            self.global_n_past_chars = full_prompt_len + len(text)
+
             self._decode_end()
             self._end_profiling()
 

nexaai/runtime.py

@@ -28,6 +28,10 @@ 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/utils/quantization_utils.py

@@ -25,12 +25,15 @@ class QuantizationType(str, Enum):
     F16 = "F16"
     Q2_K = "Q2_K"
     Q2_K_L = "Q2_K_L"
+    Q3_K = "Q3_K"
     Q3_K_M = "Q3_K_M"
     Q3_K_S = "Q3_K_S"
     Q4_0 = "Q4_0"
     Q4_1 = "Q4_1"
+    Q4_K = "Q4_K"
     Q4_K_M = "Q4_K_M"
     Q4_K_S = "Q4_K_S"
+    Q5_K = "Q5_K"
     Q5_K_M = "Q5_K_M"
     Q5_K_S = "Q5_K_S"
     Q6_K = "Q6_K"
@@ -67,12 +70,15 @@ def extract_quantization_from_filename(filename: str) -> Optional[QuantizationTy
         'f16.': QuantizationType.F16,  # Add F16 support
         'q2_k_l.': QuantizationType.Q2_K_L,  # Check Q2_K_L before Q2_K to avoid partial match
         'q2_k.': QuantizationType.Q2_K,
+        'q3_k.': QuantizationType.Q3_K,
         'q3_k_m.': QuantizationType.Q3_K_M,
-        'q3_ks.': QuantizationType.Q3_K_S,
+        'q3_k_s.': QuantizationType.Q3_K_S,
         'q4_k_m.': QuantizationType.Q4_K_M,
         'q4_k_s.': QuantizationType.Q4_K_S,
         'q4_0.': QuantizationType.Q4_0,
         'q4_1.': QuantizationType.Q4_1,
+        'q4_k.': QuantizationType.Q4_K,
+        'q5_k.': QuantizationType.Q5_K,
         'q5_k_m.': QuantizationType.Q5_K_M,
         'q5_k_s.': QuantizationType.Q5_K_S,
         'q6_k.': QuantizationType.Q6_K,

{nexaai-1.0.16rc11.dist-info → nexaai-1.0.17rc1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexaai
-Version: 1.0.16rc11
+Version: 1.0.17rc1
 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-1.0.16rc11.dist-info → nexaai-1.0.17rc1.dist-info}/RECORD

@@ -1,25 +1,26 @@
-nexaai/__init__.py,sha256=jXdC4vv6DBK1fVewYTYSUhOOYfvf_Mk81UIeMGGIKUg,2029
-nexaai/_stub.cpython-310-darwin.so,sha256=6-Cum2FTCp9SJUfkSFZji8GUnxR_835-V6_VYrU3lmw,66768
-nexaai/_version.py,sha256=f7rjwo4Z-jPnRW2BO8DA1Z8FCkrghe6e-uQJVLsIGLE,144
+nexaai/__init__.py,sha256=L8oB7GFZZMGnUpCg0PecDbI_ycKuQak-ZEJ4Y12_QIw,2184
+nexaai/_stub.cpython-310-darwin.so,sha256=-U41Rg1210R-O8tEXhmXA-PPJ5tIk48yx-v6HoJD660,66768
+nexaai/_version.py,sha256=rKQTMRjwBe37A1MYCov01rf2P607gRZoI-XbYCSIcoA,143
 nexaai/asr.py,sha256=NljMXDErwPNMOPaRkJZMEDka9Nk8xyur7L8i924TStY,2054
 nexaai/base.py,sha256=N8PRgDFA-XPku2vWnQIofQ7ipz3pPlO6f8YZGnuhquE,982
-nexaai/common.py,sha256=yBnIbqYaQYnfrl7IczOBh6MDibYZVxwaRJEglYcKgGs,3422
+nexaai/common.py,sha256=Y0NJNLTi4Nq4x1WL6PQsSvGUto0eGmWhjpsC6jcekfA,3444
 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=mxxHYsb5iBUAm2K_u-XJWr_U-spJ9S4eApc8kf9myjw,1957
+nexaai/runtime.py,sha256=JvllhlNPgYGLbgGyX2yNvmGzT0lZ5XbvTvEo8sZG_Ho,2067
 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=hVxY76tn7hN6uHDIgM7LWNvgoudHgNZVoaygM9X1RWE,217232
+nexaai/binds/common_bind.cpython-310-darwin.so,sha256=ya8gz7NnUtHlniGnRyExdwWfdHimEHiJ7Dry4I7_y44,235264
 nexaai/binds/embedder_bind.cpython-310-darwin.so,sha256=tPa0c0Dv_GiW66fgmAGWGCHXRGNApznqoQS0eQx9GFM,202064
-nexaai/binds/libnexa_bridge.dylib,sha256=Tr5qwX_FgsEBf8xbqtvA1wEj6eChdQLBuzQOChMWydk,251144
-nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=6vRO5eOsL6s-hCKdz6onIjBKR_wcpCIsXHtUJY9vJBI,182704
+nexaai/binds/libnexa_bridge.dylib,sha256=4VzUccl3U2lTlNEK7-Q1szFpdE7HKRiPmJwQ5Y-VgqM,251192
+nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=TAWfa1Hzq00TjtC1xVsiAeLp6hv2LrL5afDz4omUghc,182784
 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
@@ -27,7 +28,7 @@ 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=4rTxwHGSVUfGnvA6laPpLztMMFooy-inG2798HXqKPY,1862104
+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_mlx/py-lib/profiling.py,sha256=Dc-mybFwBdCIKFWL7CbSHjkOJGAoYHG7r_e_XPhzwBU,9361
@@ -181,6 +182,16 @@ nexaai/binds/nexa_mlx/py-lib/mlx_audio/tts/tests/test_base.py,sha256=dAeEA0IsBnU
 nexaai/binds/nexa_mlx/py-lib/mlx_audio/tts/tests/test_convert.py,sha256=79ddUhtTUlElD9NvBupUxl-MV4HKFM9PUxQpTwKLnBA,5804
 nexaai/binds/nexa_mlx/py-lib/mlx_audio/tts/tests/test_interpolate.py,sha256=9dNmH03C46HtxwesH2DpT2oTNEG1KCZWYEKq6UQ3vfk,3536
 nexaai/binds/nexa_mlx/py-lib/mlx_audio/tts/tests/test_models.py,sha256=12RiOfPtSZQj5g5JM-yCJk3uGQfM3OdmRiPt5uUDE4E,35096
+nexaai/binds/nexa_nexaml/libggml-base.dylib,sha256=JM4oOkie1su0ES5hMdtILeQHlRukRzH1vTleTupUXhg,650736
+nexaai/binds/nexa_nexaml/libggml-cpu.so,sha256=qiYxbTe4Nt7n36zJVvq3zovgSZEmrN2is6gzTern7UI,677728
+nexaai/binds/nexa_nexaml/libggml-metal.so,sha256=zfaX7rIBYQazH2lf-vza007BMhPTK1ASd2T0HLLIA4E,673104
+nexaai/binds/nexa_nexaml/libggml.dylib,sha256=aOTj_6RrAMkfDO0ZI28_3nfcC-l4Y3dRCiS3C0d0_eI,58592
+nexaai/binds/nexa_nexaml/libnexa-mm-process.dylib,sha256=3mITty3oYhoi5yc690CxS0tf4LGanqrPCR0meE-h9HQ,8568120
+nexaai/binds/nexa_nexaml/libnexa-sampling.dylib,sha256=OliTBr-r2y22Ebjc1Y33Ax56XpLMFQKtsHgfouj6TQM,7865400
+nexaai/binds/nexa_nexaml/libnexa_plugin.dylib,sha256=rkTrXOKH71m0SmUP5IBLsJrLTbCElcWX3GKn33BPZpY,224056
+nexaai/binds/nexa_nexaml/libnexaproc.dylib,sha256=Pnxgm2k29hfedLjNP4t44gvioh-NWbE_3BW-VsCdHp8,880560
+nexaai/binds/nexa_nexaml/libqwen3-vl.dylib,sha256=hd8VPINYTH8mbU3BPxnjrPOKTw-NkWn7zkDB7p074-U,753168
+nexaai/binds/nexa_nexaml/libqwen3vl-vision.dylib,sha256=AmUk-DBtBAp5DOR14ICZcsiAcKjxxAx1y6mHAaETQu8,570024
 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
@@ -399,7 +410,7 @@ nexaai/mlx_backend/tts/interface.py,sha256=0FvZbIyOvg8jERZEQ6bygbv7v02O9xHO4-TPU
 nexaai/mlx_backend/vlm/__init__.py,sha256=_25kvMEviX16Hg3bro8Ws70V0eeIEqYKV8ZDXqYzKew,73
 nexaai/mlx_backend/vlm/generate.py,sha256=DqHFEAuqk-nko8ho6U9GAXTDAWz4d8GTe_hCt-XFyCw,19071
 nexaai/mlx_backend/vlm/generate_qwen3_vl.py,sha256=undjso1mfxqpd6FMTksSA5qagRttxAGbOBj1x7cqI1s,9211
-nexaai/mlx_backend/vlm/interface.py,sha256=vFTzJCbqq55ybv_tbDBC9NVn1_sXgCfqXdsV-3ia8vo,16177
+nexaai/mlx_backend/vlm/interface.py,sha256=0BLfodbYOU71jFvAvv01FuLBE_KBtyB-8Cd7LqzzRHY,17450
 nexaai/mlx_backend/vlm/main.py,sha256=nPcg25jupeDD74uvRoxpWp3Dsulw7WddI7vll6zejak,10664
 nexaai/mlx_backend/vlm/modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nexaai/mlx_backend/vlm/modeling/convert.py,sha256=ia5i9cgTufFGmKyhkYUaW0nfNqT_bMo8i-Hg_zy5JC4,1863
@@ -536,11 +547,11 @@ nexaai/utils/manifest_utils.py,sha256=sR9Nme4GbD3Cb3fMd55yLvGZpqxb71vd6b2XZTsrIG
 nexaai/utils/model_manager.py,sha256=p2kJKK63Zk-rEUucFsgY0T5PyXi_IvJY0gKewUVcAV4,56081
 nexaai/utils/model_types.py,sha256=-DER8L4lAUR_iLS99F0r57avwqWtuN21ug5pX2p24_E,1369
 nexaai/utils/progress_tracker.py,sha256=jdUqtmPqyhwC9uSKvQcJEYETwSt-OhP4oitdJ94614o,15394
-nexaai/utils/quantization_utils.py,sha256=4gvp6UQfSO9G1FYBwnFtQspTzH9sDbi1PBXw2t1N69M,7650
+nexaai/utils/quantization_utils.py,sha256=FYcNSAKGlBqFDUTx3jSKOr2lnq4nyiyC0ZG8oSxFwiU,7825
 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-1.0.16rc11.dist-info/METADATA,sha256=ZBygPzSF5yfMU8SgBMTS8HmrlM4Rx_vAwQsoX8rZIks,1202
-nexaai-1.0.16rc11.dist-info/WHEEL,sha256=T2p57lol9__xkoU6aJTyN1Pm43ZpRU3q6km7mIbrAMs,114
-nexaai-1.0.16rc11.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
-nexaai-1.0.16rc11.dist-info/RECORD,,
+nexaai-1.0.17rc1.dist-info/METADATA,sha256=-o225-FK1Adyvf4f8tzZM7FMyeq9g9CJthuqKuNKvG4,1201
+nexaai-1.0.17rc1.dist-info/WHEEL,sha256=T2p57lol9__xkoU6aJTyN1Pm43ZpRU3q6km7mIbrAMs,114
+nexaai-1.0.17rc1.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
+nexaai-1.0.17rc1.dist-info/RECORD,,