lollms-client 0.15.2__py3-none-any.whl → 0.17.0__py3-none-any.whl

lollms_client/stt_bindings/whispercpp/__init__.py

@@ -0,0 +1,380 @@
+# lollms_client/stt_bindings/whispercpp/__init__.py
+import os
+import shutil
+import subprocess
+import tempfile
+from pathlib import Path
+from typing import Optional, List, Union, Dict, Any
+
+from ascii_colors import trace_exception, ASCIIColors
+# No pipmaster needed here as whisper.cpp is a C++ executable.
+# Python dependencies are assumed to be handled by the environment or a higher level.
+
+from lollms_client.lollms_stt_binding import LollmsSTTBinding
+
+BindingName = "WhisperCppSTTBinding"
+
+DEFAULT_WHISPERCPP_EXE_NAMES = ["main", "whisper-cli", "whisper"] # Common names for the executable
+
+class WhisperCppSTTBinding(LollmsSTTBinding):
+    def __init__(self,
+                 model_path: Union[str, Path], # Path to the GGUF Whisper model
+                 whispercpp_exe_path: Optional[Union[str, Path]] = None, # Path to whisper.cpp executable
+                 ffmpeg_path: Optional[Union[str, Path]] = None, # Path to ffmpeg executable (if not in PATH)
+                 models_search_path: Optional[Union[str, Path]] = None, # Optional dir to scan for more models
+                 default_language: str = "auto",
+                 n_threads: int = 4,
+                 # Catch LollmsSTTBinding standard args even if not directly used by this local binding
+                 host_address: Optional[str] = None, # Not used for local binding
+                 service_key: Optional[str] = None, # Not used for local binding
+                 verify_ssl_certificate: bool = True, # Not used for local binding
+                 **kwargs): # Catch-all for future compatibility or specific whisper.cpp params
+
+        super().__init__(binding_name="whispercpp") 
+        
+        # --- Validate FFMPEG ---
+        self.ffmpeg_exe = None
+        if ffmpeg_path:
+            resolved_ffmpeg_path = Path(ffmpeg_path)
+            if resolved_ffmpeg_path.is_file() and os.access(resolved_ffmpeg_path, os.X_OK):
+                self.ffmpeg_exe = str(resolved_ffmpeg_path)
+            else:
+                raise FileNotFoundError(f"Provided ffmpeg_path '{ffmpeg_path}' not found or not executable.")
+        else:
+            self.ffmpeg_exe = shutil.which("ffmpeg")
+        
+        if not self.ffmpeg_exe:
+            ASCIIColors.warning("ffmpeg not found in PATH or explicitly provided. Audio conversion will not be possible for non-WAV files or incompatible WAV files.")
+            ASCIIColors.warning("Please install ffmpeg and ensure it's in your system's PATH, or provide ffmpeg_path argument.")
+            # Not raising an error here, as user might provide perfectly formatted WAV files.
+
+        # --- Validate Whisper.cpp Executable ---
+        self.whispercpp_exe = None
+        if whispercpp_exe_path:
+            resolved_wcpp_path = Path(whispercpp_exe_path)
+            if resolved_wcpp_path.is_file() and os.access(resolved_wcpp_path, os.X_OK):
+                self.whispercpp_exe = str(resolved_wcpp_path)
+            else:
+                raise FileNotFoundError(f"Provided whispercpp_exe_path '{whispercpp_exe_path}' not found or not executable.")
+        else:
+            for name in DEFAULT_WHISPERCPP_EXE_NAMES:
+                found_path = shutil.which(name)
+                if found_path:
+                    self.whispercpp_exe = found_path
+                    ASCIIColors.info(f"Found whisper.cpp executable via PATH: {self.whispercpp_exe}")
+                    break
+        
+        if not self.whispercpp_exe:
+            raise FileNotFoundError(
+                f"Whisper.cpp executable (tried: {', '.join(DEFAULT_WHISPERCPP_EXE_NAMES)}) not found in PATH or explicitly provided. "
+                "Please build/install whisper.cpp (from https://github.com/ggerganov/whisper.cpp) "
+                "and ensure its main executable is in your system's PATH or provide its path via whispercpp_exe_path argument."
+            )
+
+        # --- Validate Model Path ---
+        self.model_path = Path(model_path)
+        if not self.model_path.is_file():
+            # Try to resolve relative to models_search_path if provided and model_path is not absolute
+            if models_search_path and not self.model_path.is_absolute() and Path(models_search_path, self.model_path).is_file():
+                self.model_path = Path(models_search_path, self.model_path).resolve()
+            else:
+                raise FileNotFoundError(f"Whisper GGUF model file not found at '{self.model_path}'. Also checked in models_search_path if applicable.")
+        
+        self.models_search_path = Path(models_search_path).resolve() if models_search_path else None
+        self.default_language = default_language
+        self.n_threads = n_threads
+        self.extra_whisper_args = kwargs.get("extra_whisper_args", []) # e.g. ["--no-timestamps"]
+
+        ASCIIColors.green(f"WhisperCppSTTBinding initialized with model: {self.model_path}")
+
+    def _convert_to_wav(self, input_audio_path: Path, output_wav_path: Path) -> bool:
+        if not self.ffmpeg_exe:
+            ASCIIColors.error("ffmpeg is required for audio conversion but was not found or configured.")
+            return False
+        try:
+            command = [
+                self.ffmpeg_exe,
+                "-i", str(input_audio_path),
+                "-ar", "16000",          # 16kHz sample rate
+                "-ac", "1",              # Mono channel
+                "-c:a", "pcm_s16le",     # Signed 16-bit PCM little-endian
+                "-y",                    # Overwrite output file if it exists
+                str(output_wav_path)
+            ]
+            ASCIIColors.info(f"Converting audio with ffmpeg: {' '.join(command)}")
+            # Using Popen to better handle stderr/stdout if needed for detailed logging
+            process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
+            stdout, stderr = process.communicate()
+            
+            if process.returncode != 0:
+                ASCIIColors.error(f"ffmpeg conversion failed (exit code {process.returncode}).")
+                ASCIIColors.error(f"ffmpeg stdout:\n{stdout}")
+                ASCIIColors.error(f"ffmpeg stderr:\n{stderr}")
+                return False
+            ASCIIColors.info(f"ffmpeg conversion successful: {output_wav_path}")
+            return True
+        except FileNotFoundError: # Handle case where ffmpeg command itself is not found
+            ASCIIColors.error(f"ffmpeg command '{self.ffmpeg_exe}' not found. Ensure ffmpeg is installed and in PATH or ffmpeg_path is correct.")
+            return False
+        except Exception as e:
+            ASCIIColors.error(f"An error occurred during ffmpeg conversion: {e}")
+            trace_exception(e)
+            return False
+
+    def transcribe_audio(self, audio_path: Union[str, Path], model: Optional[str] = None, **kwargs) -> str:
+        input_audio_p = Path(audio_path)
+        if not input_audio_p.exists():
+            raise FileNotFoundError(f"Input audio file not found: {input_audio_p}")
+
+        current_model_path = self.model_path
+        if model: # User specified a different model for this transcription
+            potential_model_p = Path(model)
+            if potential_model_p.is_absolute() and potential_model_p.is_file():
+                current_model_path = potential_model_p
+            elif self.models_search_path and (self.models_search_path / model).is_file():
+                current_model_path = self.models_search_path / model
+            elif Path(model).is_file(): # Relative to current working directory?
+                 current_model_path = Path(model)
+            else:
+                ASCIIColors.warning(f"Specified model '{model}' not found as absolute path, in models_search_path, or current dir. Using default: {self.model_path.name}")
+        
+        language = kwargs.get("language", self.default_language)
+        threads = kwargs.get("n_threads", self.n_threads)
+        extra_args_call = kwargs.get("extra_whisper_args", self.extra_whisper_args)
+
+        with tempfile.TemporaryDirectory(prefix="lollms_whispercpp_") as tmpdir:
+            tmp_dir_path = Path(tmpdir)
+            
+            # Always convert to ensure 16kHz mono WAV, unless explicitly told not to by a kwarg (e.g. assume_wav=True)
+            force_conversion = not kwargs.get("assume_compatible_wav", False)
+            
+            if force_conversion or input_audio_p.suffix.lower() != ".wav":
+                if not self.ffmpeg_exe:
+                    raise RuntimeError("ffmpeg is required for audio pre-processing but is not configured. "
+                                       "Please provide a 16kHz mono WAV file or configure ffmpeg.")
+                converted_wav_path = tmp_dir_path / (input_audio_p.stem + "_16khz_mono.wav")
+                if not self._convert_to_wav(input_audio_p, converted_wav_path):
+                    raise Exception(f"Audio conversion to compatible WAV failed for {input_audio_p}.")
+                target_audio_file = converted_wav_path
+            else: # Input is WAV, assume it's compatible (user's responsibility if assume_compatible_wav=True)
+                target_audio_file = input_audio_p
+            
+            command = [
+                self.whispercpp_exe,
+                "-m", str(current_model_path),
+                "-f", str(target_audio_file),
+                "-l", language,
+                "-t", str(threads),
+                "-otxt" # Output as a .txt file in the same dir as input wav
+            ]
+            if isinstance(extra_args_call, list):
+                command.extend(extra_args_call)
+            elif isinstance(extra_args_call, str):
+                command.extend(extra_args_call.split())
+
+            ASCIIColors.info(f"Executing Whisper.cpp: {' '.join(command)}")
+            try:
+                # Run whisper.cpp, making it output its .txt file into our temp directory.
+                # To do this, we can copy the target_audio_file into tmp_dir_path if it's not already there,
+                # then run whisper.cpp with CWD as tmp_dir_path.
+                
+                final_target_audio_in_tmp: Path
+                if target_audio_file.parent != tmp_dir_path:
+                    final_target_audio_in_tmp = tmp_dir_path / target_audio_file.name
+                    shutil.copy2(target_audio_file, final_target_audio_in_tmp)
+                    # Update command to use the path within tmp_dir_path if we copied it.
+                    # The -f argument should be relative to the CWD if CWD is set.
+                    command[command.index("-f")+1] = str(final_target_audio_in_tmp.name) 
+                else:
+                    final_target_audio_in_tmp = target_audio_file
+                    command[command.index("-f")+1] = str(final_target_audio_in_tmp.name) 
+
+
+                process = subprocess.run(command, capture_output=True, text=True, check=True, cwd=str(tmp_dir_path))
+                
+                output_txt_file = tmp_dir_path / (final_target_audio_in_tmp.name + ".txt")
+
+                if output_txt_file.exists():
+                    transcribed_text = output_txt_file.read_text(encoding='utf-8').strip()
+                    ASCIIColors.green(f"Whisper.cpp transcription successful for {input_audio_p.name}.")
+                    return transcribed_text
+                else:
+                    ASCIIColors.error(f"Whisper.cpp did not produce the expected output file: {output_txt_file.name} in {tmp_dir_path}")
+                    ASCIIColors.info(f"Whisper.cpp stdout:\n{process.stdout}")
+                    ASCIIColors.info(f"Whisper.cpp stderr:\n{process.stderr}")
+                    raise Exception("Whisper.cpp execution failed to produce output text file.")
+
+            except subprocess.CalledProcessError as e:
+                ASCIIColors.error(f"Whisper.cpp execution failed with exit code {e.returncode} for {input_audio_p.name}")
+                ASCIIColors.error(f"Command: {' '.join(e.cmd)}")
+                ASCIIColors.error(f"Stdout:\n{e.stdout}")
+                ASCIIColors.error(f"Stderr:\n{e.stderr}")
+                trace_exception(e)
+                raise Exception(f"Whisper.cpp execution error: {e.stderr or e.stdout or 'Unknown whisper.cpp error'}") from e
+            except Exception as e:
+                ASCIIColors.error(f"An error occurred during Whisper.cpp transcription for {input_audio_p.name}: {e}")
+                trace_exception(e)
+                raise
+
+    def list_models(self, **kwargs) -> List[str]:
+        models = []
+        # 1. Add the default configured model's name
+        if self.model_path and self.model_path.exists():
+            # For consistency, list by name. The 'model' arg in transcribe_audio can take this name.
+            models.append(self.model_path.name)
+
+        # 2. Scan models_search_path if provided
+        if self.models_search_path and self.models_search_path.is_dir():
+            for item in self.models_search_path.iterdir():
+                if item.is_file() and item.suffix.lower() == ".gguf":
+                    # Add name if not already listed (default model might be in search path)
+                    if item.name not in models:
+                        models.append(item.name)
+        
+        return sorted(list(set(models))) # Ensure uniqueness and sort
+
+
+# --- Main Test Block ---
+if __name__ == '__main__':
+    ASCIIColors.yellow("--- WhisperCppSTTBinding Test ---")
+
+    # --- USER CONFIGURATION REQUIRED FOR TEST ---
+    # Find your whisper.cpp build directory and the 'main' or 'whisper-cli' executable.
+    # Example: If you built whisper.cpp in /home/user/whisper.cpp, the exe might be /home/user/whisper.cpp/main
+    TEST_WHISPERCPP_EXE = None # SET THIS: e.g., "/path/to/whisper.cpp/main" or "whisper-cli" if in PATH
+    
+    # Download a GGUF model from Hugging Face: https://huggingface.co/ggerganov/whisper.cpp/tree/main
+    # Place it somewhere accessible.
+    TEST_MODEL_GGUF = "ggml-tiny.en.bin" # SET THIS: e.g., "/path/to/models/ggml-tiny.en.bin"
+                                        # If just a name, expects it in CWD or models_search_path.
+
+    # Optional: Path to ffmpeg if not in system PATH
+    TEST_FFMPEG_EXE = None # e.g., "/usr/local/bin/ffmpeg"
+
+    # Optional: A directory to put other .gguf models for testing list_models
+    TEST_MODELS_SEARCH_DIR = Path("./test_whisper_models_cpp") 
+    # --- END USER CONFIGURATION ---
+
+
+    # Create a dummy audio file for testing (requires scipy and numpy)
+    dummy_audio_file = Path("dummy_whispercpp_test.wav")
+    if not dummy_audio_file.exists():
+        try:
+            import numpy as np
+            from scipy.io.wavfile import write as write_wav
+            samplerate = 44100; duration = 1.5; frequency = 330 # A bit longer, different freq
+            t = np.linspace(0., duration, int(samplerate * duration), endpoint=False)
+            amplitude = np.iinfo(np.int16).max * 0.3
+            data = amplitude * np.sin(2. * np.pi * frequency * t)
+            # Add some noise
+            data += (amplitude * 0.05 * np.random.normal(size=data.shape[0])).astype(data.dtype)
+            write_wav(dummy_audio_file, samplerate, data.astype(np.int16))
+            ASCIIColors.green(f"Created dummy audio file for testing: {dummy_audio_file}")
+        except ImportError:
+            ASCIIColors.warning("SciPy/NumPy not installed. Cannot create dummy audio file for testing.")
+            ASCIIColors.warning(f"Please manually place a test audio file named '{dummy_audio_file.name}' in the current directory.")
+        except Exception as e_dummy:
+            ASCIIColors.error(f"Could not create dummy audio file: {e_dummy}")
+    
+    # Prepare model search directory for list_models test
+    if TEST_MODELS_SEARCH_DIR:
+        TEST_MODELS_SEARCH_DIR.mkdir(exist_ok=True)
+        # Create another dummy GGUF (just an empty file for listing purposes)
+        (TEST_MODELS_SEARCH_DIR / "ggml-base.en.bin").touch(exist_ok=True) 
+
+
+    # Basic check for prerequisites before attempting to initialize
+    if TEST_WHISPERCPP_EXE is None and not any(shutil.which(name) for name in DEFAULT_WHISPERCPP_EXE_NAMES):
+        ASCIIColors.error(f"Whisper.cpp executable not found in PATH and TEST_WHISPERCPP_EXE not set. Aborting test.")
+        exit(1)
+    
+    if not Path(TEST_MODEL_GGUF).exists() and not (TEST_MODELS_SEARCH_DIR and (TEST_MODELS_SEARCH_DIR / TEST_MODEL_GGUF).exists()):
+         # Check if model is just a name and exists in current dir if TEST_MODELS_SEARCH_DIR is not set or model not in it
+        if not (Path().cwd()/TEST_MODEL_GGUF).exists():
+            ASCIIColors.error(f"Test model GGUF '{TEST_MODEL_GGUF}' not found. Please download/place it or update TEST_MODEL_GGUF path. Aborting test.")
+            exit(1)
+        else: # Found in CWD
+            TEST_MODEL_GGUF = str((Path().cwd()/TEST_MODEL_GGUF).resolve())
+
+
+    stt_binding = None
+    try:
+        ASCIIColors.cyan("\n--- Initializing WhisperCppSTTBinding ---")
+        stt_binding = WhisperCppSTTBinding(
+            model_path=TEST_MODEL_GGUF,
+            whispercpp_exe_path=TEST_WHISPERCPP_EXE,
+            ffmpeg_path=TEST_FFMPEG_EXE,
+            models_search_path=TEST_MODELS_SEARCH_DIR,
+            default_language="en",
+            n_threads=os.cpu_count() // 2 or 1, # Use half CPU cores or at least 1
+        )
+        ASCIIColors.green("Binding initialized successfully.")
+
+        ASCIIColors.cyan("\n--- Listing available models ---")
+        models = stt_binding.list_models()
+        if models:
+            print(f"Available models: {models}")
+        else:
+            ASCIIColors.warning("No models listed. Check paths and models_search_path.")
+
+        if dummy_audio_file.exists():
+            ASCIIColors.cyan(f"\n--- Transcribing '{dummy_audio_file.name}' (expected 16kHz mono after conversion) ---")
+            transcription = stt_binding.transcribe_audio(str(dummy_audio_file))
+            print(f"Transcription: '{transcription}'")
+
+            # Test with a different model if listed and available
+            if "ggml-base.en.bin" in models and "ggml-base.en.bin" != Path(TEST_MODEL_GGUF).name :
+                if (TEST_MODELS_SEARCH_DIR / "ggml-base.en.bin").exists(): # Ensure it's actually there
+                    ASCIIColors.cyan(f"\n--- Transcribing with model 'ggml-base.en.bin' from search path ---")
+                    transcription_base = stt_binding.transcribe_audio(str(dummy_audio_file), model="ggml-base.en.bin")
+                    print(f"Transcription (ggml-base.en.bin): '{transcription_base}'")
+                else:
+                    ASCIIColors.warning("Model 'ggml-base.en.bin' listed but not found in search path for test.")
+            
+            # Test assume_compatible_wav (if user has a 16kHz mono WAV already)
+            # Create a specific 16kHz mono wav file for this
+            compatible_wav_file = Path("compatible_test.wav")
+            try:
+                import numpy as np
+                from scipy.io.wavfile import write as write_wav
+                samplerate = 16000; duration = 1.0; frequency = 550
+                t_compat = np.linspace(0., duration, int(samplerate * duration), endpoint=False)
+                data_compat = (np.iinfo(np.int16).max * 0.2 * np.sin(2. * np.pi * frequency * t_compat)).astype(np.int16)
+                write_wav(compatible_wav_file, samplerate, data_compat)
+                ASCIIColors.green(f"Created compatible 16kHz mono WAV: {compatible_wav_file}")
+
+                ASCIIColors.cyan(f"\n--- Transcribing '{compatible_wav_file.name}' with assume_compatible_wav=True ---")
+                transcription_compat = stt_binding.transcribe_audio(str(compatible_wav_file), assume_compatible_wav=True)
+                print(f"Transcription (compatible WAV): '{transcription_compat}'")
+
+            except ImportError: ASCIIColors.warning("SciPy/NumPy not available, skipping compatible WAV test.")
+            except Exception as e_compat: ASCIIColors.error(f"Error in compatible WAV test: {e_compat}")
+            finally:
+                if compatible_wav_file.exists(): compatible_wav_file.unlink(missing_ok=True)
+
+
+        else:
+            ASCIIColors.warning(f"Dummy audio file '{dummy_audio_file}' not found. Skipping main transcription test.")
+
+    except FileNotFoundError as e:
+        ASCIIColors.error(f"Initialization or transcription failed due to FileNotFoundError: {e}")
+        ASCIIColors.info("Please ensure whisper.cpp/ffmpeg executables are in PATH or paths are correctly set in the test script, and the GGUF model file exists.")
+    except RuntimeError as e:
+        ASCIIColors.error(f"Runtime error: {e}")
+    except Exception as e:
+        ASCIIColors.error(f"An unexpected error occurred: {e}")
+        trace_exception(e)
+    finally:
+        # Clean up dummy files created by this test script
+        if "samplerate" in locals() and dummy_audio_file.exists(): # Heuristic: if we created it
+             dummy_audio_file.unlink(missing_ok=True)
+        if TEST_MODELS_SEARCH_DIR:
+            if (TEST_MODELS_SEARCH_DIR / "ggml-base.en.bin").exists():
+                 (TEST_MODELS_SEARCH_DIR / "ggml-base.en.bin").unlink(missing_ok=True)
+            # Remove dir only if it's empty (or was created by this script and now empty)
+            try:
+                if not any(TEST_MODELS_SEARCH_DIR.iterdir()):
+                    TEST_MODELS_SEARCH_DIR.rmdir()
+            except OSError: pass # Ignore if not empty or other issues
+
+    ASCIIColors.yellow("\n--- WhisperCppSTTBinding Test Finished ---")

lollms_client/tti_bindings/lollms/__init__.py

@@ -16,7 +16,6 @@ class LollmsTTIBinding_Impl(LollmsTTIBinding):
 
     def __init__(self,
                  host_address: Optional[str] = "http://localhost:9600", # Default LOLLMS host
-                 model_name: Optional[str] = None, # Default service name (server decides if None)
                  service_key: Optional[str] = None,
                  verify_ssl_certificate: bool = True):
         """
@@ -24,14 +23,13 @@ class LollmsTTIBinding_Impl(LollmsTTIBinding):
 
         Args:
             host_address (Optional[str]): Host address for the LOLLMS service.
-            model_name (Optional[str]): Default service/model identifier (currently unused by LOLLMS TTI endpoints).
             service_key (Optional[str]): Authentication key (used for client_id verification).
             verify_ssl_certificate (bool): Whether to verify SSL certificates.
         """
-        super().__init__(host_address=host_address,
-                         model_name=model_name, # model_name is not directly used by LOLLMS TTI API yet
-                         service_key=service_key,
-                         verify_ssl_certificate=verify_ssl_certificate)
+        super().__init__(binding_name="lollms")
+        self.host_address=host_address
+        self.verify_ssl_certificate = verify_ssl_certificate
+
         # The 'service_key' here will act as the 'client_id' for TTI requests if provided.
         # This assumes the client library user provides their LOLLMS client_id here.
         self.client_id = service_key