Add automatic architecture detection and binary selection for llama.cpp

llm_clients/qwen_translator.py

@@ -97,8 +97,38 @@ class QwenTranslatorClient(LlmClient):
         # If not found, we'll search after extraction
         binary_path = binary_dir / "main"  # Default to 'main' (standard llama.cpp binary name)
         
-        # Download the zip file
-        zip_url = "https://github.com/ggml-org/llama.cpp/releases/download/b6995/llama-b6995-bin-ubuntu-x64.zip"
+        # Detect architecture and select appropriate binary
+        machine = platform.machine().lower()
+        is_linux = sys.platform.startswith('linux')
+        
+        # Map architectures to available binaries
+        # Available: ubuntu-x64, ubuntu-s390x, ubuntu-vulkan-x64, macos-arm64, macos-x64, win-cpu-x64, win-cpu-arm64
+        binary_name = "unknown"
+        if is_linux:
+            if machine in ['x86_64', 'amd64']:
+                # Try standard Ubuntu x64 first (most common)
+                zip_url = "https://github.com/ggml-org/llama.cpp/releases/download/b6996/llama-b6996-bin-ubuntu-x64.zip"
+                binary_name = "ubuntu-x64"
+            elif machine == 's390x':
+                zip_url = "https://github.com/ggml-org/llama.cpp/releases/download/b6996/llama-b6996-bin-ubuntu-s390x.zip"
+                binary_name = "ubuntu-s390x"
+            elif machine in ['aarch64', 'arm64']:
+                # No ARM Linux binary available, will need to handle this
+                raise RuntimeError(
+                    f"ARM64 Linux architecture detected, but no pre-built ARM64 Linux binary is available. "
+                    f"Available binaries are for x86_64, s390x, macOS ARM64, and Windows. "
+                    f"Translation feature cannot work on ARM64 Linux without compiling from source."
+                )
+            else:
+                # Try x64 as fallback
+                print(f"   ⚠️  Unknown Linux architecture '{machine}', trying x64 binary...", flush=True)
+                zip_url = "https://github.com/ggml-org/llama.cpp/releases/download/b6996/llama-b6996-bin-ubuntu-x64.zip"
+                binary_name = "ubuntu-x64"
+        else:
+            # Non-Linux systems (shouldn't reach here due to earlier check, but just in case)
+            raise RuntimeError(f"Unsupported platform: {sys.platform}")
+        
+        print(f"   Selected binary: {binary_name} for architecture: {machine}", flush=True)
         zip_path = binary_dir / "llama-binary.zip"
         
         try:
@@ -191,8 +221,8 @@ class QwenTranslatorClient(LlmClient):
             
             # Test if binary can actually run (check architecture compatibility)
             print(f"   🔍 Testing binary compatibility...", flush=True)
-            machine = platform.machine()
-            print(f"   System architecture: {machine}", flush=True)
+            detected_machine = platform.machine()
+            print(f"   System architecture: {detected_machine}", flush=True)
             
             try:
                 # Try to run the binary with --help to verify it works
@@ -230,10 +260,11 @@ class QwenTranslatorClient(LlmClient):
                     
                     raise RuntimeError(
                         f"Binary architecture mismatch. The downloaded binary is not compatible with this system.\n"
-                        f"System architecture: {machine}\n"
+                        f"System architecture: {detected_machine}\n"
+                        f"Binary selected: {binary_name}\n"
                         f"Binary info: {file_info}\n"
                         f"Error: {error_msg}\n"
-                        f"The Ubuntu x64 binary may not be compatible with this system. "
+                        f"The binary may require specific shared libraries or a different architecture. "
                         f"Translation feature requires a compatible llama.cpp binary for this architecture."
                     ) from os_error
                 else: