2026-04-16 22:54:19.800 [Info] (main) Starting Lemonade Server...
2026-04-16 22:54:20.049 [Info] (main)   Version: 10.2.0
2026-04-16 22:54:20.049 [Info] (main)   Cache dir: /Library/Application Support/lemonade/.cache
2026-04-16 22:54:20.049 [Info] (main)   Port: 13305
2026-04-16 22:54:20.049 [Info] (main)   Host: localhost
2026-04-16 22:54:20.049 [Info] (main)   Log level: info
2026-04-16 22:54:20.156 [Info] (ModelManager) Loading "recipe_options.json"
2026-04-16 22:54:20.608 [Info] (WebSocket) Configured port: 9000
2026-04-16 22:54:20.608 [Info] (Server) Starting HTTP server on localhost:13305
2026-04-16 22:54:20.836 [Info] (Server) Host resolution: IPv4=127.0.0.1, IPv6=::1
2026-04-16 22:54:21.082 [Info] (WebSocket) Server started on port 9000
2026-04-16 22:54:21.082 [Info] (Server) WebSocket server started on port 9000
2026-04-16 22:54:21.083 [Info] (Server) Binding IPv4 HTTP server to 127.0.0.1:13305...
2026-04-16 22:54:21.083 [Info] (Server) IPv4 HTTP server listening on 127.0.0.1:13305
2026-04-16 22:54:21.083 [Info] (Server) Binding IPv6 HTTP server to [::1]:13305...
2026-04-16 22:54:21.083 [Info] (Server) IPv6 HTTP server listening on [::1]:13305
2026-04-16 22:54:54.948 [Info] (ModelManager) Building models cache...
2026-04-16 22:54:56.680 [Info] (ModelManager) Backend availability:
2026-04-16 22:54:56.680 [Info] (ModelManager)   - NPU hardware: No
2026-04-16 22:54:56.680 [Info] (ModelManager)   - System RAM: 16.0 GB (max model size: 12.8 GB)
2026-04-16 22:54:56.976 [Info] (ModelManager) Cache built: 44 total, 3 downloaded
2026-04-16 22:55:59.155 [Info] (Server) Installing backend: llamacpp:metal
2026-04-16 22:55:59.159 [Info] (llamacpp Server) Upgrading llama-server from b8460 to b8779
2026-04-16 22:55:59.218 [Info] (llamacpp Server) Installing llama-server (version: b8779)
2026-04-16 22:56:01.034 [Info] (ModelManager) Building models cache...
2026-04-16 22:56:01.067 [Info] (ModelManager) Cache built: 44 total, 3 downloaded
2026-04-16 22:56:01.067 [Info] (Server) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:01.071 [Info] (LlamaCpp) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:01.072 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:56:01.073 [Info] (ProcessManager) Process started successfully, PID: 3331
2026-04-16 22:56:03.249 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:56:03.249 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_library_init: loaded in 0.147 sec
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:56:03.396 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:56:03.397 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:56:03.557 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:56:03.557 [Error] (WrappedServer) This usually means:
2026-04-16 22:56:03.557 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:56:03.557 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:56:03.557 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:56:11.516 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:56:11.520 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:56:12.152 [Info] (Server) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:12.153 [Info] (Router) Another load is in progress, waiting...
2026-04-16 22:56:13.589 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:56:13.589 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:56:13.589 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:56:13.589 [Info] (Router) All models evicted
2026-04-16 22:56:13.589 [Info] (LlamaCpp) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:13.590 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:56:13.592 [Info] (ProcessManager) Process started successfully, PID: 3352
2026-04-16 22:56:13.633 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:56:13.633 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_library_init: loaded in 0.011 sec
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:56:13.644 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:56:13.746 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:56:13.746 [Error] (WrappedServer) This usually means:
2026-04-16 22:56:13.746 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:56:13.746 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:56:13.746 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:56:21.982 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:56:21.983 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:56:24.057 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:56:24.057 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:56:24.057 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:56:24.057 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:56:24.057 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:56:24.057 [Info] (LlamaCpp) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:24.058 [Error] (Server) Per-model settings: ctx_size=4096, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-16 22:56:24.059 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:56:24.061 [Info] (ProcessManager) Process started successfully, PID: 3391
2026-04-16 22:56:24.168 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:56:24.168 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_library_init: loaded in 0.063 sec
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:56:24.231 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:56:24.352 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:56:24.352 [Error] (WrappedServer) This usually means:
2026-04-16 22:56:24.352 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:56:24.352 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:56:24.352 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:56:32.399 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:56:32.400 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:56:34.471 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:56:34.471 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:56:34.471 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:56:34.472 [Info] (Router) All models evicted
2026-04-16 22:56:34.472 [Info] (LlamaCpp) Loading model: Gemma-4-E4B-it-GGUF
2026-04-16 22:56:34.474 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:56:34.477 [Info] (ProcessManager) Process started successfully, PID: 3402
2026-04-16 22:56:34.586 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:56:34.586 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_library_init: loaded in 0.091 sec
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:56:34.677 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:56:34.764 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:56:34.764 [Error] (WrappedServer) This usually means:
2026-04-16 22:56:34.764 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:56:34.764 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:56:34.764 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:56:42.584 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:56:42.585 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:56:44.633 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:56:44.633 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:56:44.633 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:56:44.633 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:56:44.633 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:56:44.633 [Error] (Server) Error 500: POST /api/v1/load
2026-04-16 22:56:57.751 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:56:57.755 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:56:57.756 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:56:57.757 [Info] (ProcessManager) Process started successfully, PID: 3415
2026-04-16 22:56:57.850 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:56:57.850 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_library_init: loaded in 0.072 sec
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:56:57.922 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:56:58.016 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:56:58.017 [Error] (WrappedServer) This usually means:
2026-04-16 22:56:58.017 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:56:58.017 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:56:58.017 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:57:05.566 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:57:05.586 [Info] (Router) Another load is in progress, waiting...
2026-04-16 22:57:06.250 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:57:06.251 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:57:08.316 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:57:08.316 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:57:08.316 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:57:08.316 [Info] (Router) All models evicted
2026-04-16 22:57:08.316 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:57:08.316 [Info] (Router) Another load is in progress, waiting...
2026-04-16 22:57:08.316 [Info] (Router) Per-model settings: ctx_size=2048, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-16 22:57:08.318 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:57:08.320 [Info] (ProcessManager) Process started successfully, PID: 3418
2026-04-16 22:57:08.386 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:57:08.387 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:57:08.402 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:57:08.492 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:57:08.492 [Error] (WrappedServer) This usually means:
2026-04-16 22:57:08.492 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:57:08.492 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:57:08.492 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:57:16.607 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:57:16.607 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:57:18.683 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:57:18.684 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:57:18.684 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:57:18.684 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:57:18.684 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:57:18.684 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:57:18.684 [Error] (Server) Error 500: POST /api/v1/load
2026-04-16 22:57:18.684 [Error] (Server) Per-model settings: ctx_size=2048, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-16 22:57:18.685 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:57:18.687 [Info] (ProcessManager) Process started successfully, PID: 3425
2026-04-16 22:57:18.753 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:57:18.753 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:57:18.768 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:57:18.859 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:57:18.859 [Error] (WrappedServer) This usually means:
2026-04-16 22:57:18.859 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:57:18.859 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:57:18.859 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:57:27.219 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:57:27.220 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:57:29.249 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:57:29.249 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:57:29.250 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:57:29.250 [Info] (Router) All models evicted
2026-04-16 22:57:29.250 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:57:29.253 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:57:29.256 [Info] (ProcessManager) Process started successfully, PID: 3436
2026-04-16 22:57:29.417 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:57:29.417 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:57:29.486 [Info] (Process) ggml_metal_library_init: loaded in 0.069 sec
2026-04-16 22:57:29.486 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:57:29.486 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:57:29.486 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:57:29.487 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:57:29.604 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:57:29.604 [Error] (WrappedServer) This usually means:
2026-04-16 22:57:29.604 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:57:29.604 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:57:29.604 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:57:37.816 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:57:37.817 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:57:39.892 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:57:39.892 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:57:39.892 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:57:39.892 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:57:39.892 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:57:39.892 [Error] (Server) Error 500: POST /api/v1/load
2026-04-16 22:57:43.398 [Info] (Server) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:57:43.398 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:57:43.400 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:57:43.401 [Info] (ProcessManager) Process started successfully, PID: 3438
2026-04-16 22:57:43.485 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:57:43.485 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_library_init: loaded in 0.068 sec
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:57:43.553 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:57:43.577 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:57:43.577 [Error] (WrappedServer) This usually means:
2026-04-16 22:57:43.577 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:57:43.577 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:57:43.577 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:57:46.815 [Info] (Server) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:57:46.816 [Info] (Router) Another load is in progress, waiting...
2026-04-16 22:57:51.637 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:57:51.637 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:57:53.704 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:57:53.705 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:57:53.705 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:57:53.705 [Info] (Router) All models evicted
2026-04-16 22:57:53.705 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:57:53.705 [Info] (Router) Another load is in progress, waiting...
2026-04-16 22:57:53.705 [Info] (Router) Per-model settings: ctx_size=64000, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-16 22:57:53.707 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:57:53.709 [Info] (ProcessManager) Process started successfully, PID: 3443
2026-04-16 22:57:53.771 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:57:53.771 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:57:53.787 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:57:53.866 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:57:53.866 [Error] (WrappedServer) This usually means:
2026-04-16 22:57:53.866 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:57:53.866 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:57:53.866 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:58:02.199 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:58:02.201 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:58:04.266 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:58:04.267 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:58:04.267 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:58:04.267 [Error] (Router) Failed to load model: llama-server failed to start
%Y-%m-%d %H:%M:%S.#ms [Error] (log) Loading model: Qwen3.5-2B-GGUF (checkpoint: unsloth/Qwen3.5-2B-GGUF:Qwen3.5-2B-UD-Q4_K_XL.gguf, recipe: llamacpp, type: llm, device: gpu)
2026-04-16 22:58:04.267 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:58:04.267 [Error] (Server) Per-model settings: ctx_size=64000, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-16 22:58:04.267 [Error] (Server) Error 500: POST /api/v1/load
2026-04-16 22:58:04.270 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:58:04.272 [Info] (ProcessManager) Process started successfully, PID: 3444
2026-04-16 22:58:04.338 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:58:04.338 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_library_init: loaded in 0.014 sec
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:58:04.352 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:58:04.353 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:58:04.432 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:58:04.432 [Error] (WrappedServer) This usually means:
2026-04-16 22:58:04.432 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:58:04.432 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:58:04.432 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:58:12.683 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:58:12.685 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:58:14.760 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:58:14.760 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:58:14.760 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:58:14.760 [Info] (Router) All models evicted
2026-04-16 22:58:14.760 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-16 22:58:14.762 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:58:14.764 [Info] (ProcessManager) Process started successfully, PID: 3447
2026-04-16 22:58:14.820 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:58:14.820 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_library_init: loaded in 0.013 sec
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:58:14.833 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:58:14.834 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:58:14.834 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:58:14.834 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:58:14.834 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:58:14.932 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:58:14.932 [Error] (WrappedServer) This usually means:
2026-04-16 22:58:14.932 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:58:14.932 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:58:14.932 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:58:23.086 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:58:23.087 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:58:25.153 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:58:25.153 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:58:25.153 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:58:25.153 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:58:25.153 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:58:25.153 [Error] (Server) Error 500: POST /api/v1/load
2026-04-16 22:59:29.202 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:59:29.203 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:59:29.204 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:59:29.206 [Info] (ProcessManager) Process started successfully, PID: 3492
2026-04-16 22:59:29.285 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:59:29.285 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_library_init: loaded in 0.049 sec
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:59:29.334 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:59:29.378 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:59:29.378 [Error] (WrappedServer) This usually means:
2026-04-16 22:59:29.378 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:59:29.378 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:59:29.378 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:59:37.451 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:59:37.452 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:59:39.523 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-16 22:59:39.523 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-16 22:59:39.523 [Info] (Router) Evicting all models (0 total)
2026-04-16 22:59:39.524 [Info] (Router) All models evicted
2026-04-16 22:59:39.524 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-16 22:59:39.526 [Info] (LlamaCpp) Starting llama-server...
2026-04-16 22:59:39.528 [Info] (ProcessManager) Process started successfully, PID: 3493
2026-04-16 22:59:39.603 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-16 22:59:39.603 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-16 22:59:39.617 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-16 22:59:39.618 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-16 22:59:39.618 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-16 22:59:39.618 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-16 22:59:39.704 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-16 22:59:39.704 [Error] (WrappedServer) This usually means:
2026-04-16 22:59:39.704 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-16 22:59:39.704 [Error] (WrappedServer)   - Incompatible model file
2026-04-16 22:59:39.704 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-16 22:59:48.064 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-16 22:59:48.065 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-16 22:59:50.136 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-16 22:59:50.136 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:59:50.136 [Info] (LlamaCpp) Unloading model...
2026-04-16 22:59:50.136 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-16 22:59:50.137 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-16 22:59:50.137 [Error] (Server) Error 500: POST /api/v1/load
2026-04-17 10:14:21.041 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:21.045 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:21.049 [Info] (LlamaCpp) Starting llama-server...
2026-04-17 10:14:21.052 [Info] (ProcessManager) Process started successfully, PID: 10249
2026-04-17 10:14:21.117 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-17 10:14:21.117 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_library_init: loaded in 0.064 sec
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-17 10:14:21.182 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-17 10:14:21.219 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-17 10:14:21.219 [Error] (WrappedServer) This usually means:
2026-04-17 10:14:21.219 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-17 10:14:21.219 [Error] (WrappedServer)   - Incompatible model file
2026-04-17 10:14:21.219 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-17 10:14:22.649 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:22.649 [Info] (Router) Another load is in progress, waiting...
2026-04-17 10:14:29.370 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-17 10:14:29.371 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-17 10:14:31.440 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-17 10:14:31.440 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-17 10:14:31.440 [Info] (Router) Evicting all models (0 total)
2026-04-17 10:14:31.440 [Info] (Router) All models evicted
2026-04-17 10:14:31.440 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:31.440 [Info] (Router) Another load is in progress, waiting...
2026-04-17 10:14:31.440 [Info] (Router) Per-model settings: ctx_size=2048, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-17 10:14:31.441 [Info] (LlamaCpp) Starting llama-server...
2026-04-17 10:14:31.442 [Info] (ProcessManager) Process started successfully, PID: 10256
2026-04-17 10:14:31.489 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-17 10:14:31.489 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_library_init: loaded in 0.011 sec
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-17 10:14:31.501 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-17 10:14:31.608 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-17 10:14:31.608 [Error] (WrappedServer) This usually means:
2026-04-17 10:14:31.608 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-17 10:14:31.608 [Error] (WrappedServer)   - Incompatible model file
2026-04-17 10:14:31.608 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-17 10:14:39.748 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-17 10:14:39.749 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-17 10:14:41.815 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-17 10:14:41.815 [Info] (LlamaCpp) Unloading model...
2026-04-17 10:14:41.815 [Info] (LlamaCpp) Unloading model...
2026-04-17 10:14:41.815 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-17 10:14:41.816 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:41.817 [Error] (Server) Error 500: POST /api/v1/load
2026-04-17 10:14:41.818 [Info] (LlamaCpp) Starting llama-server...
2026-04-17 10:14:41.819 [Info] (ProcessManager) Process started successfully, PID: 10257
2026-04-17 10:14:41.867 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-17 10:14:41.867 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_library_init: loaded in 0.009 sec
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-17 10:14:41.877 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-17 10:14:41.989 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-17 10:14:41.989 [Error] (WrappedServer) This usually means:
2026-04-17 10:14:41.989 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-17 10:14:41.989 [Error] (WrappedServer)   - Incompatible model file
2026-04-17 10:14:41.989 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-17 10:14:49.727 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-17 10:14:49.728 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-17 10:14:51.798 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-17 10:14:51.799 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-17 10:14:51.799 [Info] (Router) Evicting all models (0 total)
2026-04-17 10:14:51.799 [Info] (Router) All models evicted
2026-04-17 10:14:51.799 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-17 10:14:51.800 [Info] (LlamaCpp) Starting llama-server...
2026-04-17 10:14:51.801 [Info] (ProcessManager) Process started successfully, PID: 10263
2026-04-17 10:14:51.896 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-17 10:14:51.896 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_library_init: loaded in 0.059 sec
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-17 10:14:51.955 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-17 10:14:51.965 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-17 10:14:51.965 [Error] (WrappedServer) This usually means:
2026-04-17 10:14:51.965 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-17 10:14:51.965 [Error] (WrappedServer)   - Incompatible model file
2026-04-17 10:14:51.965 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-17 10:14:59.932 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-17 10:14:59.932 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-17 10:15:01.997 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-17 10:15:01.997 [Info] (LlamaCpp) Unloading model...
2026-04-17 10:15:01.997 [Info] (LlamaCpp) Unloading model...
2026-04-17 10:15:01.997 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-17 10:15:01.997 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-17 10:15:01.997 [Error] (Server) Error 500: POST /api/v1/load
2026-04-18 15:33:50.081 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:33:50.093 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:33:50.097 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 15:33:50.100 [Info] (ProcessManager) Process started successfully, PID: 45487
2026-04-18 15:33:50.244 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 15:33:50.244 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_library_init: loaded in 0.074 sec
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 15:33:50.318 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 15:33:50.390 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 15:33:50.390 [Error] (WrappedServer) This usually means:
2026-04-18 15:33:50.390 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 15:33:50.390 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 15:33:50.390 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 15:33:55.654 [Info] (Server) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:33:55.654 [Info] (Router) Another load is in progress, waiting...
2026-04-18 15:33:58.694 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 15:33:58.695 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 15:34:00.771 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-18 15:34:00.771 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-18 15:34:00.771 [Info] (Router) Evicting all models (0 total)
2026-04-18 15:34:00.771 [Info] (Router) All models evicted
2026-04-18 15:34:00.771 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:34:00.771 [Info] (Router) Another load is in progress, waiting...
2026-04-18 15:34:00.771 [Info] (Router) Per-model settings: ctx_size=2048, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-18 15:34:00.774 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 15:34:00.776 [Info] (ProcessManager) Process started successfully, PID: 45492
2026-04-18 15:34:00.848 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 15:34:00.848 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 15:34:00.863 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 15:34:00.952 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 15:34:00.952 [Error] (WrappedServer) This usually means:
2026-04-18 15:34:00.952 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 15:34:00.952 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 15:34:00.952 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 15:34:09.288 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 15:34:09.289 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 15:34:11.361 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-18 15:34:11.361 [Info] (LlamaCpp) Unloading model...
2026-04-18 15:34:11.361 [Info] (LlamaCpp) Unloading model...
2026-04-18 15:34:11.362 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-18 15:34:11.363 [Error] (Server) Loading model: Gemma-4-E2B-it-GGUF (checkpoint: unsloth/gemma-4-E2B-it-GGUF:Q4_K_M, recipe: llamacpp, type: llm, device: gpu)
2026-04-18 15:34:11.363 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:34:11.364 [Error] (Server) Error 500: POST /api/v1/load
2026-04-18 15:34:11.373 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 15:34:11.376 [Info] (ProcessManager) Process started successfully, PID: 45494
2026-04-18 15:34:11.440 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 15:34:11.440 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_library_init: loaded in 0.013 sec
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 15:34:11.453 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 15:34:11.554 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 15:34:11.554 [Error] (WrappedServer) This usually means:
2026-04-18 15:34:11.554 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 15:34:11.554 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 15:34:11.554 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 15:34:19.911 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 15:34:19.911 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 15:34:21.991 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-18 15:34:21.991 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-18 15:34:21.991 [Info] (Router) Evicting all models (0 total)
2026-04-18 15:34:21.991 [Info] (Router) All models evicted
2026-04-18 15:34:21.991 [Info] (LlamaCpp) Loading model: Gemma-4-E2B-it-GGUF
2026-04-18 15:34:21.993 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 15:34:21.994 [Info] (ProcessManager) Process started successfully, PID: 45499
2026-04-18 15:34:22.084 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 15:34:22.084 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_library_init: loaded in 0.084 sec
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 15:34:22.169 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 15:34:22.272 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 15:34:22.272 [Error] (WrappedServer) This usually means:
2026-04-18 15:34:22.272 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 15:34:22.272 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 15:34:22.272 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 15:34:30.567 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 15:34:30.568 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 15:34:32.638 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-18 15:34:32.639 [Info] (LlamaCpp) Unloading model...
2026-04-18 15:34:32.639 [Info] (LlamaCpp) Unloading model...
2026-04-18 15:34:32.639 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-18 15:34:32.639 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-18 15:34:32.639 [Error] (Server) Error 500: POST /api/v1/load
2026-04-18 17:49:43.921 [Info] (Server) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:49:43.923 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:49:43.927 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 17:49:43.930 [Info] (ProcessManager) Process started successfully, PID: 48551
2026-04-18 17:49:44.075 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 17:49:44.075 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_library_init: loaded in 0.064 sec
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 17:49:44.139 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 17:49:44.140 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 17:49:44.261 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 17:49:44.261 [Error] (WrappedServer) This usually means:
2026-04-18 17:49:44.261 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 17:49:44.261 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 17:49:44.261 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 17:49:48.456 [Info] (Server) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:49:48.457 [Info] (Router) Another load is in progress, waiting...
2026-04-18 17:49:52.643 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 17:49:52.644 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 17:49:54.705 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-18 17:49:54.706 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-18 17:49:54.706 [Info] (Router) Evicting all models (0 total)
2026-04-18 17:49:54.706 [Info] (Router) All models evicted
2026-04-18 17:49:54.707 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:49:54.707 [Info] (Router) Another load is in progress, waiting...
2026-04-18 17:49:54.707 [Info] (Router) Per-model settings: ctx_size=64000, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-18 17:49:54.709 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 17:49:54.712 [Info] (ProcessManager) Process started successfully, PID: 48554
2026-04-18 17:49:54.787 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 17:49:54.787 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_library_init: loaded in 0.015 sec
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 17:49:54.802 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 17:49:54.890 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 17:49:54.890 [Error] (WrappedServer) This usually means:
2026-04-18 17:49:54.890 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 17:49:54.890 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 17:49:54.890 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 17:50:03.135 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 17:50:03.136 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 17:50:05.210 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-18 17:50:05.211 [Info] (LlamaCpp) Unloading model...
2026-04-18 17:50:05.211 [Info] (LlamaCpp) Unloading model...
2026-04-18 17:50:05.211 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-18 17:50:05.212 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:50:05.212 [Error] (Server) Error 500: POST /api/v1/load
2026-04-18 17:50:05.212 [Error] (Server) Per-model settings: ctx_size=64000, llamacpp_backend=metal, llamacpp_args=(none)
2026-04-18 17:50:05.214 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 17:50:05.215 [Info] (ProcessManager) Process started successfully, PID: 48555
2026-04-18 17:50:05.277 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 17:50:05.278 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_library_init: loaded in 0.017 sec
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 17:50:05.294 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 17:50:05.384 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 17:50:05.384 [Error] (WrappedServer) This usually means:
2026-04-18 17:50:05.384 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 17:50:05.384 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 17:50:05.384 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 17:50:13.569 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 17:50:13.571 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 17:50:15.638 [Error] (Router) Backend load failed: llama-server failed to start
2026-04-18 17:50:15.638 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-04-18 17:50:15.638 [Info] (Router) Evicting all models (0 total)
2026-04-18 17:50:15.638 [Info] (Router) All models evicted
2026-04-18 17:50:15.639 [Info] (LlamaCpp) Loading model: Qwen3.5-2B-GGUF
2026-04-18 17:50:15.641 [Info] (LlamaCpp) Starting llama-server...
2026-04-18 17:50:15.644 [Info] (ProcessManager) Process started successfully, PID: 48560
2026-04-18 17:50:15.711 [Info] (Process) ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
2026-04-18 17:50:15.711 [Info] (Process) ggml_metal_library_init: using embedded metal library
2026-04-18 17:50:15.727 [Info] (Process) ggml_metal_library_init: loaded in 0.017 sec
2026-04-18 17:50:15.727 [Info] (Process) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
2026-04-18 17:50:15.727 [Info] (Process) ggml_metal_device_init: GPU name:   MTL0
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4  (5002)
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: simdgroup reduction   = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: simdgroup matrix mul. = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: has unified memory    = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: has bfloat            = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: has tensor            = false
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: use residency sets    = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: use shared buffers    = true
2026-04-18 17:50:15.728 [Info] (Process) ggml_metal_device_init: recommendedMaxWorkingSetSize  = 12713.12 MB
2026-04-18 17:50:15.820 [Error] (WrappedServer) llama-server process has terminated with exit code: -1
2026-04-18 17:50:15.820 [Error] (WrappedServer) This usually means:
2026-04-18 17:50:15.820 [Error] (WrappedServer)   - Missing required drivers or dependencies
2026-04-18 17:50:15.820 [Error] (WrappedServer)   - Incompatible model file
2026-04-18 17:50:15.820 [Error] (WrappedServer)   - Try running the server manually to see the actual error
2026-04-18 17:50:24.236 [Warn] (ProcessManager) Process did not respond to SIGTERM, using SIGKILL
2026-04-18 17:50:24.238 [Info] (ProcessManager) Process terminated, waiting for GPU driver cleanup...
2026-04-18 17:50:26.312 [Error] (Router) Retry also failed: llama-server failed to start
2026-04-18 17:50:26.312 [Info] (LlamaCpp) Unloading model...
2026-04-18 17:50:26.312 [Info] (LlamaCpp) Unloading model...
2026-04-18 17:50:26.312 [Error] (Router) Failed to load model: llama-server failed to start
2026-04-18 17:50:26.313 [Error] (Server) Failed to load model: llama-server failed to start
2026-04-18 17:50:26.313 [Error] (Server) Error 500: POST /api/v1/load
2026-04-18 20:42:26.751 [Info] (WebSocket) New connection from: 127.0.0.1 (id: 14)