Menu

#202 feat(installer): stream MLX server logs during model load wait

closed
nobody
released (243)
2026-06-09
2026-06-09
Anonymous
No

Originally created by: adityaharishch

Problem

After → Installing MLX inference server launchd agent… the installer went silent for up to 5 minutes. Users had no idea which model was being loaded, whether a large download was happening, or how long to wait.

Changes

install-from-bundle.sh — tail both MLX log files while health-check polling:

  • mlx-server.log (JSON structured logs) — parsed with Python to extract the message field, shown with · / prefixes
  • mlx-server-error.log (raw stderr) — passed through directly so huggingface_hub tqdm download bars render with file name, size, transfer speed, and ETA

Both tail processes are killed cleanly once the health check passes or times out.

bootstrap.sh — remove stale "Python venv + Node runtime (~200 MB)" text (venv is no longer shipped; Python deps install from PyPI at setup time per [#201]).

services/pyproject.toml — update package description (hermes removed) and [tool.uv] comment (no longer references "shipped venv").

What users will see

 Waiting for the MLX server to load the model
  · llm_selector: selecting mlx-community/Qwen3-5B-4bit (will download)
  · run_task_linker_mlx: loading mlx-community/Qwen3-5B-4bit
  Fetching 8 files: 100%|████████████| 8/8 [00:42<00:00]
  Downloading: 100%|████████████| 2.4G/2.4G [01:23<00:00, 28.7MB/s]
  · run_task_linker_mlx: model loaded in 12.3s
   MLX server ready (97s)

Test plan

  • [ ] bash -n scripts/install-from-bundle.sh passes ✓
  • [ ] Fresh install: confirm model name + download progress visible in terminal
  • [ ] Re-install (model cached): confirm fast path shows "model loaded" without download bars
  • [ ] Timeout path: confirm tail processes are killed cleanly on 300s timeout

🤖 Generated with Claude Code

Related

Tickets: #201

Discussion

  • Anonymous

    Anonymous - 2026-06-09

    Ticket changed by: adityaharishch

    • status: open --> closed
     
  • Anonymous

    Anonymous - 2026-06-09

    Originally posted by: adityaharishch

    🎉 This PR is included in version 1.33.0 🎉

    The release is available on:

    Your semantic-release bot 📦🚀

     

Log in to post a comment.