Atomic-Germ/flm-add.py

You can install this with PyPI

(recommended) uv tool install flm-add

pip install flm-add

Install a pre-converted FLM (Q4NX) model and register it with FastFlowLM.

positional arguments:
  repo                  Hugging Face repo id (Org/Name), ModelScope id
                        (with --modelscope), URL, or local directory

options:
  -h, --help            show this help message and exit
  --tag TAG             Registry tag (default: derived from the repo
                        name, e.g. qwen3.5-claude:9b)
  --family FAMILY       details.family for engine dispatch (default:
                        from matching official entry)
  --config CONFIG       model_list.json to update (default:
                        $FLM_CONFIG_PATH or
                        ~/.config/flm/model_list.json)
  --models-root MODELS_ROOT
                        models directory (default: $FLM_MODEL_PATH or
                        ~/.config/flm/models)
  --xclbin-dir XCLBIN_DIR
                        user xclbins directory (default:
                        ~/.config/flm/xclbins)
  --xclbin-from XCLBIN_FROM
                        official model directory name to link xclbins
                        from (default: best match, e.g.
                        Qwen3.6-35B-A3B-NPU2)
  --system-list SYSTEM_LIST
                        official model_list.json used for defaults
                        (default: auto-detect)
  --modelscope          Treat REPO as a ModelScope repo id
  --no-xclbin           Do not create the xclbins symlink
  --no-verify           Skip sha256 verification of downloads
  --force               Overwrite existing model files/links
  --dry-run             Print the plan and exit
  --quiet               Less output

Example:

$ flm-add Atomic-Germ/DynaGuard-4B-NPU2 --tag dynaguard:4b --family qwen3.5
[INFO] xclbins from official qwen3.5:4b
[INFO] Downloading model files from Hugging Face: Atomic-Germ/DynaGuard-4B-NPU2
Downloading config.json (0.00 GB)...
    100% (0.00 GB / 0.00 GB)
Downloading model.q4nx (3.29 GB)...
      0% (0.00 GB / 3.29 GB)
      5% (0.16 GB / 3.29 GB)
     10% (0.33 GB / 3.29 GB)
     15% (0.49 GB / 3.29 GB)
     20% (0.66 GB / 3.29 GB)
     25% (0.82 GB / 3.29 GB)
     30% (0.99 GB / 3.29 GB)
     35% (1.15 GB / 3.29 GB)
     40% (1.32 GB / 3.29 GB)
     45% (1.48 GB / 3.29 GB)
     50% (1.65 GB / 3.29 GB)
     55% (1.81 GB / 3.29 GB)
     60% (1.98 GB / 3.29 GB)
     65% (2.14 GB / 3.29 GB)
     70% (2.30 GB / 3.29 GB)
     75% (2.47 GB / 3.29 GB)
     80% (2.63 GB / 3.29 GB)
     85% (2.80 GB / 3.29 GB)
     90% (2.96 GB / 3.29 GB)
     95% (3.13 GB / 3.29 GB)
    100% (3.29 GB / 3.29 GB)
Downloading tokenizer.json (0.01 GB)...
     45% (0.01 GB / 0.01 GB)
     55% (0.01 GB / 0.01 GB)
    100% (0.01 GB / 0.01 GB)
Downloading tokenizer_config.json (0.00 GB)...
    100% (0.00 GB / 0.00 GB)
Downloading chat_template.jinja (0.00 GB)...
    100% (0.00 GB / 0.00 GB)
[INFO] Registered tag 'dynaguard:4b' in /home/atomic-germ/.config/flm/model_list.json
[INFO] Linked xclbins: /home/atomic-germ/.config/flm/xclbins/DynaGuard-4B-NPU2 -> /opt/fastflowlm/share/flm/xclbins/Qwen3.5-4B-NPU2

Done: DynaGuard-4B-NPU2 installed to /home/atomic-germ/.config/flm/models/DynaGuard-4B-NPU2
Run:  flm run dynaguard:4b   (or: flm serve dynaguard:4b)

FLM_CONFIG_PATH="$HOME/.config/flm/model_list.json" FLM_XCLBIN_PATH="$HOME/.config/flm" flm run dynaguard:4b