Error: oMLX startup failed
oMLX output tail:
INFO:omlx.settings:Initialized settings with base_path: /Users/philipbodenbach/.local/share/werk1112/backends/omlx/workers/5d5f16c75a26622bbb19988f1bbf610812cfee3d50511e45
INFO:omlx.settings:Saved settings to /Users/philipbodenbach/.local/share/werk1112/backends/omlx/workers/5d5f16c75a26622bbb19988f1bbf610812cfee3d50511e45/settings.json
INFO:     Uvicorn running on http://127.0.0.1:60395 (Press CTRL+C to quit)
INFO:omlx.settings:Saved settings to /Users/philipbodenbach/.local/share/werk1112/backends/omlx/workers/5d5f16c75a26622bbb19988f1bbf610812cfee3d50511e45/settings.json
INFO:omlx.server:Generated and saved new auth secret key
INFO:omlx.server:CORS origins: ['*']
INFO:omlx.model_discovery:Discovered model: files (type: llm, engine: batched, size: 94.39GB)
INFO:omlx.model_discovery:Discovered model: mlx-community--Llama-3.2-3B-Instruct-4bit (type: llm, engine: batched, size: 1.77GB)
INFO:omlx.engine_pool:Discovered 2 models
INFO:omlx.server:Server initialized with 2 models
INFO:omlx.server:Default model: files
INFO:omlx.server:Memory guard tier: balanced (guard on)
INFO:omlx.server:Default max tokens: 32768
INFO:omlx.server:API key authentication: enabled
INFO:omlx.server:HF Downloader initialized
INFO:omlx.server:ModelScope SDK not installed, MS downloader disabled
INFO:omlx.server:oQ Quantizer initialized
INFO:omlx.server:HF Uploader initialized
INFO:     Started server process [10416]
INFO:     Waiting for application startup.
INFO:omlx.settings:Saved settings to /Users/philipbodenbach/.local/share/werk1112/backends/omlx/workers/5d5f16c75a26622bbb19988f1bbf610812cfee3d50511e45/settings.json
INFO:omlx.server:Auto-detected server aliases: ['localhost', '127.0.0.1', 'MacBook-Pro-von-Philip.local', '192.168.200.198']
WARNING:omlx.process_memory_enforcer:Metal cap (37.4GB, Apple max_recommended_working_set_size) is below the oMLX static ceiling (42.0GB); leaving Apple's default Metal cap active because iogpu.wired_limit_mb is unset. Raise it with: sudo sysctl iogpu.wired_limit_mb=43008
INFO:omlx.process_memory_enforcer:Process memory enforcer started (tier=balanced, ceiling=21.0GB, interval=1.0s)
INFO:     Application startup complete.
WARNING:omlx.server:POST /v1/models/files/load → 507: Model 'files' (29.02GB) does not fit under the dynamic memory ceiling (21.01GB). Close other apps to free RAM (static cap is 42.00GB but only 20.89GB is reclaimable right now), raise memory_guard_tier (safe → balanced → aggressive), or use a smaller model.

Caused by:
    0: oMLX POST /v1/models/files/load failed
    1: OpenAI backend HTTP 507: {"error":{"message":"Model 'files' (29.02GB) does not fit under the dynamic memory ceiling (21.01GB). Close other apps to free RAM (static cap is 42.00GB but only 20.89GB is reclaimable right now), raise memory_guard_tier (safe → balanced → aggressive), or use a smaller model.","type":"server_error","param":null,"code":null}}