Open jwvanderbeck opened 1 year ago
Is there a stats.json
file in the extension's folder? If so what are its contents?
Pulled it into vscode and looks to be valid json but I only looked real quick. It has a lot of data though so attaching it here rather than pasting it stats.zip
Seems like it's not saving any results for the latent benchmark for some reason, it's just an empty array. The Latent
sampler in highres fix for txt2img
works right? Were there any errors printed during the benchmark?
Oh wait I see, it ran out of memory during the latent benchmark. I guess that should be made clearer. Maybe reduce the maximum size/batch count to the amount you can use with Latent upscalers?
I don't normally use it but it does work (I've used it by accident once or twice). I don't recall any errors other than the one I posted.
I did originally try running a benchmark with different settings, with a 2048 size and only 2 count which failed pretty quick with the error which is when I then reset the settings to the defaults thinking it might be a bug with only 2 count, but then got the same error.
I don't know if that had anything to do with it.
I will try deleting the stats.json file and rerunning with the default settings
Well I deleted the file and reran it with the default settings and it worked, so it must be as you say when I tried to run it at 2048 and then the file got corrupted?
That said isn't the whole point here that this extension goes until it runs out of memory? Is there something special about this stage that you can't detect the oom condition?
I tried doing that earlier, the problem is sometimes it doesn't throw an OOM exception if you just barely cross the VRAM limit and instead the program just hangs indefinitely, even all the way overnight
benchmark error: CUDA out of memory. Tried to allocate 7.15 GiB (GPU 0; 24.00 GiB total capacity; 10.23 GiB already allocated; 2.73 GiB free; 17.70 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF Traceback (most recent call last): File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 337, in run_predict output = await app.get_blocks().process_api( File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1015, in process_api result = await self.call_function( File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 833, in call_function prediction = await anyio.to_thread.run_sync( File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\anyio\to_thread.py", line 31, in run_sync return await get_asynclib().run_sync_in_worker_thread( File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 937, in run_sync_in_worker_thread return await future File "D:\Developer\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 867, in run result = context.run(func, *args) File "D:\Developer\stable-diffusion-webui\extensions\a1111-stable-diffusion-webui-vram-estimator\scripts\vram_estimator.py", line 201, in run_benchmark curves[k] = VRAMCurve(v) File "D:\Developer\stable-diffusion-webui\extensions\a1111-stable-diffusion-webui-vram-estimator\scripts\vramestimator.py", line 71, in init A = np.c[np.ones(data.shape[0]), data[:,:2], data[:,0]**2, np.prod(data[:,:2], axis=1), \ IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed