Phan Tuấn Anh
Phan Tuấn Anh
hmmm the docker image image has recent nvidia driver ? the tensorrt speed up requires driver version ≥ 537.58
 source: https://nvidia.custhelp.com/app/answers/detail/a_id/5487/~/tensorrt-extension-for-stable-diffusion-web-ui
u r sure about that console message output ? the input model is `turbovisionxlSuperFastXL_tvxlV431vae.safetensors` but the message is about `sd_xl_turbo_1.0_fp16`
seem like problem with turbo, u may want to try other models not turbo, there's already another problem with turbo in #4 i may need to remove turbo from supported...
what onnx version and tensorrt version u have ?
it's weird ... i cannot reproduce those error u have enough vram and/or disk space ?
if u dont mind can u delete 2 folders `Unet-onnx` & `Unet-trt` then try convert again ? for any SDXL model it should create 2 files: 1 `.onnx` file 3...
do u see any onnx file as i said ?
no onnx file at all 😰 seem like something mess up with `xformers` ... if there is nothing rely on `xformers` can u try remove it ? comfyui doesnt rely...
> TensorRT engines has been saved to disk. it worked 😉 now u can check if the engine actually work in comfyui