ERROR:system_services.cc(34)] SetApplicationIsDaemon: Error Domain=NSOSStatusErrorDomain Code=-50 "paramErr: error in user parameter list" (-50)

[openSourcerer9000]

git clone ...
pnpm setup:dev:mac
pnpm dev

Fires it up, says it's DLing the model, but seems to be stuck on 0 bytes with the following cryptic error.

[Python] Started server process [86387]
Waiting for application startup.

[Python] Application startup complete.

[Python] Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)

[Renderer] Python backend started successfully
[Renderer] Python backend started successfully
[Renderer] Checking backend health at: http://localhost:8000
[Renderer] Checking backend health at: http://localhost:8000
[Renderer] Backend health: {"status":"ok","models_loaded":false,"active_model":null,"gpu_info":{"name":"Apple M3 Ultra (MPS)","vram":262144,"vramUsed":0},"sage_attention":false,"models_status":[{"id":"fast","name":"LTX-2 Fast (Distilled)","loaded":false,"downloaded":false}]}
[Renderer] Backend health: {"status":"ok","models_loaded":false,"active_model":null,"gpu_info":{"name":"Apple M3 Ultra (MPS)","vram":262144,"vramUsed":0},"sage_attention":false,"models_status":[{"id":"fast","name":"LTX-2 Fast (Distilled)","loaded":false,"downloaded":false}]}
[Python] INFO:_routes.settings:Applied settings patch (changed=none)

[Python] INFO:handlers.download_handler:Downloading ltx-2.3-22b-distilled.safetensors from Lightricks/LTX-2.3

[86570:0314/212944.889847:ERROR:system_services.cc(34)] SetApplicationIsDaemon: Error Domain=NSOSStatusErrorDomain Code=-50 "paramErr: error in user parameter list" (-50)

[karlwhite]

I am getting this too, and spent a while trying to debug it (thinking that the download had stalled/failed), but I discovered if I left it long enough it does eventually update (it seems it actually is downloading, but just not updating the progress indicator).

Maybe try leaving it for a (long) while, and see if it works.

[openSourcerer9000]

You're right, an error log and no progress bar feedback is not expected behavior though, so I'll leave this open.

It still doesn't work though, per the other open issue.

[lmangani]

There's a bug causing the update bar not to update until halfway through, but the download should be ongoing.
Downloading the huuuuuge models takes a REALLY long time. Watch the python process in your resource monitor network tab to see if its downloading to confirm.

[lmangani]

I am getting this too, and spent a while trying to debug it (thinking that the download had stalled/failed), but I discovered if I left it long enough it does eventually update (it seems it actually is downloading, but just not updating the progress indicator).

Maybe try leaving it for a (long) while, and see if it works.

That's indeed the bug but downloading should happen. Did you eventually complete and start?

[openSourcerer9000]

Well, it said it was finished, but now I'm wondering if the two issues are actually related. The model folder is empty after the long "download". The weird part is that my network was indeed showing full download activity for an hour or so.

#10

[karlwhite]

For me it did eventually (fully) download. So the bug is definitely with the progress indicator not updating.

[openSourcerer9000]

What's the default download location? I got it to work (see PR) but the location I sent for download on an external SSD is clearly empty, so I'm not sure where it put the model

[openSourcerer9000]

What are the three processes here? I remember LTX one had an upscaling step, but this won't even include includes audio??? it looks to me like the video step is only 36 seconds, 1/4 of the total generation time for 5 sec 504p. It would be quite handy to be able to decouple each of these steps, only upscaling clips worth keeping and only adding audio as needed.

Does this app have an API?

  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:04<00:31,  4.56s/it]
 25%|██▌       | 2/8 [00:09<00:27,  4.56s/it]
 38%|███▊      | 3/8 [00:13<00:22,  4.56s/it]
 50%|█████     | 4/8 [00:18<00:18,  4.56s/it]
 62%|██████▎   | 5/8 [00:22<00:13,  4.56s/it]
 75%|███████▌  | 6/8 [00:27<00:09,  4.56s/it]
 88%|████████▊ | 7/8 [00:31<00:04,  4.56s/it]
100%|██████████| 8/8 [00:36<00:00,  4.55s/it]
100%|██████████| 8/8 [00:36<00:00,  4.56s/it]

  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:22<00:44, 22.12s/it]
 67%|██████▋   | 2/3 [00:43<00:21, 21.95s/it]
100%|██████████| 3/3 [01:05<00:00, 21.84s/it]
100%|██████████| 3/3 [01:05<00:00, 21.89s/it]

  0%|          | 0/3 [00:00<?, ?it/s]
 67%|██████▋   | 2/3 [00:16<00:08,  8.10s/it]
100%|██████████| 3/3 [00:16<00:00,  4.82s/it]
100%|██████████| 3/3 [00:16<00:00,  5.47s/it]

[lmangani]

Please refer to the original LTX-Desktop for any question, this is just a fork to enable basic generation on MPS and doesn't change how models are executed, apis, etc ☮️

[gabrielrocco]

It is downloading but not showing the progress, I have changed

in file:
backend/services/model_downloader/hugging_face_downloader.py

the function
@contextlib.contextmanager def _patch_http_get_progress(callback: Callable[[int, int], None]) ->

to
@contextlib.contextmanager
def _patch_http_get_progress(callback: Callable[[int, int], None]) -> Iterator[None]:
"""Temporarily monkey-patch huggingface_hub.file_download.http_get
(and xet_get when hf_xet is installed) to inject a custom tqdm
bar that forwards progress to callback.

``hf_hub_download`` does not expose a ``tqdm_class`` parameter (unlike
``snapshot_download``), but its internal ``http_get`` accepts a private
``_tqdm_bar`` kwarg.  We wrap ``http_get`` to inject our own bar when
the caller hasn't already provided one.

When ``hf_xet`` is installed, HuggingFace routes downloads through
``xet_get`` instead of ``http_get``.  We patch both so progress is
reported regardless of the download backend.

See ``test_http_get_accepts_tqdm_bar`` — if that test breaks after a
huggingface_hub upgrade, this patch needs to be revisited.
"""
tqdm_cls = _make_progress_tqdm_class(callback)
original_http_get: Callable[..., Any] = file_download.http_get  # type: ignore[reportUnknownMemberType]

def _wrapped_http_get(*args: Any, **kwargs: Any) -> None:
    if kwargs.get("_tqdm_bar") is None:
        kwargs["_tqdm_bar"] = tqdm_cls(total=kwargs.get("expected_size"), disable=True)
    return original_http_get(*args, **kwargs)

original_xet_get: Callable[..., Any] | None = getattr(file_download, "xet_get", None)

def _wrapped_xet_get(*args: Any, **kwargs: Any) -> None:
    if kwargs.get("_tqdm_bar") is None:
        kwargs["_tqdm_bar"] = tqdm_cls(total=kwargs.get("expected_size"), disable=True)
    return original_xet_get(*args, **kwargs)  # type: ignore[misc]

with patch.object(file_download, "http_get", _wrapped_http_get):
    if original_xet_get is not None:
        with patch.object(file_download, "xet_get", _wrapped_xet_get):
            yield
    else:
        yield

And it started showing the progress

[lmangani]

It is downloading but not showing the progress, I have changed

Please open a PR with the change and take the glory for fixing this 👍